Exploring the Uncanny Valley of .

An Investigation of Human Uniqueness , Control Perceptions, and Threat Experience in the Face of -Sensitive Artificial

Der Fakultät für Human- und Sozialwissenschaften der Technischen Universität Chemnitz vorgelegte

Dissertation

zur Erlangung des akademischen Grades doctor rerum naturalium (Dr. rer. nat.) im Promotionsfach Psychologie

Jan-Philipp Stein (geboren am 15.06.1988 in Mannheim)

Tag der Einreichung 30.04.2018 Tag der Disputation 24.04.2019

Erstgutachter Prof. Dr. Peter Ohler, Technische Universität Chemnitz Zweitgutachter Prof. Dr. Josef Krems, Technische Universität Chemnitz

I

Acknowledgments

“We give each other souls.”

So beantwortet zumindest Rodney Brooks, emeritierter Professor des Massachusetts Institute

of Technology, eine grundlegende metaphysische Streitfrage. Seelen sind letztlich wie

Farben: Nehmen wir sie wahr, weil es sie gibt – oder gibt es sie nur, weil wir sie

wahrnehmen?

Bevor sich die folgende Arbeit (im weiteren Sinne) mit diesem Gedankenexperiment

auseinandersetzt, möchte ich sie den Menschen widmen, die mir meine Seele nicht nur

verleihen, sondern sie beflügeln, anspornen und mir teuer werden lassen.

Allen voran gilt das meinem Mann, dessen Liebe stets Kraft für neue Bemühungen spendet:

사랑해요, 여보! Ebenso danke ich meiner Familie – meinen Eltern und Großeltern, Heide,

Lina, Frank, Markus, Stefanie, Inken, Matthias, sowie den Schwiegereltern und Daeyeon in

Korea – für die nicht minder wertvolle Gewissheit, ein Zuhause in allen Himmelsrichtungen

zu haben.

Einen ganz besonderen Dank möchte ich natürlich meinem Doktorvater Prof. Dr. Ohler

aussprechen. Danke, Peter, für die Freiheiten, die Du mir gelassen hast, wenn ich es mir

wünschte, und jeden kritischen Ratschlag, den Du bereithieltest, wenn ich es brauchte! Die

Zeit bei Dir war eine einmalige Erfahrung, die mich immer begleiten wird. Für die

Zweitbegutachtung dieser Dissertation danke ich ebenfalls Prof. Dr. Krems ganz herzlich. II

Der Deutschen Forschungsgemeinschaft und der TU Chemnitz bleibe ich tief verbunden für ihre großzügige Unterstützung, die mich drei Jahre lang angstbefreit und fokussiert an diesem

Vorhaben arbeiten ließ. Ganz sicher werde ich die vielen Eindrücke aus anderen Teilen der

Welt, die mir das DFG-Graduiertenkolleg „CrossWorlds“ schenkte, niemals vergessen.

Selbiges gilt natürlich auch für die anregenden und aufmunternden Gespräche mit

Kolleginnen und Kollegen. Ich bedanke mich vor allem bei Timo für seine unerschütterliche

Ruhe und Hilfsbereitschaft, sowie bei Benny, Daniel und Kevin; nicht nur für die gemeinsamen wissenschaftlichen Erfolge, sondern vor allem für viele lustige Momente in

Chemnitz, Fukuoka und San Diego! Zugleich sollen auch die Begegnungen mit Georg,

Katrin, Madlen, Matthias, Niki, Sabine und Susanne nicht vergessen sein, die alle auf ihre

Weise zu einem entspannten Arbeitsalltag beigetragen haben. Danke dafür.

Völlig außer Frage steht, dass das Abenteuer „CrossWorlds“ ohne Alexander, Benedikt und

Ingmar nicht halb so unbeschwert gewesen wäre. Euer scharfsinniger Humor war unbezahlbar – egal ob er bei Team-Meetings, Seilbahnfahrten oder der obligatorischen

Feierabendpizza zum Einsatz kam.

„Freunde sind die Familie, die man sich aussucht.“ Auch wenn meine Herkunftsfamilie wenig Wünsche offenlässt, erfüllt es mich mit tiefer Dankbarkeit, dass sich Martin, Gregor,

Charlaura, Eva, Marijke, Caro, Romy, Kai, Marlen, Heajung, Jiao und Katharina dazugesellt haben. Ihr seid der Jackpot! (Und ohne Def geht seit zwanzig Jahren sowieso nichts.)

Zu guter Letzt widme ich diese Arbeit dem Kater, der jedes an diesem Schreibtisch geschriebene Wort kritisch gegengelesen hat. Ich hoffe, Du bist mit dem Ergebnis einverstanden, Gilbert. III

Summary (German)

Inspiriert von den enormen technologischen Fortschritten der letzten Jahrzehnte beschäftigt sich die vorliegende Dissertation mit der Wahrnehmung emotionssensitiver künstlicher Intelligenz (KI) und nimmt dabei vor allem die Bedeutung von erlebter

Menschenhaftigkeit in den Fokus. Bei der Entwicklung meiner Hypothesen finden Befunde von kognitionswissenschaftlichen Studien ebenso Beachtung wie philosophische und kulturpsychologische Literatur. Tatsächlich gipfelt die interdisziplinäre Recherche in einer deutlichen Vermutung: Da das Erkennen von Geistes- und Gefühlszuständen in anderen

Lebewesen von vielen Personen als Kernkomponente ihrer menschlichen Einzigartigkeit betrachtet wird, laufen affektive Technologien mit ähnlichen Fähigkeiten Gefahr, unheimlich oder abstoßend zu wirken – sie werden zu unliebsamen ‚Eroberern‘ einer urmenschlichen

Domäne.

In einer ersten empirischen Studie (N = 92) finde ich Hinweise auf die Gültigkeit der getroffenen Annahme. Es zeigt sich, dass empathisch agierende Figuren in einer virtuellen

Umgebung deutlich stärkere Ablehnung hervorrufen, wenn Probanden dahinter keine

Menschen (Avatare), sondern ein hochkomplexes KI-System (autonome Agenten) vermuten.

Nachdem persönliche Beobachtungen und Aussagen der Versuchspersonen wiederholt auf die Rolle von Bedrohungswahrnehmnungen hindeuten, stelle ich diese Variable ins Zentrum einer zweiten Untersuchung. Mithilfe vertiefender Literaturrecherche entwickle ich ein

„Model of Autonomous Technology Threat“ („Modell der Bedrohung durch autonome

Technologie“), welches zwei potentielle Determinanten von Technologieaversion kombiniert:

Während die distale Facette übergeordnete Einstellungen zu menschlicher Einzigartigkeit bündelt, wird zudem ein starker Einfluss von unmittelbaren (proximalen) Wahrnehmungen postuliert, etwa dem erlebten Ausmaß an situativer Kontrolle oder der Sorge um die eigene physische Unversehrtheit. IV

Ein neues Virtual-Reality-Experiment, in dem Probanden (N = 125) mit einem

angeblich autonomen – tatsächlich jedoch ferngesteuerten – Agenten interagieren, dient dazu,

das Modell statistisch zu überprüfen. Obgleich sich durchaus Hinweise zur Bedeutsamkeit

des distalen Pfades abzeichnen, erweisen sich vor allem proximale Faktoren als signifikanter

Prädiktor von Bedrohungsempfinden und infolgedessen der Nutzerakzeptanz.

Mit einer dritten Studie soll schließlich exploriert werden, inwieweit die

Wahrnehmung von Menschenartigkeit auch bei nicht verkörperlichten, emotionssensitiven

KIs Relevanz besitzt. Zugleich rücke ich nun Kultureinflüsse in den Fokus, indem ich

Versuchspersonen verschiedener kultureller Herkunft (N = 89) eine automatisierte

Emotionserkennungssoftware kennenlernen lasse. Während chinesische Probanden, deren

kulturelle Sozialisation tendenziell ein breiteres Verständnis von ‚Beseeltheit‘ umfasst, nur

kurzzeitig vom Feedback der affektiven Software in Erregung versetzt werden, lässt sich bei

deutschen Versuchspersonen ein deutlich längerer Anstieg physiologischer Aktivierung

feststellen. Zugleich zeichnet die Messung subjektiver Empfindungen ein überraschendes

Bild: Die emotionssensitive Maschine – in diesem Fall ein abstrakter, mechanischer Kasten –

wird umso positiver bewertet, je mehr Menschenhaftigkeit Probanden in ihr erkennen.

Angesichts der drei durchgeführten Studien komme ich zu dem Schluss, dass lediglich

Szenarien mit verkörperlichter KI im Sinne des entwickelten Bedrohungsmodells beide Pfade

bedienen, was für eine eindeutig aversive Reaktion erforderlich sein könnte. Die Gestaltung

der Software in Studie 3 spielte unterdessen für das unmittelbare Kontrollerleben der

Versuchspersonen keine Rolle; eine mögliche Erklärung, warum sich Attributionen von

Menschenähnlichkeit hier sogar positiv in den subjektiven Evaluationen niederschlugen.

V

Summary (English)

Inspired by the enormous technological advancements of previous decades, this

doctoral thesis revolves around users’ perception of emotion-sensitive

(AI), with particular focus on the role of human likeness attributions. For the development of

my hypotheses, I acknowledge both cognitive scientific as well as philosophical and cultural

psychological literature. Eventually, my interdisciplinary review culminates in one central

assumption: Since many people regard the recognition of mental and emotional states in other

entities as a core component of their human uniqueness, affective technology with similar capabilities runs the risk of being seen as uncanny or aversive—turning into a discomforting

‘challenger’ of an inherently human domain.

Indeed, a first empirical study (N = 92) provides evidence for the validity of my

hypothesis. My findings show that empathically acting characters in a virtual environment are

met with much stronger aversion if participants suspect them to be controlled by highly

complex AI (autonomous agents) instead of other humans (avatars). Acknowledging

statements from my participants which repeatedly hint towards the importance of threat

perceptions, I turn this into the main subject of a second study. Based on additional

literature research, I develop a “Model of Autonomous Technology Threat”, which combines

two potential determinants of technology aversion: Whereas the model’s distal facet

summarizes overarching attitudes about human uniqueness, I also postulate a strong influence

of immediate (proximal) perceptions, including the experience of situational control or users’

concern about their immediate physical well-being.

Using yet another (VR) setting, I ask participants (N = 125) to interact

with an allegedly autonomous—in remotely controlled—digital agent under different

conditions, which allow for a statistical comparison of my proposed model. Although the

yielded results do lend some support to the validity of the distal path, it is mostly the VI proposed proximal factor that connects to participants’ threat experience, which in turn emerges as a negative predictor of technology acceptance.

Lastly, a third study is designed to investigate whether perceptions of human likeness possess any relevance in the context of disembodied emotion-sensitive AI. Moreover, I now focus on potential cross-cultural differences, inviting participants from different cultural backgrounds (N = 89) to familiarize themselves with automatic software.

Whereas Chinese participants, whose cultural socialization encompasses a much broader of ‘animacy’, only show brief arousal after feedback from the affective software, I observe a significantly longer increase in physiological activity among German participants. At the same time, the obtained subjective measures paint a surprising picture:

The more participants attribute human likeness to the emotion-sensitive machine—an abstract mechanical box in this scenario—the higher they actually rate the technology’s attractiveness.

By summarizing the three conducted studies, I reach the conclusion that only AI stimuli involving elaborate embodiments activated both paths of the developed threat model, which might be a requirement for a distinctly aversive reaction. The presentation of the abstract software in Study 3, on the other hand, did not influence participants’ immediate control perceptions—a possible explanation as to why attributions of human likeness turned out as a positive predictor for the subjective evaluations in this experiment. VII

List of Abbreviations

AFER automatic facial emotion recognition

ANOVA analysis of variance

ANS autonomic nervous system

AI artificial intelligence bpm beats per minute

CFA confirmatory factor analysis

HCI human-computer interaction

HMD head-mounted display

HRI human- interaction

MANOVA multivariate analysis of variance

ToM Theory of Mind

UV uncanny valley

VR virtual reality

VIII

List of Included Publications

This cumulative dissertation contains three articles that have already been published in peer- reviewed scientific journals. The following list provides the formal references for all included articles, including their digital object identifiers.

Chapter 3: Stein, J.-P., & Ohler, P. (2017). Venturing into the uncanny valley of mind— The influence of mind attribution on the acceptance of human-like characters in a virtual reality setting. , 160, 43–50. https://doi.org/10.1016/j.cognition.2016.12.010

Chapter 4: Stein, J.-P., Liebold, B., & Ohler, P. (2019). Stay back, clever thing! Linking situational control and human uniqueness concerns to the aversion against autonomous technology. Computers in Human Behavior, 95, 73–82. https://doi.org/10.1016/j.chb.2019.01.021

Chapter 5: Stein, J.-P., & Ohler, P. (2018). Saving face in front of the computer? Culture and attributions of human likeness influence users’ experience of automatic facial emotion recognition. Frontiers in Digital Humanities, 7, 18. https://doi.org/10.3389/fdigh.2018.00018

IX

Table of Contents

Acknowledgments I Summary (German) III Summary (English) V

List of Abbreviations VII List of Included Publications VIII Index of Figures XI Index of Tables XII

1 Theoretical Background 1

1.1 Animacy and Sentience 3 1.2 Human Uniqueness 5 1.3 Control and Threat Perceptions Towards Technology 8 1.4 The Uncanny Valley (of Mind) 11 1.5 State of the Art in Artificial Intelligence Research 14

2 Overview of the Dissertation Studies 16

2.1 Studying the Uncanniness of Artificial 18 2.2 Assembling the Model of Autonomous Technology Threat 19 2.3 Advancing to Disembodied AI and the Role of Culture 21

3 Study I: “Venturing into the Uncanny Valley of Mind—The Influence of Mind Attribu- tion on the Acceptance of Human-Like Characters in a Virtual Reality Setting” 25

3.1 Introduction 27 3.2 Methods 33 3.3 Results 39 3.4 Discussion 43

4 Study II: “Stay Back, Clever Thing! Linking Situational Control and Human Uniqueness Concerns to the Aversion Against Autonomous Technology” 48

4.1 Introduction 50 4.2 Method 58 4.3 Results 66 4.4 Discussion 72 X

5 Study III: “Saving Face in Front of the Computer? Culture and Attributions of Human Likeness Influence Users’ Experience of Automatic Facial Emotion Recognition” 78

5.1 Introduction 81 5.2 Method 91 5.3 Results 100 5.4 Discussion 107

6 General Discussion 112

6.1 A Tentative “Uncanny Valley of Mind” 115 6.2 Limitations and Future Work 116 6.3 Concluding Remarks 118

References 121

Curriculum Vitae 149 List of Publications 152

Selbstständigkeitserklärung (Affidavit) 154

XI

Index of Figures

Figure 1 Initial concept of the Uncanny Valley of Mind, including examples for 12 both biological (green) and artificial (black) entities.

Figure 2 Extended concept of the Uncanny Valley of Mind, suggesting the 13 additional influence of an entity’s physicality.

Figure 3 Uncanny valley model (redrawn from Mori, 1970). 27

Figure 4 Screenshot from the presented VR scene. 36

Figure 5 The study’s 2  2 factorial design and excerpts from the corresponding 37 narratives.

Figure 6 Average eeriness ratings for the different mind attributions (error bars 41 reflect +/– 1 standard error of the means).

Figure 7 Threat proximity as a common dimension of previous conceptualizations 54 (e.g., MacDorman & Entezari, 2015; Złotowski et al., 2017).

Figure 8 Model of Autonomous Technology Threat. 55

Figure 9 The study’s 2×2 between-subject design. 58

Figure 10 Experimental manipulation of interpersonal distance, 4 meters (left) vs. 60 0.8 meters (right).

Figure 11 Coefficients obtained from path analysis (* p < .05, ** p < .01). 69

Figure 12 The study’s between-subject design. Table cells contain each condition’s 92 theoretical implications for cultural display rules.

Figure 13 MultiSense used for the deceptive narrative of emotion 94 recognition.

Figure 14 Procedure to convey the deceptive result sheet after the experimental 95 task. (A) The login screen hosted on a local web server. (B) Result sheet with fictitious data.

Figure 15 The four experimental groups’ heart rate changes in bpm, compared to 103 their respective baseline value.

Figure 16 Standardized regression coefficients for the relationship between culture 106 and perceived technology attractiveness as mediated by ascribed technology human likeness. (*p < .05; **p < .01).

Figure 17 Modified concept of the Uncanny Valley of Mind. 115

XII

Index of Tables

Table 1 Research foci and bibliographic references of the studies included in this 17 thesis.

Table 2 Methodological comparison of the studies included in this thesis. 17

Table 3 Means and standard deviations on the three uncanny valley indices for 41 each condition.

Table 4 Zero-order correlations between measured variables. 67

Table 5 Means and standard deviations obtained for the measured variables. 68

Table 6 Multi-group confirmatory factorial analyses to check translated scales for 99 measurement invariance.

Table 7 Descriptive for heart rates and relative heart rate changes 102 between the three measuring points.

Table 8 Descriptive statistics for self-report measures. 105

THEORETICAL BACKGROUND 1

1 | Theoretical Background

Lifeless objects turning into animate, human-like entities have long evoked marvel and hope, but also deep-rooted anxiety among human society. Accordingly, the ambiguous fascination of things becoming beings can be traced throughout millennia of arts, media, and folklore—from the golems in ancient Jewish mythology or Goethe’s broomsticks disobeying the eponymous “Sorcerer’s Apprentice” to hordes of Frankensteinian creations and roomfuls of children toys coming to life in the beloved movies of modern times. It is on account of this timelessness that the spectacle of animated inanimacy has become all but brandished into the collective cultural mind—in particular among members of Western societies.

Despite people’s untiring interest in creations of unexpected aliveness, however, actual occurrences of living objects have mostly been confined to the realms of imagination or the hands of skilled puppeteers for the better part of history—if fleeting moments of doubt about new inventions, such as ‘fire-breathing’ steam trains, may be ignored. Even considering the first sophisticated humanoid of the 20th century, observers would quickly find themselves back in the world of machines, stuck between mechanistic boxes and pre-programmed behavioral patterns. Arguably, a 1970’s could make people suspend their disbelief for a while (Borody, 2013), but it never truly contested the man-machine distinction that had persisted throughout the ages.

It is all the more astounding, in light of this ancient living versus non-living dichotomy, that a single handful of decades has now been sufficient to challenge fundamental definitions such as animacy, humanness, or having a mind—as advances in the field of have been giving birth to the first truly intelligent and self-controlled technologies. With the invention of neural network technology and deep in particular, digital systems are now able to behave in ways that many had deemed impossible for an artificial creation just 50 years ago. Inspired by the design of the human brain, these THEORETICAL BACKGROUND 2

new forms of artificial intelligence (AI) not only allow computers to learn about the world in

an autonomous way, but may sometimes even produce completely different outcomes than what was anticipated by their human creators (Horton, 2016). As soon as a digital creation of this complexity is also coupled with the advanced visualization methods of the 21st century

(e.g., nearly photorealistic virtual embodiments), no one could fault observers for coming

under the impression that they interact with just another human being; especially since most people have an innate tendency to treat computers as social actors anyway (Reeves & Nass,

1996).

Despite the impressive processing power of contemporary AI, however, recent research has argued that it is actually not the high agency of artificial systems (defined as their capability to learn, plan, and act autonomously), but in fact the impression of new-found experience—i.e., their ability to feel—that truly challenges people’s understanding of the world (Appel, Weber, Krause, & Mara, 2016; Gray & Wegner, 2012). Since experience constitutes a facet of mind that is usually attributed to animalistic beings only (Gray, Gray &

Wegner, 2007; Wegner & Gray, 2016), cognitive scientists have suggested that emotional AI approaches an insurmountable acceptance threshold: Society may approve of a technology executing a given task on its own—but appreciate it much less if the system feels something while doing so.

Admittedly, considering even the most sophisticated computer’s ‘emotional’ programming as sincere, human-like feeling might be a bit of a stretch, at least at this point in time. In most cases, current artificial intelligence is merely designed to emulate processes that are typically human, pursuing the vague objective to facilitate more efficient means of human-computer interaction (HCI). Nevertheless, this thesis argues that the remaining artificiality of emotion-sensitive AI is of little relevance for the cognitive and affective reactions of the common user. Considering that humans heuristically rely on mental THEORETICAL BACKGROUND 3

categories to anticipate possible behaviors from different types of entities (Cohen &

Lefebvre, 2017), it becomes less important whether a computer actually feels or simply

pretends to do so—its observable actions should suffice to defy previous expectations about

machines and trigger new attribution processes. Once this reevaluation process veers too close to the dystopian scenarios depicted in the arts and media, users may feel a strong impulse to insist on the human dominance over technology—which might open up a proverbial Pandora’s box of HCI-related problems. At the same time, the potential tension

between humans and emotionally capable AI also offers an intriguing field of research,

especially since society can still decide on the restrictions it wants to impose on digital .

To fulfill this task in a well-informed manner, psychological investigations of people’s needs,

fears and desires in a life surrounded by technology are absolutely indispensable; a fact that

greatly motivated the preparation of this dissertation.

1.1 | Animacy and Sentience

Among the most basic attributions that occur as humans perceive any given stimulus

are those of animacy—“Does it live?”—and sentience—“Does it experience?”. However,

since both concepts reach far into various areas of expertise (including psychology, biology, neuroscience, , philosophy, and religion), there is little academic consensus on the boundaries that confine both terms. Complicating matters further is a cultural rift that runs through potential definitions of animacy and sentience, as the Western rooting in Cartesian dualism, which sorts physical matter and spiritual essence into two distinct categories and exclusively reserves the latter for humanity (Descartes & Cottingham, 1986), contrasts starkly with conceptualizations that have developed in other parts of the world (e.g., Descola,

1994; Ojalehto & Medin, 2015; Varela, Rosch, & Thompson, 2016). As such, the question whether an entity possesses any semblance of aliveness might yield quite contrary answers depending on the responder’s cultural and philosophic background. Even more so, these THEORETICAL BACKGROUND 4

fundamental differences seem to persist despite the on-going globalization and blurring of

cultural differences. For instance, several East Asian cultures and religions have preserved

their connection to animism, a belief system that refrains from a dichotomization between

things that are ‘supposed’ to live and those that are not (Århem, 2016; Kitano, 2007).

Although modern-day animism has somewhat evolved from its more esoteric origins, its

remanining derivatives such as Japanese Shintō mythology still feature countless phenomena

that elude a clear description as an object or being, instead providing a vast spectrum of

animacy (Kazuhiko, 2017). Arguably, some members of current Japanese society might dismiss this aspect of their culture as outdated or folkloristic (Yamakage, 2012), but anthropologists nevertheless suggest that animistic beliefs have forever permeated the East

Asian worldview—living on through novel forms such as techno-animism (Allison, 2006).

Irrespective of this cultural variety, however, English-speaking psychological literature has still attempted to posit some core components that may be used to identify both animacy and sentience in a stimulus, at least when following the strict perspective of scientific . According to these criteria, entities that move of their own volition and demonstrate basic interactive qualities will generally be regarded as animate (Michotte, 1963;

Tremoulet & Feldman, 2006), whereas those supplementing their autonomy with phenomenal —i.e., feelings and subjective experience (Gray, Gray & Wegner, 2007)—may be considered as sentient. And indeed, laboratory experiments have demonstrated that even simple objects such as marbles or two-dimensional geometric shapes evoke strong animacy perceptions among participants when moving around in an autonomous, goal-directed manner (e.g., Barrett & Johnson, 2003; Gao & Scholl, 2011). Still, in terms of the experience possessed by such stimuli, most people with a natural scientific worldview would eventually come to the conclusion of non-sentience; current neurobiological knowledge clearly dictates that a lack of cortical (or, at the very least, subcortical) structures also means no experience THEORETICAL BACKGROUND 5

(Panksepp, 2011; Steck & Steck, 2016). Followers of pantheistic beliefs or alternative

metaphysical theories, on the other hand, might firmly disagree.

As a contribution to the fields of psychology and cognitive science, this dissertation

naturally leans towards a rationalistic worldview. At the same time, I considered it important

to acknowledge that previous attempts to create indisputable definitions of animacy and sentience have simply shifted their conceptual ambiguity to the next lower level of abstraction. In all , people’s understanding of sentience might really depend on the agency or experience they perceive in an entity—it is just that these terms are once again completely open to interpretation. Thus, any definition of ‘having a mind’ that claims universal applicability will only hold up within the philosophical or scientific it was developed for. In consequence of this limitation, some theorists have even questioned whether attempts to define sentience might be meritless whatsoever (Searle, 1998). Instead, literature suggests that cognitive scientific research might be better off with a phenomenological approach (Gallagher & Zahavi, 2008), which focuses more on people’s individual perception of mind than on the development of supposedly objective definitions that are destined to fail.

1.2 | Human Uniqueness

A that directly builds upon the discussed perceptions and therefore inherits

most of their conceptual complexity is anthropomorphism—defined as the attribution of

human-like characteristics, , motivations, and intentions to non-human entities

(Epley, Waytz & Cacioppo, 2007). Etymologically, the term already bespeaks its embedment

in Western homocentrism, as it collects the possession of volition and emotionality under the

mantle of strictly anthropomorphous, i.e., human-natured qualities. Considering the reviewed

variety of animacy and sentience conceptualizations, this designation certainly becomes

rather questionable. Nonetheless, it has to be noted that anthropomorphism (both as a THEORETICAL BACKGROUND 6

dispositional and a process-related variable) has provided a rather practicable approach for

researchers and technology developers (e.g., Epley, Waytz, Akalis, & Cacioppo, 2008; Tam,

Lee, & Chao, 2013; Waytz, Heafner, & Epley, 2014), often uncovering surprising effects that

result from people ‘recognizing themselves’ in non-human creations. Ultimately, the concept does work; researchers just need to acknowledge that the level of self-recognition is not only determined by the subject matter at hand, but also intrinsically connected to culturally acquired worldviews.

Again and again, the introduction of this thesis leads back to contemplations of culture, but I argue that the importance of this subject for my research questions cannot be stressed enough. Based on my interdisciplinary review of literature, it actually seems that all of the discussed perceptions—animacy, sentience, and anthropomorphism—as well as their role in successful HCI might involve one crucial, cultural antecedent: The extent to which people insist on humanity’s uniqueness as a species.

“The Western man puts all his pride in this delta which is supposed to be specifically human, a testimony of its divine origins,” argues Kaplan (2004, p. 477) in a rather compact description of Christian anthropocentrism. And indeed, despite the on-going religious decline observable in the West (Franck & Iannaccone, 2014), empirical research shows strong support for Kaplan’s argument, as US-American and European HCI studies repeatedly document participants’ concerns about human distinctiveness in relation to their technology acceptance (Ferrari, Paladino, & Jetten, 2016; MacDorman & Entezari, 2015; Złotowski,

Yogeeswaran, & Bartneck, 2017). More often than not, the apprehensiveness towards digital minds seems to revolve around their level of emotional experience, as people consider the expression and recognition of feelings a differentia specifica of humans—or at least as

privileges that are reserved for biological minds. Admittedly, from a 21st century perspective,

it has become common practice to ascribe at least a limited form of emotionality to other THEORETICAL BACKGROUND 7

animal species as well; the medieval interpretation of livestock as insensate ‘meat bags’ has

mostly been overwritten by the hands of scientific enlightenment. At the same time,

psychological research indicates that most people still assume only the simplest forms of emotional experience to occur among animal life: Unconscious streams of temporary states that are completely devoid of meanings and values, morality, or any form of mutual awareness (Cartmill, 1990). Again, a significant percentage of variance in these perceptions has been shown to stem from principles of anthropocentrism, as species with human-like movement speed (Morewedge, Preston, & Wegner, 2007), anatomical design (Mitchell,

Thompson, & Miles, 1997), or lifespan (Wegner & Gray, 2016) are typically ascribed more ability to experience than, for instance, insects or sea animals. Unsurprisingly, these effects seem to turn out much weaker for animals that are traditionally used in human consumption, since many people prefer to recognize only little sentience in these species irrespective of their biological features (Bastian, Loughnan, Haslam, & Radke, 2011).

In any case, a traditional view on all non-human species that has remained to this day is that—apart from feeling emotions—they are usually ascribed only limited capabilities to recognize or interpret such states in each other (Heyes, 2015). Even though zoological experiments actually provide evidence for social in some animals, such as great apes (Hare, Call, Agnetta, & Tomasello, 2000; Suddendorf, 2013) or dolphins (Tomonaga &

Uwano, 2010), the reported studies are often inconclusive and subject to methodological critique (Shumaker & Swartz, 2002) so that they have not yet sufficed to change the mainstream perception of animal minds. Surely, in layman terms, people might say that a dog

can ‘sense’ the mood of its owner, or that a lioness ‘worries’ about the well-being of her

cubs; but considering the depth and complexity of such empathic behaviors, the skeptics’

conclusion is usually the same: In dubio contra reum—without proper evidence, attributions

of mind are denied. THEORETICAL BACKGROUND 8

However, there might also be another explanation accounting for people’s resistance to share abilities such as social cognition with non-human entities, one that relies less on hard empirical data than on abstract principles of human . Specifically, I postulate that many individuals (at least those with a Western socialization) may have an inherent resistance against perceiving empathy in creations other than themselves, because it means protecting one of the last ‘bastions’ of their human uniqueness. After several industrial revolutions, society may have come to terms with the fact that machines are much faster, stronger, and more effective than a human being—once planes surpassed the speed of sound and microchips executed a billion calculations per second, some kind of inferiority just had to be accepted. But still, throughout countless iterations of technological innovation, the human species could always hold on to one exclusive characteristic that epitomized its significance on Earth: The possession of complex feelings, emotional awareness, and empathy. In a more provocative reading, this could even be described as a form of species-related ‘narcissism’— a manifestation of human exceptionalism that not only determines people’s view on plants and animals, but on machines, technology, and virtual entities as well. Combined with strict

Christian tradition, the resulting mindset carries an almost normative connotation: Intelligent machinery should never reach human levels of mindfulness because it would signify a blasphemous attempt to become God-like (Gee, Browne, & Kawamura, 2005).

1.3 | Control and Threat Perceptions Towards Technology

Following the presented line of thought, my thesis offers its presumed answer to the

question how people might feel as soon as novel AI systems acquire their own forms of

emotional understanding. Unlike animals, whose alleged simple-mindedness and lack of

articulate language imply little threat to humanity’s role as the ‘crown of creation’, emotion- sensitive technology is usually designed to broadcast its impressive abilities in an open manner—leaving less doubt about its potential to challenge a final pillar of human identity. THEORETICAL BACKGROUND 9

Proceeding from this argument, I further suggest that the empathic machine inevitably

turns into an agent of morality; after all, it is a basic ethical principle that the understanding

of another being’s experience creates moral responsibilities for the consequences of one’s own behavior (Young & Waytz, 2013). Assuming that digital systems are indeed on the brink of genuine perspective-taking—i.e., about to develop their own Theory of Mind (ToM)— people have every reason to hope that this new-found insight will inspire the technology to behave in a benevolent and caring manner. At the same time, the autonomous and often unpredictable of AI systems is becoming increasingly well-known to the broader public, so that observers might also come to fear the emergence of new artificial moral systems that do not match the human way of thinking. Up until recently, this sorrow has mostly belonged to the realms of science fiction—a genre that is, after all, characterized by a rather dystopian outlook on the future of HCI. However, with the technological breakthroughs of the current decade, the prospect of potentially diverging human–machine ethics has actually turned into a highly relevant research area (Hauer, 2018), for instance in regard to driverless cars (Lin, 2016). For human society, even the most vague possibility that machines might someday devalue the significance of human life can only herald disastrous outcomes; in consequence, I suggest that concerns about human uniqueness are not only a matter of hazy philosophical debate, but mostly a very ‘real’ manifestation of people’s desire to remain unharmed and in control of their machinery.

This is not to say that the loss of human identity could, in itself, ever constitute a pleasant thought—it is hard to imagine that anyone would enjoy the of becoming exchangeable for or even inferior to a digital counterpart. Accordingly, a recent study about potential job replacements by robots revealed that participants were significantly more anxious about the idea of losing emotion-related jobs to machines than when they considered abandoning physical work (Waytz & Norton, 2014). Clearly, findings like these illustrate that THEORETICAL BACKGROUND 10 people’s contemplation of lost human distinctiveness cannot be separated from more palpable fears—which is why several authors have suggested that truly autonomous AI technology will automatically be seen as a threat to human safety and resources as well (Kang, 2009;

Złotowski, Yogeeswaran, & Bartneck, 2017).

Highlighting yet another crucial influence, previous research has shown that it is mostly scenarios involving robots and virtual agents—i.e., AI with an actual embodiment—in which people’s perception of digital minds translates into measureable aversion effects

(Kim & McGill, 2011; MacDorman & Entezari, 2015; Stein & Ohler, 2017a; Złotowski,

Yogeeswaran, & Bartneck, 2017). While an emotionally aware desktop computer might already seem ‘creepy’ just by foreshadowing its potential to reshape societal arrangements

(an effect that I have termed distal threat, see Stein, Liebold, & Ohler, 2019), observers’ reactions are likely to turn out much more negative once the artificial mind is put into a tangible body—with arms that can grab or legs that may step just a little too close (proximal threat). This effect may become even stronger once mind–body interaction effects are considered: Since entities with a humanoid design are usually attributed with much more emotional experience in the first place (Broadbent et al., 2013; Gray & Wegner, 2012), it may be assumed that mind- and body-related threat perceptions would ultimately reinforce each other. In theoretical consequence, I suggest that AI developers will have to consider both abstract of human uniqueness as well as many situational factors (e.g., interpersonal distance, type of embodiment) if they want to keep their customers’ threat experience—thus, their resentment towards new technologies—at a minimum.

THEORETICAL BACKGROUND 11

1.4 | The Uncanny Valley (of Mind)

A model from the area of human-robot interaction (HRI), which I found very suitable

to illustrate the developed assumptions, is the well-known Uncanny Valley (UV) hypothesis by Japanese roboticist Masahiro Mori (1970). Inspired by the fearful reactions he often observed towards prosthetic limbs, Mori suggested that increasing the human likeness of an artificial object could only foster people’s fondness of it up to a threshold of near-perfect realism—before any effort to make the creation yet more human-like would inevitably push it into a valley of uncanniness; a somewhat familiar, but eventually unpleasant weirdness1.

For most of the time since its inception, the UV model has mainly been used to address potential problems in the visual design of human-like replicas. In the corresponding studies, researchers have offered an abundance of explanations as to why people might

experience the cold, unpleasant feeling towards certain stimulus configurations, including

cognitive dissonance effects (Yamada, Kawabe, & Ihaya, 2013), evolutionary fears of

pathogens (Ho, MacDorman, & Pramono, 2008) and unburied corpses (Moosa & Ud-Dean,

2010), the confrontation with animalistic mortality and subsequent terror management

techniques (MacDorman & Ishiguro, 2006), as well as the perception of emotional

detachment due to an entity’s lifeless eyes (Tinwell, 2014). Regardless of their individual

reasoning, however, most authors have focused exclusively on visual—i.e., aesthetic or

gestalt psychological—factors in their pursuit of Mori’s phenomenon. Although this

approach has led to a large number of insightful and often surprising findings, I still consider

it an unnecessary limitation in the application of the UV model; a shortcoming which I

wanted my thesis to overcome.

1 In the original UV publication (1970), Mori uses the Japanese neologism shinwakan to describe the outcome variable of his hypothesis, which was initially translated as “familiarity“. However, the semantic equivalence of this translation has been criticized by several authors (e.g., Bartneck, Kanda, Ishiguro, & Hagita, 2009; Ho & MacDorman, 2010), so that the model’s y-axis is now usually described as “affinity”, “(lack of) eeriness”, or “valence of response”. THEORETICAL BACKGROUND 12

Pending some minor adjustments, I postulate that Masahiro Mori’s hypothesized

pattern may work just as well to illustrate non-visual factors, in particular regarding the role of mind perception. If the x-axis is used to plot the complexity of attributed mental abilities instead of the visual human likeness of a stimulus, I expect the traditional UV pattern to emerge rather unchanged (Fig. 1), offering a graphic juxtaposition of artificial (and, if needed, biological) creations as they increase in emotional prowess.

Figure 1. Initial concept for an Uncanny Valley of Mind, including examples for both biological (green) and artificial (black) entities.

Apart from the UV’s eponymous shape, Mori further differentiated between static and

moving stimuli in his original model (1970) because his research had indicated that adding

movement to a human-like creation could significantly increase the slopes (both positive and

negative) of the hypothesized graph. For the emerging notion of an Uncanny Valley of Mind,

I propose a similar two-fold gestalt—albeit not in terms of entity movement. Adhering to the

developed theory that both distal and proximal threat perceptions have to work together to

manifest a significant aversion against emotion-sensitive AI, I instead suggest that the depth

of the mind-related UV increases with a stimulus’s amount of physical features, i.e., its THEORETICAL BACKGROUND 13

physicality (Fig. 2). After all, it should be expected that a sophisticated AI lingering in the cloud evokes much less experience of threat than the same synthetic mind put into a complete with hydraulic extremities and actual body strength. On the other hand, I consciously chose not to call this dimension “human likeness” or “human-like embodiment”—even though previous studies argue that highly anthropomorphic machines will trigger stronger threat perceptions than mechanical or mixed designs (Ferrari,

Paladino, & Jetten, 2016; Kim & McGill, 2011). In my understanding, it seems just as likely that certain mechanomorphic shapes could surpass bipedal humanoids in terms of proximal threat experience; clearly, a robotic with six powerful extremities should evoke much more threat than a room service android with a deliberately cute design. Consequently, my extended Uncanny Valley of Mind conceptualization uses the notion of physicality as a relatively broad umbrella term to encompass factors such as body shape and size, virtual or actual embodiment, physical strength, and spatial arrangement—without denying that human- like features could further boost the observed effects.

Figure 2. Extended concept for an Uncanny Valley of Mind, hypothesizing an additional influence of an entity’s physicality on users’ response.

THEORETICAL BACKGROUND 14

1.5 | State of the Art in Artificial Intelligence Research

Participants of psychological studies in the 20th century would have had some

difficulty believing a study conductor who introduced them to a “fully autonomous” or

“emotion-sensitive” AI in a laboratory experiment. Although the theoretical development of

elaborate digital minds reaches back as far as the 1940s and 1950s (McCulloch & Pitts, 1943;

Rosenblatt, 1958), it is only with the dramatic technological advancements of recent years that AI systems have actually proven successful in emulating the tremendously complex abilities of the human brain (Hassabis, Kumaran, Summerfield, & Botvinick, 2017; LeCun,

Bengio, & Hinton, 2015). At last, developers of digital minds not only have the knowledge, but also the computational power to implement neuroscientific insights about human information processing into their technology—constructing so-called artificial neural networks. Among many other astonishing abilities, these AI systems may include selective attention mechanisms (e.g., Hong, Yamins, Majaj, & DiCarlo, 2016) as well as highly parallelized stochastic memory structures that constantly readjust the interpretation of new stimuli (e.g., Mnih et al., 2015). Furthermore, unlike earlier forms of machine learning technology, they are usually extremely multi-layered—so-called deep learners—which lets them approach a core principle of true intelligence: To understand the world in terms of hierarchical concepts. Based on this crucial advancement, modern-day AI continues to surpass previous variants by far, especially in regard to language understanding, sentiment analysis, and topic classification (LeCun, Bengio, & Hinton, 2015). Not only are current neural networks much more efficient in clustering sensory input on their own—which in turn minimizes the time-consuming teaching they require from their developers—they also show promising results in potentially life-saving use cases. For instance, an increasing number of studies have highlighted the successful application of deep learning technology as a diagnostic tool for cancer detection from tomographic images, which may be on par (Litjens THEORETICAL BACKGROUND 15

et al., 2016; Wang et al., 2016) or even outperform (Wang et al., 2017) human doctors, at

least in terms of sensitivity (rate of correct positives).

Naturally, the ability to execute such intricate tasks does not come without its price:

Most of today’s more capable AI systems are still housed by large-sized supercomputers and may take weeks to finish a proper initial training. At the same time, simpler forms of task- specific AI (sometimes called narrow AI) can already be encountered in much smaller and faster forms, for example as customer service agents, smartphone operating systems, or

Internet chatbots. Although these systems typically employ simpler machine learning techniques—so that they could hardly be described as AI with its ‘own mind’—their access to Big Data and ever-growing semantic still allows them to achieve astounding interactive qualities (Saenz, 2010), not least in the recognition and interpretation of human emotion. As millions of users provide their phones and other smart devices with never-ending streams of data on a daily basis—including behavioral habits, speech patterns, physiological parameters, semantic nuances, and so on—even AI hardware in the size of a 90-millimetre cube (Adee, 2015) is now able achieve a valid interpretation of human affect. Sometimes, the technology may even learn from people when it is supposed to be sleeping, as was the case with a smart speaker device developed by Google (Burke, 2017)—much to the chagrin of the technology’s customer base.

Despite my personal curiosity in the most complex achievements accomplished by modern AI, the empirical work of this thesis was mainly built around scenarios that could be found at the intersection between narrow AI and more professional neural network solutions.

Most of all, this decision was made after considerations of plausibility: Having researched the level of competence that is currently achieved by deep learning AI in the university context2,

2 For their provided insight, I especially thank René Richter and Tobias Höppner from the chair of Artificial Intelligence, Chemnitz University of Technology. OVERVIEW OF THE DISSERTATION STUDIES 16

my major goal was to devise technological settings of strong external validity and believability. For practical reasons, however, the designed stimuli were not presented in fully

functional versions, but had to be introduced by using a Wizard-of-Oz approach, in which

numerous efforts are expended to convince participants of a stimulus’s autonomy while

actually training remote operators to control the technology. To facilitate the necessary

deception, I relied on the common knowledge that universities tend to share large data

networks and often conduct joint research ventures; by embedding my scenarios within the

plausible narrative of large-scale international AI research, I argue that the shown

technologies became much more believable to the recruited participants. Supporting this

notion from an empirical perspective, all studies’ manipulation checks confirmed the success

of the respective deceptions.

2 | Overview of the Dissertation Studies

Based upon the theoretical and empirical findings reviewed in Chapter 1, I aimed at

advancing the understanding of causes, moderators, and consequences that might be relevant

within interactions between humans and emotional AI systems. As an advancement of extant

literature, I devised my experiments to focus less on computers with their own emotional

experience and more on technology that appears capable to recognize—and empathize

with—the emotional state of others. The rationale behind this decision was to highlight a

form of mental competence that not only violates the traditional differentiation between

computers and animalistic beings in general (e.g., the possession of emotions), but to explore

a concept that people consider as exclusively human, even when compared to other animals.

Based on the that empathy and emotional recognition are often considered in this

regard, I assume that they relate to a much greater perceived ‘distance’ between humans and

other creations than, for instance, the mere experience of affect. In consequence, empathic OVERVIEW OF THE DISSERTATION STUDIES 17 computers should appear particularly threatening, because they overcome an immense gap that traditionally separates man from everything else, at least in the Western mind.

In summary, three studies were conducted to scrutinize this overarching hypothesis, each focusing on a distinct sub-topic as listed in Table 1. By combining the insight gained from all three experiments, my thesis offers answers not only to the question if emotion- sensitive HCI might be problematic in terms of daily-life adoption, but also suggests a preliminary suggestion as to why this lack of acceptance may occur. For a first overview of methodological parameters of the conducted experiments, Table 2 can be consulted.

Table 1. Research foci and bibliographic references of the studies included in this thesis.

Study Research focus Reference

1 . Acceptance of embodied AI that Stein, J.-P., & Ohler, P. (2017). Venturing into the demonstrates artificial empathy uncanny valley of mind—The influence of mind attribution on the acceptance of human-like characters in a virtual reality setting. Cognition, 160, 43–50.

2 . Model of Autonomous Stein, J.-P., Liebold, B., & Ohler, P. (2019). Stay back, Technology Threat as a clever thing! Linking situational control and human potential explanation for the uniqueness concerns to the aversion against aversion against emotion- autonomous technology. Computers in Human sensitive AI Behavior, 95, 73–82.

3 . Acceptance of disembodied Stein, J.-P., & Ohler, P. (2018). Saving face in front of emotion-sensitive AI the computer? Culture and attributions of human . Cultural differences likeness influence users’ experience of automatic facial . The role of display rules emotion recognition. Frontiers in Digital Humanities, 7, 18.

All studies included in this thesis have been published in international peer-reviewed journals (Stein, Liebold, & Ohler, 2019; Stein & Ohler, 2017a; Stein & Ohler, 2018a).

Moreover, scientific talks on the conducted experiments have been held at the 67th Annual

Conference of the International Communication Association (Stein, Liebold & Ohler, 2017), the 68th Annual Conference of the International Communication Association (Stein & Ohler,

2018b), and the 10th Conference of the Media Psychology Division of the German Society OVERVIEW OF THE DISSERTATION STUDIES 18

for Psychology (Stein & Ohler, 2017b). All of these conference contributions were subject to peer-review processes.

Table 2. Methodological comparison of the studies included in this thesis.

Study Sample size Statistical approach Outcome measures

1 N = 92 . Two-factorial ANOVA . Uncanny valley indices (Ho & MacDorman, 2010)

2 N = 125 . Two-factorial MANOVA . Uncanny valley indices (Ho & MacDorman, . Mediation analysis 2010) . Human uniqueness concerns (self- . Path analysis developed) . Hierarchical regression . Situational control (self-developed) . Threat experience (self-developed)

3 N = 89 . Two-factorial MANOVA . Attractiveness and human likeness indices . Hierarchical regression (Ho & MacDorman, 2010) . Cardiovascular activity (changes in heart rate)

2.1 | Studying the Uncanniness of Artificial Empathy

Serving as the ignition spark for my argumentation, the first developed study tied directly into earlier work by Gray and Wegner (2012), who had reported that computers expressing their own emotional states had felt highly aversive to the participants of a laboratory experiment. Expanding on the authors’ paradigm, I pondered that similar (or potentially even stronger) effects might occur if the digital system not only serves as the

sender, but also as the receiver within an emotional exchange, hence being capable of

interpreting affective states and empathizing with other entities.

By manipulating participants’ perception of two highly realistic VR characters in a

between-subject design—i.e., introducing them as neural network based AI, simple chatbots, or human avatars of varying autonomy—Study 1 indeed revealed that empathic behavior was rated as more unsettling if it originated from an autonomous computer system instead of a human operator. As hypothesized, participants considered the impression of a mindful OVERVIEW OF THE DISSERTATION STUDIES 19

machine as a somewhat ‘creepy’ and ‘goosebumps-inducing’ occurrence3. Moreover, when I

asked them for a more detailed description of their experience in my informal debriefing

sessions, many students reported their desire to instinctively move away from the alleged AI

characters due to a rather ambiguous sense of feeling threatened. In fact, I had repeatedly

observed this impulse during my work as the study’s conductor, often noticing that

participants physically shied away from the virtual entities (e.g., by means of head

movements), especially once the supposed agents walked past the virtual camera at the end of

the prepared scene.

In my interpretation, this was a particularly interesting finding, which subsequently

sparked the conceptualization of the follow-up study. Obviously, participants’ evaluation of

the autonomous technology had depended on feelings of situational control; even though the

AI’s embodiment was not physically present but only part of an immersive virtual environment, the mere impression of corporeal proximity to this obscure, ‘thinking’ creation

had sufficed to trigger some form of avoidance response. In turn, this also posed the question

how such situational factors compared to the discussed influence of human uniqueness

concerns in people’s perceptions of technology. Were immediate control and threat

perceptions as important as the contemplation of human identity? Or could their influence be

even stronger, maybe even up to a full mediation of the users’ experience? To provide a first

answer to these questions, I subsequently conceived Study 2 to include a direct comparison of

attitudinal and situational variables, which were then manipulated in a novel laboratory

experiment.

3 Both of these attributes are actual items of the eeriness index that was used in the study (Ho & MacDorman, 2010). OVERVIEW OF THE DISSERTATION STUDIES 20

2.2 | Assembling the Model of Autonomous Technology Threat

Participants’ personal explanations during my pilot study had pointed me towards the

concept of threat, and as I monitored new additions to the body of HCI literature, I noticed

that this variable had indeed been turning into a current ‘hot topic’ among technology

acceptance researchers. To contribute to this exciting field of research, this dissertation’s

second study also served to scrutinize the role of threat, albeit with a special focus on

emotion-sensitive technology.

Confronted with the rather different typologies of threat that had been used in

previous HCI studies (e.g., Kang, 2009; MacDorman & Entezari, 2015; Wang & Rochat,

2017; Złotowski, Yogeeswaran, & Bartneck, 2017), I considered it a valuable first effort to find a common dimension on which the various threat-related findings could be plotted. In cooperation with my co-authors, I eventually developed the novel taxonomy of threat proximity (Stein, Liebold, & Ohler, 2019) as a suitable framework for the classification of

even the most diverse threat forms. In the emerging system, proximal threats would

encompass all situational factors in a human-technology interaction, including the fear of physical harm or other cues for social anxiety; on the opposite end of the spectrum, we conceived of distal threats as more vague, overarching expectations about negative future outcomes for society, which might result from the technology’s application (e.g., loss of jobs, resources, and human uniqueness). However, inspired by recent neuroscientific and evolutionary biological findings (Chekroud, Everett, Bridge, & Hewstone, 2014; Connolly et al., 2016; Fessler, Holbrook, & Snyder, 2012), we also assumed that all possible forms of threat on the suggested continuum should ultimately feed into a singular neurological representation of threat experience, which would further predict aversive emotional reactions.

To subject the developed model to an empirical test, Study 2 subsequently proceeded with a novel VR experiment. Replacing the passive observation of the previous study with an OVERVIEW OF THE DISSERTATION STUDIES 21

actual human-technology interaction, participants were now invited to engage in a

conversation with a virtual agent, which we introduced as a sophisticated personality

assessment AI (while actually having the system controlled by trained study conductors). In accordance with our theoretical underpinnings, we then manipulated both situational threat— in terms of interpersonal distance to the allegedly emotion-sensitive agent—and the activation of human uniqueness concerns by means of a priming procedure. Although the necessary deception required much more preparation than the design of Study 1, our manipulation checks, as well as personal during the data collection, confirmed the plausibility of the developed scenario.

Finally, the analysis of the obtained data revealed a distinct pattern: Whereas the

hypothesized proximal path was reproduced rather neatly—threat experience indeed

mediated the effect of perceived situational control on users’ attractiveness ratings—the

influence of more abstract contemplations and attitudes had emerged to a much smaller

extent. On the one hand, people’s individual insistence on human distinctiveness could not be

disregarded completely as a contribution to their threat experience (as indicated by a

significant correlation); on the other, a direct path from this variable to the eventual outcome

measures (i.e., attractiveness or eeriness) remained missing from the statistical results.

Having finished the work on Study 2, I became aware of a potential shortcoming of

the research I had conducted so far. Up to this point of my dissertation, all empirical efforts

had revolved around settings with at least some form of AI embodiment, either focusing on

mind perceptions (Study 1) or the comparison of attitudinal and situational factors (Study 2).

In my understanding, it was only logical that I now needed to proceed to a disembodied form

of sophisticated technology in order to explore how people’s response would change if

situational threat perceptions were taken out of the equation—i.e., to find out whether the OVERVIEW OF THE DISSERTATION STUDIES 22 distal path alone would suffice to provoke any form of threat response. For this purpose, I developed Study 3 around the interaction with an abstract (‘bodiless’) computer interface.

2.3 | Advancing to Disembodied AI and the Role of Culture

In stark contrast to the immersive VR experiences of my previous experiments, the participants in Study 3 merely had to solve a pen-and-paper test in front of a machine-like automatic facial emotion recognition (AFER) setup, which consisted of a large desktop PC, a video camera, and several monitors. Of course, in favor of my thesis’s overarching topic, the experimental introduction still made sure to highlight the strong autonomy and complexity of the shown technology, once again embedding it in the deceptive narrative of intelligent deep learning technology—when, in fact, all results were fully standardized.

In an additional modification of my previous work, the participants for this study were recruited from two cultural backgrounds, as I wanted to address the large cross-cultural differences in people’s tendency to anthropomorphize that had been suggested by previous

HCI literature (e.g., Bennett, & Šabanović, 2015; Kaplan, 2004; Katagiri, Nass & Takeuchi,

2001). Focusing on individuals of either Chinese or German descent, I further used the study as an opportunity to add the discussion of emotional display rules to my theoretical framework, which is explained in greater detail in the introduction of the respective paper

(see Chapter 5.1).

Unfortunately, the addition of culture as a quasi-experimental factor led to conceptual difficulties during the study’s immediate preparation, as the required measures now had to be translated from German language into Mandarin Chinese. Meritless translation efforts by

Chinese assistants and colleagues suggested that the eeriness scale (Ho & MacDorman, 2010) that had formed an essential outcome variable of the first two studies could not be converted into a perfect Chinese equivalent—especially when used in the context of a disembodied stimulus. Consequently, I decided to focus on the more salient items offered by Ho and OVERVIEW OF THE DISSERTATION STUDIES 23

MacDorman’s technology-related attractiveness scale (2010). To compensate for this

adjustment, Study 3 further included physiological measurements of cardiovascular activity,

which were supposed to provide additional insight into potentially arousing or threatening

reactions to the presented technology.

Having collected data from 89 participants, the statistical analysis showed that both

Chinese and German individuals evaluated the affective technology in this experiment rather

positively, with only negligible differences between the cultural groups. However, when

entering participants’ attributions of human likeness into a regression model, the data also

revealed that individual levels of anthropomorphism—which were generally higher in the

Chinese sample—worked as a significant predictor of the attractiveness ascribed to the AFER

system. Surprisingly, this relationship turned out positively linear: The more human-like a

person considered the software, the higher was their rating in terms of technology

attractiveness. Lastly, concerning physiological activity, I observed that Chinese participants quickly returned to their initial heart rate after feedback from the emotion-sensitive computer, whereas German students remained in a state of sustained cardiovascular arousal for several minutes.

For the general theme of this dissertation, these results added a potential twist to the first two experiments. Although the obtained data had demonstrated that East Asian individuals might indeed perceive more humanness in artificially intelligent computers—a perception which partially influenced the valence of their response—the study also showed that, for disembodied technology, more perceived human likeness was ultimately a good thing. At the same time, I have to acknowledge that only a fraction of the variance of the participants’ ratings was actually explained by their level of anthropomorphism, suggesting that the final affinity towards the abstract, emotion-sensitive computer had depended on numerous other factors as well. Similarly, the obtained physiological data remains open to a OVERVIEW OF THE DISSERTATION STUDIES 24 variety of interpretations. On the one hand, German participants had been significantly more aroused by the provided scenario; on the other, the lacking emotional specificity of autonomic reactions (Mendes, 2016) makes a qualitative understanding of this result (e.g., as fear or joy) virtually impossible. In the end, the reading of the observed arousal effect may depend upon the reliability one suspects in the simultaneous self-report measurements. From a more skeptical perspective, the longer cardiovascular increase among German participants could indeed be a manifestation of subconscious anxiety—which would mean that their positive subjective evaluations had stemmed from social desirability or a failure to process the ambiguous fear that is typical for the UV. On the other hand, it is just as likely that these participants were sincerely attracted to the technology and found the idea of emotion- sensitive computers exciting in the most positive sense. In any case, it has to be noted that hardly any participant reported feelings such as fear or discomfort in our post-experiment discussions, so that I lean towards the interpretation that the stimulus in this study did not fall into an Uncanny Valley of Mind; in the end, the distal path might not have been strong enough in this specific scenario.

The following three chapters consist of the articles that have been written about the conducted studies, which are either published in peer-reviewed journals or currently submitted for publication. Afterwards, a “General Discussion” chapter will integrate the presented findings into a joint conclusion regarding my theoretical framework, closing with implications for future research projects and AI developments.

STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 25

3 | Study I

Venturing Into the Uncanny Valley of Mind— The Influence of Mind Attribution on the Acceptance of Human-Like Characters in a Virtual Reality Setting

Jan-Philipp Stein Peter Ohler

Status: published in the journal Cognition

Formal citation/reference:

Stein, J.-P., & Ohler, P. (2017). Venturing into the uncanny valley of mind—The influence of

mind attribution on the acceptance of human-like characters in a virtual reality setting.

Cognition, 160, 43–50. doi:10.1016/j.cognition.2016.12.010

Contribution of the authors:

JAN-PHILIPP STEIN – literature review, study conception, design and programming of the

VR environment, data collection, data analysis, writing first draft of the paper

PETER OHLER – study conception, paper review, and guidance during the revision process

STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 26

Abstract

For more than 40 years, the uncanny valley model has captivated researchers from various fields of expertise. Still, explanations as to why slightly imperfect human-like characters can

evoke feelings of eeriness remain the subject of controversy. Many experiments exploring the

phenomenon have emphasized specific visual factors in connection to evolutionary

psychological theories or an underlying categorization conflict. More recently, studies have

also shifted away focus from the appearance of human-like entities, instead exploring their mental capabilities as basis for observers' discomfort. In order to advance this perspective, we introduced 92 participants to a virtual reality (VR) chat program and presented them with two digital characters engaged in an emotional and empathic dialogue. Using the same pre-

recorded 3D scene, we manipulated the perceived control type of the depicted characters

(human-controlled avatars vs. computer-controlled agents), as well as their alleged level of

autonomy (scripted vs. self-directed actions). Statistical analyses revealed that participants

experienced significantly stronger eeriness if they perceived the empathic characters to be

autonomous artificial . As human likeness and attractiveness ratings did not

result in significant group differences, we present our results as evidence for an "uncanny

valley of mind" that relies on the attribution of emotions and social cognition to non-human

entities. A possible relationship to the philosophy of anthropocentrism and its "threat to

human distinctiveness" concept is discussed.

Keywords: uncanny valley, theory of mind, social cognition, anthropocentrism, artificial

intelligence, virtual reality STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 27

3.1 | Introduction

Computer systems have become inseparably entangled with people's daily lives, ever growing in complexity and sophistication. Apart from many beneficial effects, research has also explored unpleasant experiences that result from engaging advanced technologies. A prominent contribution to this field, the uncanny valley theory (1970) by Japanese engineer Masahiro Mori illustrates how complex human-like replicas (such as robots and digital animations) can evoke strong feelings of eeriness if they approach a high level of realism while still featuring subtle imperfections. (Fig. 3).

Figure 3. Uncanny valley model (redrawn from Mori, 1970).

Although its basic assumptions have remained mostly unchanged for more than four decades, the model has not lost any relevance due to the continued success and advancement of digital technology. Even more so, the exploration of uncanny valleys has ceased to be a merely academic venture, as modern robotics keep unfolding their economic potential and big-budget entertainment media stand and fall with the perception of their virtual characters

(Barnes, 2011; Tinwell & Sloan, 2014). STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 28

Traditionally, research on the uncanny valley effect has focused on an object's specific

appearance or motion patterns to explore which features might come across as abnormal and

unsettling (Bartneck, Kanda, Ishiguro, & Hagita, 2009; Hanson, 2006; Seyama & Nagayama,

2007). As numerous studies have succeeded in exposing such visual imperfections and

connected them to negative evaluations, the phenomenon has been framed by theories such as

pathogen avoidance (Ho, MacDorman, & Pramono, 2008), mortality salience (MacDorman &

Ishiguro, 2006) or the fear of psychopathic individuals (Tinwell, Abdel Nabi, & Charlton,

2013). Pursuant to these evolutionary psychological approaches, the aversion against human- like entities with slight defects might serve as part of a behavioral immune system

(Schaller & Park, 2011), shielding individuals against potential dangers to themselves or their

progeny.

Concurrently, another research direction has put aside evolutionary factors in favor of

an underlying cognitive dissonance effect as explanation for the uncanny valley (Ramey,

2005; Yamada, Kawabe, & Ihaya, 2013). This theory builds upon the paradigm that people

use a combination of perceptual cues and former experiences to categorize a subject (e.g., as

"human" or "robot") so that they can efficiently anticipate its behavior. Once they encounter

an entity that violates their expectations, however, observers are likely to experience

cognitive dissonance, which then manifests emotionally as uneasiness, disgust, or fear.

Notably, this line of thought corresponds to one of the first definitions of the "uncanny" term

by German psychologist Ernst Anton Jentsch, who coined it as an eerie sensation arising

from "doubts about the animation or non-animation of things" (Jentsch, 1906, p. 204). More

than a hundred years later, Jentsch's conceptualization has become firmly embedded in the

natural sciences, as studies applying eye-tracking and neuroimaging methods continue to

support the cognitive dissonance hypothesis (Cheetham, Pavlovic, Jordan, Suter, & Jancke,

2013; Saygin, Chaminade, Ishiguro, Driver, & Frith, 2012). At the same time, literature has STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 29

remarked upon stimulus novelty as an essential factor for mental categorization conflicts

(Grinbaum, 2015); given multiple interactions, people should be able to form new templates

for elements that have repeatedly defied expectations, resulting in the "infill" of previously

prevalent uncanny valleys. On the other hand, with analogue and digital human simulations

advancing constantly, categorization conflicts might just shift to higher levels of realism, as

people get increasingly sensitive in detecting visual flaws (Tinwell & Grimshaw, 2009).

Mind in a Machine

Apart from the many studies on visual influences, a large body of research has

demonstrated that the attribution of certain mental capacities (such as goal direction and

interactivity) is also an important factor in the perception of an entity's animacy and therefore

its categorization (Fukuda & Ueda, 2010; Tremoulet & Feldman, 2006). As most modern computers and robots can provide an animate impression by acting in seemingly goal-

directed ways, people have been shown to "apply social rules and expectations" to them

(Nass & Moon, 2000, p. 87), inferring ideas about a machine's "personality" or some form of

digital mind. However, research has also indicated that people tend to attribute only one of

two mind dimensions to non-human entities: Unlike experience (defined as the ability to

feel), they merely ascribe agency (the ability to plan and act) to their technology, reserving the former as a distinctively human trait (Gray, Gray, & Wegner, 2007; Knobe & Prinz,

2008). Even more so, a pioneering experiment by Kurt Gray and Daniel Wegner has illustrated that blending this differentiation—by presenting a "feeling" computer system, even without mention of a human-like appearance—could lead to significant unease among participants (Gray & Wegner, 2012). In another study of the same paper, the authors found that a human subject bereft of any emotions was also rated as eerie, hinting at the possible uncanniness of emotional experience from the other side of the man-machine continuum.

Following this groundwork, research about job replacements by robots has shown that people STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 30 feel increased discomfort if they consider losing an emotion-related job to a machine, rather than one that relies on cognitive tasks (Waytz & Norton, 2014). In contrast to this, studies on embodied conversational agents in training contexts have indicated that people might actually prefer a digital character that expresses emotions to a neutral counterpart (Creed, Beale, &

Cowan, 2014; Lim & Aylett, 2005). Recent findings from the field of social robotics even suggest that people may only rely on visual cues to assess a human-like entity, taking its presumed mental abilities into little consideration (Ferrari, Paladino, & Jetten, 2016).

Undoubtedly, the diversity of these results invites further investigation of the circumstances under which attributions of mind place a creation into an uncanny valley. It seems particularly necessary to explore different facets of artificial minds that eventually contribute to observers' discomfort. As research has indicated that people feel anxious about machines expressing their own emotional experience (Gray & Wegner, 2012), it stands to reason to focus next on machines that also understand emotional experience in others— considering that feelings are rarely confined to a single , but serve a social function between individuals (Frijda & Mesquita, 1994). Therefore, a scenario in which digital entities recognize emotional states and react to them in a socially aware manner should shed new light on the uncanny valley of mind—a phenomenon that might, after all, relate to a basic understanding of human uniqueness.

Threats to Human Distinctiveness

Throughout history, many cultures have regarded a consciousness enriched by emotional states as inherently human domain, closely related to philosophical concepts like a person's spirit or soul (Gray, 2010). Although theology, natural sciences and social studies vary in their understanding of artificiality and spiritual essence, the Cartesian interpretation of humans as "ghosts" in (bodily) machines has been a prominent philosophical consensus for many, especially Christian, civilizations (Fuller, 2014). Influenced by countless myths about STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 31 golems, homunculi and other revolting creations, "the Western man puts all his pride in [a] delta which is supposed to be specifically human" (Kaplan, 2004, p. 477)—a mental (and, to some, spiritual) component that clearly distinguishes humans from other beings. Considering the long-standing prevalence of this worldview, it can be argued that many people would sense a fundamental threat to their identity—their differentia specifica—if previously

"soulless" machines began to share their more complex mental abilities. In consequence of this threat to human distinctiveness hypothesis, the aversion against intelligent non-humans constitutes a sociocultural form of threat avoidance (MacDorman & Entezari, 2015), which serves to protect not only the individual, but also humanity in general. As culture studies reveal a more generous conceptualization of the "soul" in East Asian societies (Kaplan,

2004), this theory also accounts for the higher robot acceptance in countries like Japan; their inhabitants, influenced by everyday Buddhism and Shintoism, might simply be more accepting of "spirited" machines instead of feeling replaced or violated (Borody, 2013; Gee,

Browne, & Kawamura, 2005).

However, not only the possession of mental states, but also the ability to ascribe them to oneself and others—known as social cognition or having a theory of mind (Premack &

Woodruff, 1978)—has been discussed as essential difference between humans and other creations (Adolphs, 1999; Gallagher & Frith, 2003; Pagel, 2012; Vogeley & Bente, 2010).

Although studies continue to present evidence for a basic theory of mind in some animal species (Call & Tomasello, 2008; Tomonaga & Uwano, 2010), the declaration of humans as

"pride of creation" due to abilities like perspective-taking and empathic processing remains widespread. More recently, research on the role of a mirror neuron system in human neurophysiology has offered scientific footing to the anthropocentric idea of human uniqueness (Azar, 2005; Iacoboni, 2009), albeit not without controversy (Spaulding, 2013).

From a more philosophical standpoint, the idea of superior human minds might even emerge STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 32 as a species-related form of narcissism, which arguably presents itself in the careless destruction of other creations. But no matter if mental states and the ability to ascribe them are interpreted as a natural product of neurological processes, spiritual privilege or the essence of human exceptionalism, it seems likely that the human identity would suffer severe consequences if virtual entities demonstrated their own, sophisticated theory of mind. Even more than two decades ago, when artificial intelligence was far less refined than it is today, scientists worried about diffusing the long-standing dichotomy of man and machine, and advised caution in the development of new human-like features (Nass, Lombard,

Henriksen, & Steuer, 1995).

The Current Study

To explore the presented reasoning, this paper focuses on the perception of emotions and social cognition in a human replica as primary cause for an uncanny valley response.

Following the theoretical groundwork, we devised an experiment that manipulated the mind attribution to human-like characters, while keeping constant their visual appearance and verbal expressions. Several groups of participants observed the same friendly and empathic dialogue scene of two virtual characters, but received different instructions as we claimed the

3D models to be either human-controlled "avatars" or computer-controlled "agents".

Secondly, we manipulated the alleged autonomy of the characters, stating the dialogue to be either a work of the controller's "own imagination" or an intensely prepared script. In summary, this resulted in a 2  2 factorial design with the conditions "human, scripted",

"human, autonomous", "computer, scripted" and "computer, autonomous".

According to the interpretation of social cognition as distinct human privilege, we expected an alleged artificial intelligence ("computer, autonomous") that shows awareness of another character's emotions to strongly violate category expectations. Specifically, we theorized that only such an autonomous agent—but not a scripted one—would be attributed STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 33

the mental processes underlying its empathic behavior, and that this digital social cognition would appear uncannily human. At the same time, we suspected a human avatar that only acts as a "vessel" for scripted content to cause more eeriness than an autonomously acting one, as it approaches the category border between human and non-human from the other side of the uncanny valley.

H1a: People will perceive autonomous virtual agents that display emotions and social

cognition as more eerie than scripted virtual agents.

H1b: People will perceive autonomous human avatars that display emotions and social

cognition as less eerie than scripted human avatars.

Taking inspiration from the work by Gray and Wegner (2012), our assumptions

intended to advance their notion of uncanny minds by turning the emotional computer into a

social entity—a machine that perceives emotions, interprets them, and adapts its behavior

accordingly. As we expected a particularly strong aversion against such a category-defying

creation, we further hypothesized that autonomous virtual agents would receive the lowest

attractiveness rating due to the subconscious impulse to avoid further contact.

H2: People will perceive autonomous virtual agents that display emotions and social

cognition as more human-like than scripted virtual agents.

H3: People will perceive autonomous virtual agents that display emotions and social

cognition as least attractive among the four groups.

3.2 | Methods

Although the method of juxtaposing human avatars with virtual agents has been well-

known to gaming research (Weibel, Wissmath, Habegger, Steiner, & Groner, 2008; Lim &

Reeves, 2010) and persuasion studies (Fox et al., 2014; Guadagno, Blascovich, STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 34

Bailenson, & McCall, 2007; Patel & MacDorman, 2015), we chose to apply it to a social

media scenario, assuming that most participants would be familiar with this specific domain.

Also, with social network services turning into one of the most influential media branches

(Ngai, Tao, & Moon, 2015), they have emerged as a particularly relevant platform for future

uncanny valley occurrences. To increase the study's ecological validity even further, we chose to make use of head-mounted display (HMD) technology, which has remained in the focus of software and hardware developers in recent years (Benner & Wingfield, 2016;

Zuckerberg, 2014). Since stereoscopic HMDs can visualize a highly immersive 3D environment, we hoped that participants would be more susceptible to our deceptive instructions than they would have been in front of a 2D screen, especially in the artificial intelligence condition.

Participants

We recruited 97 students at a German university (30 male, 67 female; age: M = 23.6 years, SD = 3.32) as participants for our 30 minute experiment. Although the recruitment process targeted students from a variety of study programs, those willing to participate were predominantly enrolled in media communication and education studies. With the exclusion of two participants who reported moderate simulator sickness from the applied VR technology, as well as three participants who failed the final manipulation check, the final sample consisted of 92 students (29 male, 63 female; age: M = 23.6 years, SD = 3.41). Participants received €5 or partial course credits for taking part in the experiment. We assigned them to one of the four experimental groups by means of block randomization and provided them with extensive consent forms at the beginning of their appointments. Since our study design featured several deceptive elements, each participation ended with a thorough debriefing about the real nature of the used materials and the aim of the study.

STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 35

Stimuli

To create a customizable platform for the experiment, we programmed our own

fictitious chat software with the game engine Unity (Version 5.2.0, Unity Technologies) and

its built-in VR support. While not featuring any actual chat functionality, the program

contained several mock-up elements such as login and notification sounds, as well as eight

different virtual characters. Apart from displaying it on a 2D computer screen, our application

could be accessed through the HMD "Oculus Rift DK2", which completely blocks out

visual information and records the user's head movement in order to provide them

with 360-degree gaze orientation.

In its first stage, the chat program displayed a mostly empty urban plaza with only a

couple of non-interactive characters walking around in a distance. After a 30-second waiting

period, which allowed participants to adjust to the immersive VR environment, two high-

resolution character models—one female, one male, both middle-aged—appeared in front of the participant's point of view and approached each other. Constituting the main part of the experiment, the two characters would then engage in a casual dialogue, which we had completely scripted beforehand. Participants were told to observe the conversation silently and not to interfere for standardization reasons. Although the final version of our software allowed for two-directional movement, we did not provide mouse or keyboard to prevent motion sickness and to restrict differences in the visual stimuli to basic head movements.

The conversation scene, which lasted for 140 seconds, was composed to include expressions about the general mood ("I'm feeling droopy"), temperature sensitivity ("Today is really hot") and hunger ("Yeah, I'm quite hungry, too") of the two dialogue partners. In order to operationalize their ability for social cognition, we scripted both characters to acknowledge each other's statements and to empathize with them (e.g., "That must be annoying for you.").

As we intended to use the same scene in all four experimental conditions, we presented the STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 36

spoken dialogue with a slight electronic distortion that could appear as either artificial voice

(of a virtual agent) or inferior sound quality (of a human speaker controlling an avatar).

Additionally, both characters displayed a small set of gesture animations during their speech,

which we had implemented using a license-free sample of motion capturing data (Carnegie

Mellon University Graphics Lab, 2015). This body language included hand movements such

as waving or the wiping of sweat (Fig. 4), as well as simple head movements. To avoid

drawing too much attention to the details of our deception, we applied the animations in a

restrained manner and told participants that the software could only provide a small amount

of expressions, triggered by the human user's press of a button or the agent's programming,

respectively. At the end of the scene, the characters would decide on visiting a virtual café,

walk past the observer's point of view and fade out acoustically.

Figure 4. Screenshot from the presented VR scene.

STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 37

Procedure

According to our study design, the instructions given prior to the 3D scene provided the

only, yet crucial, manipulation in the experiment. As such, we facilitated the introduction to

the different conditions with several steps. After each participant had received the same

technical briefing about our VR software, which "could be used to meet friends and chat with

them in real-time", we told them to observe either two human confederates or a new set of virtual agents—both framed as a basic "beta test" of the platform's appeal. With the additional variation of the characters' alleged autonomy, this resulted in four different instruction narratives (Fig. 5).

alleged autonomy

scripted autonomous

"… a chat our "…a chat our confederates confederates will will improvise" human present according to alleged script, word by word" identity "…basic agents "… intelligent agents with programmed with an randomized emotional computer example script" parameters and content processed in real-time"

Figure 5. The study's 2  2 factorial design and excerpts from the corresponding narratives.

In order to prime participants for the deceptive scenarios, we then asked them to read a

single-page newspaper article about virtual reality and its applications while we pretended to

log into our software's web server. Only those in the "computer, autonomous" condition

received the article with three additional lines of text, telling them about recent breakthroughs

in the field of artificial intelligence, neural network technology and machine learning. Apart

from the statement that "AI systems could now process dialogue in real-time, using word

databases and emotional ", we also drew attention to their ability for social STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 38 cognition, emphasizing that modern systems could interpret emotional cues and demonstrate

"surprising levels of empathy".

In the conditions with supposedly human-controlled characters, we instead facilitated our deception by presenting a rigged intercom system. Doing so, we were able to stage the coordination with two confederates, who were supposed to control the avatars from another laboratory.

Measures

We asked participants to fill out a 15-minute questionnaire immediately after the 3D scene had ended. The first part of the questionnaire comprised the eeriness (eight items,

α = .85), human likeness (six items, α = .84) and attractiveness (five items, α = .81) scales by

Ho and MacDorman (2010), which feature 19 semantic differentials developed specifically for research on the uncanny valley. For sufficient differentiability, we presented the items

(e.g., "reassuring—eerie" and "artificial—lifelike") in a 7-point answer format. We asked participants not to rate each character separately, but to report their combined impression of both characters in order to average possible gender biases.

Subsequent to the emotional evaluation, participants completed the 16 items of the

Simulator Sickness Questionnaire (SSQ; Kennedy, Lane, Berbaum, & Lilienthal, 1993), a well-established instrument to register technology-induced symptoms of nausea, disorientation, and oculomotor dysfunction. Using the suggested cutoff value for "severe discomfort", we ensured that participants with strong physiological reactions to the HMD technology were excluded from the final set of data, as their ratings might have been confounded with unpleasant somatic side effects.

Further questions examined previous experience with social media and VR to monitor these variables for potential outliers, as well as a short socio-demographic . Lastly, we conducted a manipulation check that assessed the perception of the characters' behavioral STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 39

autonomy ("The chat partners act on their own accord"), emotional autonomy ("The chat

partners possess their own feelings"), and social competence ("The chat partners are socially

competent") on 5-point scales. As a final question, participants had to answer explicitly who

they thought had controlled the two dialogue partners; choosing the option that did not match

our instructions led to the exclusion from the study.

3.3 | Results

Manipulation Check

After the exclusion of two people who had experienced unpleasant side effects from

the VR display, as well as three participants who had not confirmed the alleged identity in

our final question, we calculated separate one-way ANOVAs for the three manipulation

check items.

Regarding the behavioral autonomy rating, our analysis resulted in significant

differences between conditions, F(3,88) = 5.90, p = .001. Post-hoc LSD tests revealed that participants in the "computer, autonomous" group indeed perceived more freedom of action in the characters (M = 2.64, SD = 1.00) than those in the "computer, scripted" condition

(M = 1.83, SD = 0.72), p = .004. Similarly, differences between the "human, autonomous"

(M = 2.83, SD = 1.03) and "human, scripted" conditions (M = 2.13, SD = 0.85) turned out significant, p = .010. This means that, for both virtual agents and human avatars, participants perceived the characters' actions as more independent if our instruction had claimed so. We further note that the artificial intelligence ("computer, autonomous") was ascribed nearly as much behavioral autonomy as self-directed humans, indicating the success of this difficult deception.

For emotional autonomy, another one-way ANOVA revealed significant differences between groups, F(3,88) = 6.31, p = .001. Post-hoc testing showed that the "computer, autonomous" characters were attributed more own feelings (M = 2.55, SD = 1.06) than the STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 40

"computer, scripted" characters (M = 1.78, SD = 0.80), p = .007. However, the difference between "human, autonomous" (M = 2.91, SD = 0.95) and "human, scripted" conditions

(M = 2.63, SD = 0.88) did not emerge as significant, p = .287. Thus, participants perceived

emotional autonomy in human controllers irrespective of their alleged spontaneity, whereas

only self-directed virtual agents were attributed their own feelings. Arguably, the finding that

participants tended to ascribe some form of emotionality to humans, even within the

constraints of a verbatim script, matches the idea of emotions as a very basic human quality.

Finally, we compared the perceived social competence between the four conditions.

An analysis of variance did not result in significant group differences, F(3,88) = 1.53,

p = .212. Moderate means in every condition—from "computer, scripted" (M = 2.78,

SD = 0.95) and "human, autonomous (M = 3.09, SD = 0.95) to "computer, autonomous"

(M = 3.18, SD = 1.00) and "human, scripted" (M = 3.33, SD = 0.70)—suggest that participants generally perceived the characters as empathic and friendly. Combined with the other two manipulation check items, this speaks to the successful manipulation of social cognition perceptions in our experiment. Although the characters appeared socially competent to all groups, only the supposedly autonomous entities were regarded as initiators of the displayed behavior. Therefore, we argue that participants only ascribed the reasons for the empathic conversation—i.e., social cognition—to intelligent virtual agents and human avatars, but not to scripted agents.

Uncanny Valley Indices

To investigate if different attributions of mind corresponded to differences in our participants' emotional response, we calculated two-factorial analyses of variance for each index of the Ho and MacDorman questionnaire (Tab. 3). Checking the requirements for the parametric procedure, the human likeness and attractiveness scores appeared normally distributed in Shapiro-Wilk tests, whereas the eeriness scores in one group ("human, STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 41

autonomous") deviated slightly from a normal distribution. However, as we found

homogeneity of variances for all variables, and every group featured more than 20

participants, we applied parametric tests nonetheless, as they prove robust against small

requirement violations when investigating groups of sufficient size (Stevens, 1999).

Table 3. Means and standard deviations on the three uncanny valley indices for each condition.

Attribution of mind Human, Human, Computer, Computer, scripted autonomous scripted autonomous (n = 24) (n = 23) (n = 23) (n = 22) M SD M SD M SD M SD

Eeriness 2.97 0.80 2.79 0.97 2.77 0.59 3.43 0.96

human likeness 3.10 0.90 2.96 1.26 3.11 1.04 3.46 1.08

attractiveness 4.86 0.79 4.77 0.88 4.57 0.78 4.98 0.90 Note. Indices range from 1 to 7.

Figure 6. Average eeriness ratings for the different mind attributions (error bars reflect +/–1 standard error of the means). STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 42

Eeriness scores. The two-way ANOVA conducted with participants' eeriness scores

showed no significant main effect for the factor "alleged identity", F(1,88) = 1.51, p = .223,

as well as no significant main effect for the factor "alleged autonomy", F(1,88) = 1.83,

p = .180. However, an interaction between both factors (Fig. 6) emerged as significant,

F(1,88) = 5.68, p = .019 and ηp² = 0.06, constituting a moderate effect (Cohen, 1988). While

spontaneously acting humans appeared as less eerie than those following a script, participants

experienced it as more uncanny to watch allegedly autonomous agents than scripted ones. As

such, we confirm our main hypotheses H1a and H1b.

Human likeness scores. Unlike the analysis of the eeriness ratings, a two-way

ANOVA regarding human likeness yielded no significant results. There was no significant

main effect for the factor "alleged identity", F(1,88) = 0.17, p = .258, no significant main

effect for the factor "alleged autonomy", F(1,88) = 1.64, p = .647, and no significant

interaction between both, F(1,88) = 0.48, p = .266. Although we did find slightly higher

scores in the "computer, autonomous" than in the "computer, scripted" condition—aligning with the assumption of H2—we cannot support this hypothesis due to a lack of statistical significance.

Attractiveness scores. Similar to the findings for the human likeness ratings, the two- factorial analysis of variance did not produce any significant results for the attractiveness scores, neither a main effect for "alleged identity", F(1,88) = 0.04, p = .847, nor a main effect for "alleged autonomy", F(1,88) = 0.81, p = .371, or a significant interaction, F(1,88) = 2.05, p = .156. Despite our data revealing the highest attractiveness mean in the "computer, autonomous" condition—forming a numerical pattern that contradicts H3—the non- significant results again ask for a cautious interpretation.

STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 43

3.4 | Discussion

The presented study aimed at investigating the effects of mind attribution on the well-

known uncanny valley phenomenon. In a virtual reality experiment, we equalized the factor

appearance but manipulated the perception of emotional and social abilities in human-like characters. Building upon previous research, we designed characters that not only expressed

their own emotional experience, but also empathized with the mental states of each other.

Following different introductions to our stimuli (as scripted virtual agents, autonomous

artificial intelligence, or human avatars), we were able to compare the effects of different

attributions of mind on observers' emotional response. Statistical analyses revealed a

significant interaction of the factors "identity" and "autonomy", as participants found

autonomous human avatars that showed signs of social cognition less eerie than scripted

ones, but autonomous virtual agents significantly more eerie than their scripted counterparts.

In light of the considerable effect size, we present this result as evidence for the uncanniness

that arises from emotional computer systems turning into social beings with their own theory

of mind.

Unlike our initial assumption, we found no differences for perceived human likeness

between the four experimental conditions. We assume that participants have inferred this

rating predominantly from visual features. In fact, one of the six items of the corresponding

scale by Ho and MacDorman explicitly references the visual nature of the stimuli

("mechanical movement—biological movement"), prompting similar evaluations in all

conditions. Another explanation might lie in specific features of our 3D scene, such as the

subtly distorted voices or short delays after each spoken message; these elements may have

seemed peculiar to users who expected a human avatar, preventing the emergence of

significant results. STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 44

Similar to the human likeness scores, we found no significant difference between

participants' attractiveness ratings. To our surprise, the scores on this index turned out high

throughout all conditions (M ranging from 4.57 to 4.98 on a 7-point scale). This could be

connected to the fact that participants often reported their amazement about the applied VR

technology during and after their appointments. As the concept of attractiveness and the

applied measurement clearly depend on factors such as style and aesthetics (items include

"crude—stylish" and "ugly—beautiful"), it seems likely that our visually impressive

presentation influenced the answers on this index for all groups.

These observations notwithstanding, we note that the combination of our findings

matches the suggested theory of an uncanny valley of mind. While all four groups ascribed

the same visual appeal to the depicted characters, they reported different levels of eeriness

depending on the attributed mind. With every other variable kept constant between

conditions, this result indicates an aversion that is not softened by the "stylishness" of a human-like character, but might persist because of its unexpected emotional and social skills.

As laid out in the introduction of this paper, many cultures regard emotional experience as intrinsically human privilege. Similarly, the cognitive ability to attribute mental states to oneself and others (theory of mind) constitutes a central argument for many people's

anthropocentric worldview, highlighting the superiority of human minds. Indeed, our results

support the assumption that people react with increased caution if a virtual creation starts to

resemble (or at least competently replicate) the prowess of a human brain. The revealed effect

suggests that people prefer human-like replicas to be limited to a certain set of characteristics

and might not appreciate them to behave in an empathic or social manner. It could be that

they worry about losing their supremacy as humans—as suggested by the threat to human

distinctiveness hypothesis—or even fear imminent harm from the sophisticated non-human

creation. As several participants of our study reported their unease about not being able to STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 45

anticipate the autonomous agents' next action, we note that their discomfort might indeed be

connected to a perceived loss of control, which has been suggested as uncanny valley

correlate in previous literature (Kang, 2009).

At the same time, one may consider an ethical component that contributes to the

aversion against emotionally aware computers. Since ascribing mind to an entity also means

imposing moral responsibilities on it (Waytz, Gray, Epley & Wegner, 2010; Gray & Schein,

2012), the arrival of emotional and empathic machinery can hardly result in anything but

technology-related skepticism. Digital systems that are able to reflect about mental states, or

at least calculate some form of emotional consequence, will be given new ethical standards to

abide by: Just like their human creators, "feeling" computers operate with moral gravity.

Especially in contexts that blur the physical border between human and human-like

simulation (e.g., robotics or immersive virtual reality), this sense of ethical responsibility might directly influence the comfort that people experience, as they wonder if the system's theory of mind matches their own—and what to expect if it does not. Ultimately, this relates to a basic principle that literature has suggested more than two decades ago: "Individuals

reserve the right to decide which roles computers should fill" (Nass, Lombard, Henriksen, &

Steuer, 1995, p. 237).

Limitations and Future Work

Our study has explored the uncanniness of human-like replicas that have the potential to conquer uniquely human domains in an uncontrollable way. However, we want to report several limitations of the conducted work. As most students responding to our recruitment process came from media and communication studies, it seems likely that the evaluation of the VR scenario was subject to a sampling bias. This is reflected by the moderate eeriness and high attractiveness ratings collected from our media savvy participants. Several students expressed their curiosity in the applied technology, which might have been less fascinating, STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 46

or even explicitly unpleasant, to people from other areas of expertise. A similar limitation

occurs due to the small age range of our sample, as younger people ("digital natives") can be

much more accepting of new technology than older generations (Buckingham, 2013). Hence,

we consider it possible that additional examinations among older participant samples could

reveal an even stronger aversion against the idea of emotionally aware computer systems.

Lastly, the deceptive and persuasive elements of our manipulation leave room for methodological critique. While the final manipulation check only led to the exclusion of three people, it remains unclear how the remaining participants related to our complex instruction statements. As some of the recruited students reported to be well-accustomed to experimental studies, they might have been wary of deceptions, or have answered according to social desirability. Although our explicit manipulation check resulted in different emotional autonomy scores between conditions, we collected no qualitative data that could indicate whether participants attributed the same type of genuine emotion and empathy to the artificial intelligence as they would have ascribed to humans. The participants in the respective condition might have been completely persuaded by our introduction of the "intense emotional realism of neural network systems", or they may have considered the empathic behavior merely as well-executed pretense. A more extensive inquiry into the perceived authenticity of digital emotions and social cognition might help to improve the informative value of further studies. Still, as we have provided an enriched scenario and received only affirmative feedback during the debriefing sessions, we consider our manipulation as controlled as possible.

Considering the documented impact of individual beliefs on uncanny valley sensitivity

(MacDorman & Entezari, 2015), an essential next step might be to collect data from different cultural backgrounds. As our study sample mainly consisted of atheistic students with a

Western socialization, the examination of a more diverse group of participants should yield STUDY I: “VENTURING INTO THE UNCANNY VALLEY OF MIND” 47 further insightful results. Also, with literature highlighting the influence of personality traits on the perceived uncanniness (MacDorman & Entezari, 2015; von der Pütten, Krämer, &

Gratch, 2010), new studies could include personality assessments to be used as covariates.

Additional objective measures (i.e., eye-tracking or neuroimaging methods) would help to produce results that go beyond people's conscious—and possibly biased—evaluations. By this means, research might yield interesting insights into the biological processes that underlie the aversion against emotionally aware non-humans.

Finally, we recommend that future studies consider the manipulation of both appearance and mind when evaluation non-human characters, instead of equalizing one of these factors. Since previous research has indicated that the perceived personality of a robot is connected to its facial design (Broadbent et al., 2013), additional experiments should investigate if the attribution of social cognition interacts with certain visual features to create even more distinct forms of digital eeriness. Against the background of unstoppable technological progress, any effort to explore an uncanny valley of mind will hold great value, as it supports the harmonious co-existence of humans and their machinery in the long run. STUDY II: “STAY BACK, CLEVER THING!” 48

4 | Study II

Stay Back, Clever Thing! Linking Situational Control and Human Uniqueness Concerns to the Aversion Against Autonomous Technology

Jan-Philipp Stein Benny Liebold Peter Ohler

Status: published in the journal Computers in Human Behavior

Formal citation/reference:

Stein, J.-P., Liebold, B., & Ohler, P. (2019). Stay back, clever thing! Linking situational

control and human uniqueness concerns to the aversion against autonomous technology.

Computers in Human Behavior, 95, 73–82. https://doi.org/10.1016/j.chb.2019.01.021

Contribution of the authors:

JAN-PHILIPP STEIN – literature review, development of the theoretical model, study

conception, design and programming of the VR environment, data analysis, writing the first

draft of the paper

BENNY LIEBOLD – development of the theoretical model, study conception, data analysis,

writing the first draft of the paper (methods section), paper review

PETER OHLER – study conception, paper review

STUDY II: “STAY BACK, CLEVER THING!” 49

Abstract

As artificial intelligence advances towards unprecedented levels of competence, people’s

acceptance of autonomous technology has become a hot topic among psychology and HCI

scholars. Previous studies suggest that threat perceptions—regarding observers’ immediate

physical safety (proximal) as well as their more abstract concepts of human uniqueness

(distal)—impede the positive reception of self-controlled digital systems. Developing a

Model of Autonomous Technology Threat, we propose both of these threat forms as common antecedents of users’ general threat experience, which ultimately predicts reduced technology acceptance. In a laboratory study, 125 participants were invited to interact with a virtual reality agent, assuming it to be the embodiment of a fully autonomous personality assessment system. In a path analysis, we found correlational support for the proposed model, as both situational control and human uniqueness attitudes predicted threat experience, which in turn connected to stronger aversion against the presented system. Other potential state and trait influences are discussed.

Keywords: autonomous technology, artificial intelligence, threat, control, human uniqueness, virtual reality

STUDY II: “STAY BACK, CLEVER THING!” 50

4.1 | Introduction

In a time when people have conversations with their phones’ operating systems and

healthcare robots smile reassuringly while taking blood pressure, it is clear that computers

have far surpassed their traditional role as passive machinery. Acknowledging this profound

redefinition, a growing body of research has investigated the factors that determine the

successful interaction between humans and autonomous technologies. Apart from abstract

supercomputers (Gray & Wegner, 2012), self-controlled vehicles (Nordhoff et al., 2018), or sophisticated domestic robots (Bartneck et al., 2007; Kwak, Kim, & Choi, 2017), the respective studies also focus on human-like androids and virtual agents (e.g., Krämer et al.,

2018; Złotowski et al., 2017), whose anthropomorphic design adds yet another factor defying

traditional views on human uniqueness. Mori’s uncanny valley model (1970) remains an

influential framework in this regard, as it comprehensibly illustrates the relationship between

a machine’s human likeness and the way people might react to it. According to Mori’s

observations, human-like replicas evoke an increasingly positive response the more lifelike

they appear, until a level of close—yet slightly imperfect—realism triggers intense

discomfort (i.e., fear or disgust) among observers. Although several theories have been

proposed to explain this phenomenon, their relative contributions remain the subject of on-

going scientific debate (e.g., Shimada, Minato, Itakura, & Ishiguro, 2018; Wang, Lilienfeld,

& Rochat, 2015). Moreover, due to groundbreaking innovations in the field of artificial

intelligence (AI), many scholars have started to shift their focus from the “looks” of

contemporary technology to its complex mental abilities, including emotional experience

(Gray & Wegner, 2012) and emulated empathy (Liu & Sundar, 2018; Stein & Ohler, 2017a).

Now, with the realm of uncanny minds added to that of uncanny appearances, it has become

more challenging than ever to disentangle the factors that cause people to feel wary in the

presence of sophisticated technology. STUDY II: “STAY BACK, CLEVER THING!” 51

Explanations for the Uncanniness of Technology

From the many explanations that have been suggested during nearly five decades of

uncanny valley research, two main approaches can be distilled. First, an evolutionary

psychological perspective has explored the role of specific perceptual cues (e.g., a robot’s

unrealistic eyes or movement patterns), which may prompt an aversion against pathogens (Ho

et al., 2008), unfit reproductive partners (Green et al., 2008), or psychopathic individuals

(Tinwell et al., 2013). Considering the innate nature of these avoidance mechanisms, the

uncanny valley could indeed constitute an evolutionary phenomenon—i.e., a descendent of

the inherently human fear of unfamiliarity. In a similar vein, some authors have argued that

people’s fear of faulty human-like machines might be related to the primal conflict that death

is inevitable, assuming that such creations remind observers of the vulnerability of physical

bodies (MacDorman & Ishiguro, 2006).

The second research direction, on the other hand, has focused more on the violation of

overarching mental categories as an explanation for uncanny valley observations

(MacDorman & Ishiguro, 2006; Yamada et al., 2013). Its proponents assume that an entity eluding previously acquired expectations (e.g., a computer acting emotionally) will cause

unpleasant cognitive dissonance, triggering the impulse to avoid further contact. Remarkably,

this argument is supported not only by mathematical models (Moore, 2012) but also by

neuroimaging research, as fMRI studies confirm cognitive dissonance effects on a basic

neurological level (Saygin et al., 2012; Urgen, Kutas, & Saygin, 2018). At the same time, it

has become scientific consensus to interpret expectation violations as the product of both

biological adaptation and socio-cultural influence—considering that religion, folklore, and

media all contribute to people’s understanding of man–machine distinctiveness and their

making sense of technology (e.g., MacDorman & Entezari, 2015; Sundar, Waddell, & Jung,

2016; Young & Carpenter, 2018). Especially in Western cultures, which have been shaped by STUDY II: “STAY BACK, CLEVER THING!” 52

centuries of predominant Christianity, there is still an implicit tendency to regard humans as

the unique “pride of creation,” a species of unrivaled mental prowess and privilege (Fuller,

2014). In turn, these anthropocentric attitudes have been suspected of spawning negative

views on other entities, including animals and plants, but also human-like technology

(Haslam et al., 2009; Kaplan, 2004).

Autonomous Technology and Threat Experience

From a broader view, both described interpretations of the uncanny valley— evolutionary mechanism and culture-dependent categorization conflict—seem to provide quite different explanations for the aversion to advanced technology. However, it can be noted that both perspectives suggest some sort of threat perception, may it stem from innate or socialized factors, as the underlying cause for negative user responses. Accordingly, recent research has shown that the two approaches do indeed complement each other in the genesis of people’s technology acceptance: Whereas evolutionary psychological factors predict immediate feelings of eeriness, culturally acquired attitudes might contribute indirectly by increasing people’s sensitivity for the effect (MacDorman & Entezari, 2015). In consequence, scholars have hypothesized that the aversion against advanced machinery depends on both

“sociocultural constructions and biological adaptations for threat avoidance” (MacDorman &

Entezari, 2015, p. 141).

A similar notion can also be found in the theoretical work by Kang (2009), who anticipated control and threat perceptions as the most crucial influence on the acceptance of future technologies. Only a decade later, the autonomy of contemporary machinery has advanced far enough to turn Kang’s assumptions into empirical reality; with the increasing freedom of action in modern technologies, their potential to evoke threat perceptions has grown substantially. More specifically, a recent paper by Złotowski, Yogeeswaran, and

Bartneck (2017) has linked attitudes towards autonomous robots not only to perceptions of STUDY II: “STAY BACK, CLEVER THING!” 53

realistic threats (e.g., the loss of jobs, resources, and safety), but also to more symbolic

identity threats (e.g., the loss of human uniqueness). Both of these threat categories are in fact

echoed by empirical findings from other studies, although a clear differentiation does not

always seem feasible. Concerning a more realistic form of threat, another experiment in the

field of social robotics has demonstrated that participants experienced stronger discomfort

when engaging robot groups of increased group size and coherence, as they started to expect

unfavorable treatment by the mechanical “out-group” (Fraune et al., 2017). Similarly, virtual

agents that threatened users’ autonomy while giving environmental advice instilled strong

psychological reactance among users in a previous study (Roubroeks et al., 2011). In some cases, however, observations that initially appear to involve realistic threats might also encompass more abstract, identity-related issues. For example, Waytz and Norton (2014) argued that botsourcing—the process of giving human jobs to robots—may evoke particularly strong aversion if emotion-oriented jobs are redistributed to machines, as the loss of factual resources (i.e., employment) would then be amplified by a perceived loss of identity. Similar arguments can also be found in the growing body of dehumanization research, which examines the significance of human uniqueness for people’s self-esteem

(e.g., Ferrari, Paladino, & Jetten, 2016; Haslam, 2006; Turkle, 1984; Vaes et al., 2012).

Offering a comprehensive introduction to this line of thought, Biocca’s article “’s

Dilemma” (1997) assumes that people will feel increasingly unnatural the more they are surrounded by human-like technology; in consequence, the author suggests that human identity can only be lost while it is conquered by other, non-human entities.

Taking all of the reviewed findings into account, we note that threat from autonomous technology actually serves as a two-fold term in academic literature, combining concerns about immediate physical harm with the more abstract fear of delayed negative outcomes for human society. At the same time, we find that the previously established terms of realistic STUDY II: “STAY BACK, CLEVER THING!” 54

threats and identity threats (Złotowski, Yogeeswaran, & Bartneck, 2017) may sometimes

blend with each other; as such, we suggest to reshape the dichotomy into a more flexible

continuum of threat proximity, which ranges from threats that are very close to the physical

body (“proximal threats”) to those that are more immaterial and intellectual in nature (“distal

threats”). Figure 7 illustrates how this novel terminology may be used to plot previous

theories on a common dimension.

Figure 7. Threat proximity as a common dimension of previous conceptualizations (e.g., MacDorman & Entezari, 2015; Złotowski et al., 2017).

The Model of Autonomous Technology Threat

Following our review of potential threat perceptions in the face of self-controlled technology, we propose an integrative Model of Autonomous Technology Threat that includes

(a) lost situational control as the prototypical proximal threat and (b) defied human

uniqueness as the most extreme form of distal threat. Since different types of threat

perception—both imminent and abstract in nature—have been shown to feed into a common

neurological representation of danger (Fessler, Holbrook, & Snyder, 2012), our model STUDY II: “STAY BACK, CLEVER THING!” 55 proceeds to the assumption that both specific factors contribute to a more general experience of threat within the individual (Figure 8). This underlying layer, in turn, is theorized as the core predictor of people’s (reduced) affinity for an autonomous technology. At the same time, we acknowledge that human uniqueness attitudes often revolve around strictly normative criteria such as religious taboos, which made us consider that the distal side of our model could also form a direct connection to the evaluation of autonomous technology, without the need to actually feel threatened.

Lastly, in terms of outcome variables, we focus on high eeriness and low attractiveness evaluations—the typical response pattern observed in empirical uncanny valley research. Since both variables have been interpreted as emotional qualities in previous studies

(Ho & MacDorman, 2010), our model not only acknowledges the strong empirical relationship between situational control and negative affect (Rapee, 1997), but also the previously suggested link between anthropocentrism and emotional responses to technology

(Nass et al., 1995).

Figure 8. Model of Autonomous Technology Threat.

STUDY II: “STAY BACK, CLEVER THING!” 56

The Current Study

For a validation of the proposed model, we developed a laboratory study that replaced

the hypothetical scenarios of previous studies with a naturalistic and interactive setting.

Employing virtual reality (VR) technology, we designed a human-like virtual agent and

deceptively introduced it as the embodiment of a fully autonomous personality assessment

system. After a short interaction with the allegedly self-controlled technology, participants

rated their situational control, concerns about human uniqueness, threat experience, and

aversion. According to our model, we assumed:

H1: Less perceived situational control in interactions with autonomous technology leads to stronger aversion (proximal threat).

H2: Stronger concerns about human uniqueness lead to stronger aversion against autonomous technology (distal threat).

H3: Threat experience mediates these effects.

To increase the potential variance in the obtained data, we employed a 2×2 between- subject experimental design, striving to manipulate our model’s two prototypical threat forms independently from each other. For the manipulation of situational control (as a factor contributing to proximal threat perceptions), we induced personal space violations, closely

following the definition of interpersonal distance as a buffer zone for behavioral control

(Fossataro et al., 2016; Horowitz et al., 1964; Strube & Werner, 1984). Since a pioneering

study in the field of VR has indicated that digital characters can trigger similar spatial

expectations as real-life encounters (Bailenson et al., 2003), we assumed that participants would experience less situational control (and thus higher proximal threat) in interactions with a physically close digital entity. To further strengthen our manipulation, we decided to present the allegedly autonomous technology in the form of a male agent, considering that STUDY II: “STAY BACK, CLEVER THING!” 57

personal space intrusions by male strangers have been shown to be particularly aversive to

both women and men (e.g., Rustemli, 1988).

H4: Interpersonal distance violations by an autonomous technology’s embodiment reduce the perceived situational control.

Addressing our model’s second path, we manipulated participants’ concerns about

human uniqueness (as a factor contributing to distal threat perceptions) by presenting a

specifically prepared newspaper article about the potential loss of human identity by the

hands of autonomous technology. In our expectation, reading such an article before

interacting with a virtual agent should (a) activate respective knowledge structures and (b)

affect participants’ attitudes towards autonomous technology, biasing them towards negative

views on the provided technology. As we assumed that reading the newspaper article would

also accentuate pre-existing human uniqueness concerns—which typically express

themselves as attitudinal and therefore quasi-experimental factors—our manipulation ultimately strived to increase the variance within the sample.

H5: Factual biasing by a newspaper article increases concerns about human uniqueness.

Lastly—on a more exploratory note—we were interested if several other, theoretically

relevant variables could offer a contribution to our model. Following our literature review,

we selected the personality traits need to belong, which has been shown to influence basic

animacy perceptions (Krämer et al., 2018; Powers et al., 2014), and need for control, which modulates the level of discomfort evoked by lost situational control (Leotti et al., 2010). As a contextual factor, we further looked into the predictability attributed to the allegedly autonomous system.

RQ: How do the need to belong, need for control, and perceived technology predictability affect participants’ evaluation of an autonomous technology? STUDY II: “STAY BACK, CLEVER THING!” 58

4.2 | Method

While developing the current study, we deemed it most crucial to provide participants with the credible impression of a truly autonomous technology. In order to achieve this goal, we decided to use the Wizard of Oz method (Martin & Hanington, 2012), in which a human experimenter controls the actions of a supposedly independent agent, thereby ensuring complex but also smooth interactions. Following a 2×2 factorial design (Figure 9), participants either read a news article discussing human uniqueness concerns or not (distal threat) before interacting with a virtual agent at close or medium interpersonal distance

(proximal threat).

Figure 9. The study’s 2×2 between-subject design.

Participants

Distributing invitation mails via Facebook groups and mailing lists, we recruited 126 undergraduate and graduate students at the local German university. Participants came from various study programs, with most of them being enrolled in media studies, engineering, and psychology. Every participant received €5 or partial course credit as a compensation for their time and effort. One participant had to be excluded from the data analysis due to not being STUDY II: “STAY BACK, CLEVER THING!” 59 able to correctly recall any part of the provided news article after the study. Thus, the final sample consisted of 125 participants (85 female, 39 male, 1 unspecified; M = 23.3 years,

SD = 3.71). Based on our 2×2 factorial design, participants were assigned to one of four experimental conditions by means of a block randomization.

Procedure

Participants were invited to engage in a brief interaction with a VR agent, which we deceptively introduced as the “embodiment of a novel AI-based personality assessment system.” Our cover story further claimed that the agent could assess anybody’s personality merely by talking to them, utilizing “a combination of voice recognition, movement tracking, word databases, and neural network technology.” To facilitate our deception, a Microsoft

Kinect body tracker and microphone were assembled in our VR laboratory; in reality, a well- trained study conductor controlled all of the agent’s actions remotely, and neither the prominently placed Kinect, nor the microphone actually captured any data.

Following our introduction to the study’s (alleged) scenario, participants received a full briefing on the anonymity and voluntariness of their participation. If they had been assigned to the bias condition, they were subsequently presented with a digital newspaper article on AI technology and its potential to conquer human uniqueness; participants in the non-biased group proceeded directly to the prepared virtual environment. Having put on the provided HTC Vive head-mounted display in a comfortable manner, participants could take some time to get accustomed to the spatial orientation in the VR environment. If no discomfort occurred, we remotely activated the digital agent, making it appear on the virtual stage and approach the participant from a distance. Depending on the experimental condition, the agent either stopped at an interpersonal distance of 4 meters (distant) or came as close as

80 centimeters (close)—which is slightly less than the average personal space margin, even when accounting for culture and gender differences (Sorokowska et al., 2017). Figure 10 STUDY II: “STAY BACK, CLEVER THING!” 60

depicts the participants’ point of view in both conditions. Although HTC Vive hardware

enables users to move around in the VR, we kindly asked participants to remain at their initial

observer’s point, which was monitored by the experimenter.

Figure 10. Experimental manipulation of interpersonal distance, 4 meters (left) vs. 0.8 meters (right).

Once the agent had reached its final position, its human “wizard” controlled it to ask participants for a short personal introduction, including their three favorite hobbies. Once the respective answer had been given, the scenario always resulted in the same standardized personality analysis. Depending on the participants’ talkativeness, the whole interaction lasted approximately four to five minutes, after which we asked them to complete a set of questionnaires on a laboratory PC. Concluding each appointment, participants were requested to leave their contact information so that we could inform them about the study’s results— and its true nature—at a later time. If participants preferred not to leave their information, we debriefed them directly and kindly asked them to protect the integrity of our cover story in front of other students.

STUDY II: “STAY BACK, CLEVER THING!” 61

Stimulus Materials

We used the game engine Unity (Unity Technologies, version 5.5.1, 2017) to build a

minimalistic virtual reality environment and the software Adobe Fuse (Adobe, 2017) to

design a middle-aged, male virtual agent. Additionally, the Salsa RandomEyes with LipSync

Unity plugin (Crazy Minnow Studios, 2017) was employed to synchronize the agent’s lip

movements with its spoken messages, and to provide realistic gaze tracking towards the

participant’s head position. As the core element of our Wizard of Oz deception, we coded a

multi-path interaction script that enabled the study conductor to react to a variety of situations

(while the participant was seemingly interacting with the autonomous system). Striving for at

least moderate consistency across trials, however, the final script contained no more than 39

different speech samples. In order to give the impression of actual voice and word

recognition, every interaction started with an acknowledgement of the participant’s field of study (“I have understood your voice perfectly. I have not yet talked to many students from the area of [e.g., economics]!”), which was prepared in multiple versions to capture every existing faculty of the local university. Other available actions included various signs of rapport (e.g., “U-huh”, “I understand”), clarifications of potential misunderstandings, and short motivational statements for cases of prolonged silence. Since we feared that a completely human voice would be disruptive for our cover story, all of the agent’s spoken messages were prepared as text-to-speech sound clips using Natural Reader software

(Naturalsoft Inc., 2017), which provides a warm, yet notably artificial voice. Lastly, we added subtle motion capturing animations (e.g., nodding, hand gestures) taken from the freely available Virtual Human Toolkit (Hartholt et al., 2013) to most of the prepared statements.

According to our cover story of an advanced personality assessment AI, the interaction script had to end in some form of “analysis result.” To avoid the confounding STUDY II: “STAY BACK, CLEVER THING!” 62

influence of different outcomes, we decided to compose a single results text, which only

included ambiguous (e.g., “I have not yet decided if you are ambitious or lazy”) and desirable

outcomes (e.g., “I would consider you to be a rather conscientious person”), as well as

statements matching the general situation of psychological experiments (e.g., “At the start of

this conversation you seemed a bit shy”). Doing so, we strived to make use of the “Barnum

effect” (Meehl, 1956), a phenomenon whereby vague and positive personality assessments

are typically perceived as accurate by most individuals (e.g., MacDonald & Standing, 2002).

Of course, to check if participants truly perceived the standardized result as authentic and

personal, control questions were added to the final questionnaire.

For the bias towards human uniqueness concerns in one half of our sample, we

created a detailed replica of a renowned national news website, in which we embedded a self-

written article detailing how autonomous technologies “get less distinguishable from humans by the day, both in terms of cognitive and emotional capacities.” To lend a believable voice to our biasing stimulus, we mostly used excerpts from real news publications, including

quotes from prominent human-computer interaction scientists and IT entrepreneurs. As a

closing argument, the article claimed that “according to most experts, a clean separation of

human and machine won’t be possible in the near future, resulting in countless ethical

challenges.”

Measures

For the purpose of validating our Model of Autonomous Technology Threat, we

needed robust measures for situational control and human uniqueness concerns, as well as

checks for their respective manipulations. Due to a lack of existing instruments that could be

applied to our specific scenario, we created our own questionnaires for these means; their

original German versions and ad-hoc translations can be obtained from the Supplementary

Materials. STUDY II: “STAY BACK, CLEVER THING!” 63

Situational control. To assess their situational control, participants were provided

with a six-item measure that consisted of both positively (e.g., “I felt as though I could react

to all eventualities.”) and negatively worded (e.g., “I was at the mercy of the situation”)

items, the latter of which were subsequently reverse-coded. Each item had to be filled in using a five-point response format (1 = fully disagree, 5 = fully agree). The resulting index of situational control showed acceptable internal consistency, Cronbach’s α = .74. An exploratory factor analysis further indicated that all items loaded high on a single factor, which explained 47% of the variance in our participants’ answers.

Human uniqueness concerns. Concerns about human uniqueness were assessed with 13 self-created items (e.g., “The idea that machines will someday have the same abilities as real humans makes me anxious.”) that had to be rated on a seven-point scale (1 = fully

disagree, 7 = fully agree). Internal consistency of the resulting scale turned out excellent with

a Cronbach’s alpha of .92. While an exploratory factor analysis suggested that the

questionnaire might encompass two sub-factors, all items showed a factor loading of at least

.45 on the first factor, which explained 51% of the observed variance; hence, we deemed the measure valid for the assessment of human uniqueness concerns as a singular construct.

Threat experience. Participants were asked to rate their general threat experience as

evoked by the autonomous technology with a set of ten items (e.g. “The agent was up to no

good”, “I was afraid of the agent”) on five-point scales. Positive items such as “The agent

gave a peaceful impression” were reverse-coded for our analysis. The final measure showed

good internal consistency, Cronbach’s α = .85. An exploratory factor analysis indicated that the measured construct included three sub-dimensions (i.e., kind impression with 3 items, fear/anxiety with 4 items, and suspected benevolence with 3 items). However, a subsequent confirmatory factor analysis showed that a single second-order factor could explain between

77% and 89% of the variance in these three sub-factors, with the model’s fit turning out STUDY II: “STAY BACK, CLEVER THING!” 64

excellent (χ² = 530.01; df = 45; p < .001; CFI = 0.982, TLI = 0.975, RMSEA = 0.046). Based on this, we suggest that our self-developed measure offered a sound assessment of threat experience as one overarching construct.

Technology aversion. Participants’ aversion against the presented autonomous technology was assessed using the well-established uncanny valley indices by Ho and

MacDorman (2010). According to the authors, the scales for perceived eeriness (8 items, e.g.,

“bland – uncanny”; Cronbach’s α = .77) and attractiveness (5 items, e.g., “ugly – beautiful”;

Cronbach’s α = .81) constitute two distinct affective measures within the complex conceptualization of uncanniness, so that we added both as outcome variables to our analysis.

The third index of Ho and MacDorman’s questionnaire, human likeness (6 items, e.g.,

“mechanical movement – biological movement”; Cronbach’s α = .84), focuses more on the uncanny valley’s x-axis and was therefore included to control for potential disruptions in our groups’ artificiality perceptions. However, the four conditions did not differ significantly in this regard, F(3,121) = 0.39, p = .76, meaning that we can rule out random effects in the human likeness ascribed to the presented technology.

Personal space violation and human uniqueness bias. Since we planned to manipulate the experienced situational control and human uniqueness concerns in our sample, additional measures were required to evaluate the success of both manipulations. For the induced personal space violation, we created seven items (e.g., “The virtual agent stood uncomfortably close to me.”) in a five-point answer format (1 = fully disagree, 5 = fully agree). The resulting index proved to be of excellent internal consistency, Cronbach’s

α = .95. By means of an exploratory factor analysis, we also found that all items addressed a single factor, accounting for 78% of the variance.

To make sure that our factual biasing in the form of a newspaper article had been read thoroughly, we composed a single-choice recall test with four questions about the provided STUDY II: “STAY BACK, CLEVER THING!” 65

text. By collecting answers from all participants—whether they had been biased or not—we

were able to compare the biased group’s recollection to chance level. Indeed, participants in

the newspaper condition achieved an average of M = 3.03 (SD = 1.00) correct answers, scoring significantly higher than those in the unbiased condition (M = 1.36, SD = 1.00),

Mann-Whitney U = 3.38, p < .001, r = .64. One participant in the bias group, who could not answer any of the four questions correctly, was excluded from our data analysis.

Additional state and trait variables. Reliable measures for participants’ need to belong and need for control—the trait variables addressed by our additional RQ—could be obtained from extant literature. The need to belong scale (Leary et al., 2013) consists of ten items (e.g., “I want other people to accept me”), with a good internal consistency of

Cronbach’s α = .81. The desirability of control scale (Burger & Cooper, 1979) covers 20 items regarding the individual’s need for control (e.g., “I try to avoid situations where someone else tells me what to do”); since two items of the questionnaire refer to car driving, we deemed them unsuitable for our student sample and only used the remaining 18 items, observing an acceptable Cronbach’s α of .71.

Technology predictability, the contextual aspect included in the exploratory RQ, was assessed with a self-created four-item scale (e.g., “I did not know what the agent would do next.”; 1 = fully disagree, 5 = fully agree). Although we examined a less-than-ideal internal consistency for the measure (Cronbach’s α = .61), the emergence of a single-factor solution in a subsequent exploratory factor analysis convinced us that it could still serve its exploratory purpose.

Control variables. To identify potentially problematic outliers in our sample, we asked participants about their previous experiences with VR, video games, and virtual agents

(all using one-item measures), as well as their level of public speaking anxiety using the well- established Personal Report of Communication Apprehension (PRCA-24; McCroskey, 1982) STUDY II: “STAY BACK, CLEVER THING!” 66

subscale for public speaking (6 items, e.g., “Certain parts of my body feel very tense and

rigid while giving a speech”; Cronbach’s α = .82). Neither the distribution of technological expertise nor that of public speaking anxiety indicated any notable outliers within the sample.

In the final, yet very crucial part of our questionnaire, we addressed the plausibility of the presented scenario with two questions. The first item explored the believability of the autonomous technology itself, asking participants to rate how competent they considered the

AI system on a scale from 1 (very incompetent) to 5 (very competent). Not only did we obtain a high mean of M = 3.92 (SD = 0.84) for this measure, we also found no significant differences between conditions, F(3,121) = 1.40, p = .25. The second question asked participants about the perceived appropriateness of the provided personality assessment, ranging from 1 (fully inaccurate) to 5 (the fully accurate). As an a priori exclusion criterion, we decided to regard a minimum value of 1 in at least one of the two questions as a sign of overwhelming disbelief; however, no participant met this cutoff. Instead, all participants considered the system’s personality judgment to be sufficiently accurate, M = 3.83 (SD =

0.68), with no noteworthy differences between groups, F(3,121) = 1.16, p = .33. Taken together, these results highlight the more than adequate believability of our Wizard of Oz scenario across conditions.

4.3 | Results

The threshold for statistical significance for all analyses was set to p < .05. Zero-order

correlations for all measured variables can be obtained from Table 4. Table 5 offers an

overview of the means and standard deviations of the study’s measures, broken down by

experimental group.

S Table 4. Zero-order correlations between measured variables. TUDY

II:

“ S TAY TAY variable 1 2 3 4 5 6 7 8 9 10 B ACK 1 eeriness – ,

2 attractiveness .19* – C LEVER 3 threat experience .12 -.22 * –

4 situational control -.06 .08 -.49 ** – T HING 5 human uniqueness concerns .08 .07 .20 * -.09 – ! ” 6 interpersonal distance violation .10 -.18 * .25 ** -.26** .10 – 7 newspaper test score (bias) .04 -.01 .09 .04 .05 .04 –

8 predictability -.43** -.03 -.16 .10 -.11 .01 -.13 –

9 human likeness .11 .46 ** -.11 .08 -.06 .14 -.05 -.02 – 10 need to belong .31** .11 .07 -.16 .26** .00 -.08 -.30** .08 – 11 need for control -.08 -.06 .03 -.11 -.02 -.02 .16 .11 -.05 -.21 *

Note: * p < .05, ** p < .01. 67

S Table 5. Means and standard deviations obtained for the measured variables. TUDY

II:

“

low interpersonal distance high interpersonal distance S TAY TAY no bias bias no bias bias B

M SD M SD M SD M SD ACK

a ,

eeriness 4.23 0.81 4.30 0.78 4.09 0.84 4.07 0.82 C LEVER attractivenessa 4.99 0.84 4.87 0.64 4.98 0.88 5.11 0.72

b T threat experience 2.05 0.69 2.02 0.61 1.81 0.56 2.03 0.13 HING

b ! situational control 2.67 0.69 2.89 0.74 3.13 0.68 2.98 0.79 ” human uniqueness concernsa 4.21 1.17 4.32 1.24 4.01 1.37 4.15 1.20 interpersonal distance violationb 3.63 1.02 3.89 0.94 1.50 0.48 1.68 0.54 newspaper test score (bias)c 33.6 27.4 76.6 21.3 34.4 22.7 75.0 28.6 predictabilityb 3.06 0.90 2.71 0.74 2.70 0.67 2.83 0.83 human likenessa 3.36 1.19 3.21 1.02 3.48 1.24 3.47 1.02 need to belongb 3.42 0.57 3.41 0.72 3.50 0.63 3.47 0.74 need for controlb 3.46 0.33 3.57 0.37 3.44 0.37 3.50 0.44

Notes. a Scale range from 1 to 7. b Scale range from 1 to 5. c Percentage of correct answers. 68

STUDY II: “STAY BACK, CLEVER THING!” 69

Path Analysis of the Model Variables

Based on our theoretical considerations, we conducted a path analysis using the

lavaan package in R to find out if the correlational structure of our data conformed with our

model. In this analysis, our manipulations were explored in their role as predictors of perceived situational control (H4) and human uniqueness concerns (H5), which then served as predictors for general threat experience (H1-3). In turn, threat experience was used to predict participants’ technology aversion as indicated by feelings of eeriness and attractiveness. All path coefficients can be obtained from Figure 11.

Figure 11. Coefficients obtained from path analysis (* p < .05, ** p < .01).

Although the experimental manipulation of human uniqueness concerns did not suffice to influence the underlying construct in the expected manner, the Model of

Autonomous Technology Threat itself was well reflected by the obtained data. Focusing on the situational perceptions and human uniqueness attitudes by our participants, we observed both to be significant predictors of threat experience (supporting H1 and H2), which further predicted decreased attractiveness evaluations. Only the proposed direct path from human uniqueness concerns to aversion, as well as the paths to participants’ eeriness perceptions STUDY II: “STAY BACK, CLEVER THING!” 70 turned out insignificant. Thus, we report that proximal and distal threat factors contributed to the aversion against autonomous technology as expected, albeit in a form that appears more targeted at attractiveness perceptions than at eerie feelings.

The Mediating Role of Threat Experience

A series of four mediation analyses was performed in order to investigate the potential role of general threat experience as a mediator between proximal and distal threat and the aversion against autonomous technology (H3). We used the PROCESS macro for SPSS with both eeriness and attractiveness as dependent variables and situational control and human uniqueness concerns as predictors. Indeed, the procedure revealed a significant indirect effect from situational control over threat experience on attractiveness, b = .12 [95% CI .01; .26], lending support for a mediation of our model’s proximal component. However, no significant mediation could be uncovered for the distal path, as the impact of human uniqueness concerns on technology aversion was not mediated by a general form of threat experience.

Causal Structure

Participants indeed reported a much stronger violation of personal space when the virtual agent stood right in front of them than when it kept its distance, t(123) = –15.48, p < .001, r = .81. Apart from this manipulation check, we also found that participants confronted with the close agent did in fact report significantly less situational control, t(123) = –2.195, p = .03, even though the effect turned out rather small, r = .19. Thus, while the manipulation of situational control can be considered successful in terms of statistical significance, we would like to give an only cautiously positive answer to H4.

On the other hand, we found no significant differences in participants’ human uniqueness concerns when comparing the group that had read the bias article compared to the group with no biasing, t(123) = –0.569, p = .571. This means that we were not able to bias participants’ attitudes towards an increased importance of human uniqueness by means of our newspaper article—indicating a negative answer to H5. STUDY II: “STAY BACK, CLEVER THING!” 71

Matching the reduced effectiveness of our manipulations, a multivariate analysis of

variances (MANOVA) using personal space violation scores and recall test results as

independent variables and eeriness and attractiveness as dependent variables remained

without significant results. Neither the manipulation of interpersonal distance violation, V =

.02, F(2, 120) = 1.44, p = .24, nor that of human uniqueness concerns, V < .01, F(2, 120) =

0.02, p = .98, showed a significant isolated influence on the intensity of participants’ aversion. The interaction between both factors turned out insignificant as well, V < .01, F(2,

120) = 0.49, p = .61. In light of these results, we have to refrain from interpreting our findings as causal evidence for the developed model; instead, we present our analyses strictly as correlational contribution to the current literature.

Additional State and Trait Influences

To explore additional systematic variance in the measured aversion against

autonomous technology, we conducted two step-wise hierarchical regression analyses

including the selected exploratory variables as predictors (RQ) and either eeriness or

attractiveness as the dependent variable. In both cases, all measured trait variables (human

uniqueness concerns, need to belong, need for control) were entered in a first step before the

measured state variables (situational control, perceived threat, agent predictability) were

added in a second step.

The analysis with eeriness as criterion resulted in a significant regression equation in

the first step, F(3,121) = 4.22, p < .01, explaining 7.2% of the variance in the dependent

variable. Introducing state variables increased the explained variance to 18.5%, F(6,118) =

5.69, p < .01. In this extended regression model, we observed that both the need to belong,

β = .20, t(118) = 2.25, p = .03, and agent predictability, β = –.37, t(118) = –4.26, p < .01,

significantly predicted the eeriness ascribed to the autonomous technology: The more STUDY II: “STAY BACK, CLEVER THING!” 72

participants felt the need to be socially included and the less they perceived the system’s

actions as predictable, the eerier they rated the presented stimuli.

A second hierarchical regression with attractiveness as criterion (and the same

predictor selection method) did not result in a significant regression model, neither in the first

step, F(3,121) = 0.64, p = .59, nor in the second, F(6,118) = 1.51, p = .18. As such, we

conclude that the additional state and trait variables had a more important influence on

evaluations of the agent’s eeriness, while attractiveness perceptions were more related to the

factors included in our main model.

4.4 | Discussion

The traditional understanding of the man–machine relationship has long been that of

human users and obedient tools, especially in Western cultures. However, recent advances in

the fields of social robotics and AI have started to contest these classic role assignments, as

digital entities continue to reach unprecedented levels of autonomy. In today’s technological

landscape, digital minds not only make their own meaningful decisions (Banks, 2018), they

also take on elaborate bodies (e.g., photorealistic VR agents) that allow for lifelike interactions. After millennia of much more restricted technology use, it comes as little surprise that people often engage these new forms of self-control and physicality with some reluctance. While studies also indicate that repeated interactions with autonomous technology might be enough to reduce users’ aversion (e.g. Złotowski et al., 2015), the harmony between people and their technological creations still appears inherently fragile, raising numerous practical as well as ethical questions.

We investigated how 125 participants experienced the interaction with a realistic virtual agent, which was framed as the embodiment of an autonomous assessment system.

Following the theoretical groundwork, we suspected threat experience to be the crucial mediator for potentially aversive reactions, linking both immediate (personal space invasions) STUDY II: “STAY BACK, CLEVER THING!” 73

and delayed threats (human uniqueness concerns) to a negative emotional response. A path

analysis focusing on our hypothesized model structure lent clear empirical support to our

framework. It uncovered significant paths from situational control and human uniqueness

concerns to threat experience, which in turn predicted reduced attractiveness evaluations.

Consequently, we note that threat perceptions—both proximal and distal in nature—indeed played an important role for the technology aversion of our participants. Moreover, a comparison of both factors’ effect sizes suggests that, at least in the type of scenario that we presented, situational variables might exert a notably larger influence on threat experience than general concerns about human identity. This finding was supported by a significant mediation effect for the proximal component of our model—situational control reduced threat experience and thereby influenced the attractiveness ascribed to the virtual system. However, no significant mediation effects could be observed for distal threat factors. Thus, while human uniqueness concerns did predict threat experience (to a lesser degree), the pattern found in our data was mainly determined by situational factors. In our interpretation, this conclusion actually makes sense, as attitudinal factors shape our perceptions of a situation

(e.g., to protect us from harm), but situational stimuli ultimately play a greater role in triggering actual responses. For real-life applications and the decision makers behind them, this could mean good news, considering that the situational conditions of interactive technology are mostly subject to conscious design choices. Based on our work, we argue that future technological endeavors might prove most successful in terms of user acceptance if they strive for unimpeded situational control—or, at the very least, the impression of it.

Although the current study merely examined people’s feelings towards a specific form of AI assessment system, our literature review and the obtained data clearly indicate that people’s views on technology are modulated by stable and overarching mechanisms; thus, the shown relevance of situational factors might apply to a much larger spectrum of autonomous systems as well. STUDY II: “STAY BACK, CLEVER THING!” 74

Methodologically, we adhered to the common conceptualization of uncanniness—a central outcome variable in technology acceptance research—as a blend of attractiveness and eeriness ratings. However, our path analysis showed that threat perceptions were mainly related to attractiveness, whereas eeriness relied more on other factors, such as the need to belong as a personality factor and agent predictability as a situational factor. As such, it appears that the experience of threat actually renders autonomous technology less acceptable but does not necessarily increase the strange and weird feeling reported by traditional uncanny valley experiments. In our interpretation, this finding indicates that intelligent machines falling into the “uncanny valley of mind” have reached a point beyond ambiguity and awkwardness—with their threatening gestalt merely triggering the impulse to avoid further interaction, making them appear less attractive, stylish, or beautiful.

Our participants’ perceived eeriness, on the other hand, was found to depend much more on factors such as the personality trait need to belong: The more people valued social inclusion, the more they disliked the presented AI system. We suspect a possible cause for this in the lack of interpersonal warmth during our prepared interaction scenario. Since we had scripted the agent to present a highly ambiguous personality judgement (including the rude statement “I have heard more interesting hobbies than yours”), its actions might have felt like a form of rejection to some participants. We assume that those with a higher need to belong may have been especially surprised or hurt by the analytic behavior of the assessment system, culminating in the impression of an eerie, not entirely normal human-computer interaction.

Finally, a particularly strong predictor (β = –.37) for eeriness emerged in the form of technology predictability, which we had initially conceptualized as a sub-factor of situational control. In light of this, we come under the impression that behavioral predictability may provide a much more fitting construct to explain uncanniness than situational control in general; unlike the latter, predictability focuses exclusively on the stimulus itself, setting STUDY II: “STAY BACK, CLEVER THING!” 75

aside internal attributions (such as self-efficacy) and other interfering situational variables.

Considering that predictability has also been shown to be an important antecedent of threat perceptions on a neurobiological level (e.g., Klahn et al., 2016), the concept might be another important construct to be included in our Model of Autonomous Technology Threat.

Therefore, we suggest a potential model extension for future applications of our work, although a more refined measure for technology predictability might be needed in this case.

Limitations

Familiarity, likability, affinity—uncanny valley researchers often disagree about a

meaningful conceptualization of the y-axis in Mori’s model, which he originally described

with the Japanese neologism “shinwakan.” Complicating matters even further, several

authors have argued that the concept is actually multi-dimensional (MacDorman & Ishiguro,

2006; Moore, 2012), raising doubts on the idea to simply measure technology aversion with a single variable. In consequence, it has become common practice to assess people’s disliking of human-like technology with multiple concepts, such as the presently used, two-fold operationalization as high eeriness and low attractiveness. According to Ho and MacDorman

(2010), who developed a widely used instrument to measure these two variables, this conceptualization is actually supported by high reliability coefficients and an only marginal intercorrelation. Indeed, our data showed that both dependent variables were shaped by quite different predictors, as some factors exerted a stronger influence on eeriness (need to belong, predictability) and others were more related to attractiveness evaluations (threat experience).

At first sight, these results indicate that the uncanny valley’s current operationalization provides the complexity needed to describe affective responses to sophisticated technology.

However, we think that the underlying mechanisms still remain oversimplified by this type of measurement, as uncanniness encompasses a multitude of affective (e.g., fear, disgust), physiological, and behavioral facets. At least in regard to autonomous technology, it could STUDY II: “STAY BACK, CLEVER THING!” 76

therefore be considered to examine all of these indicators separately than to summarize them

under an umbrella term.

During our effort to increase the natural variance of both types of threat perception in our sample, we encountered some difficulties, especially when trying to influence human uniqueness concerns by means of a newspaper-based bias stimulus. Unfortunately, our observation conforms with a recent critique about the many contingencies of related procedures (Cesario, 2014), indicating that attitudes cannot be sufficiently manipulated by short-term interventions. While the manipulation of situational control turned out more

successful, the influence of our personal space violations was also not as strong as expected.

Therefore, the evidence we provide for our model is correlational in nature. Additional

experimental work focusing on the importance of threat-related factors for the aversion

against autonomous technology could further strengthen this line of research.

While the natural interaction paradigm used in our experiment underscores the high

ecological validity of our findings, we still have to note several limitations due to the use of a

convenience sample of local university students. Specifically, both the similar level of

education among our participants, as well as their homogenous age arguably lessens the

generalizability of our results. In regard to our theoretical framework, we further note that the

relatively high percentage of atheists within our sample (69.6% claimed to be unreligious)

has likely weakened the impact of human uniqueness concerns—an attitudinal construct that

typically revolves around religious views and might even require a certain level of

fundamentalism (MacDorman & Entezari, 2015). In light of this, we assume that our data

might underestimate the distal path of the developed model, and that worries about human

distinctiveness should emerge as a much stronger influence among other samples, especially

those rooted in anthropocentric beliefs (e.g., conservative Christians, Muslims).

STUDY II: “STAY BACK, CLEVER THING!” 77

Conclusion

The current study set out to explore the role of perceived threat in interactions with autonomous technology. We juxtaposed two pathways that might fuel a general level of threat experience: Proximal threats that build upon the behavior of the respective technology and the surrounding environment, and distal threats that result from abstract ruminations about human identity. In a laboratory experiment, we found first support for the proposed model—but also inspiration for necessary modifications. Compared to studies that use hypothetical scenarios, our results achieved higher external validity as they emerged from an actual interaction with what was believed to be a fully autonomous technology. At the same time, the reported evidence is mostly correlational in nature; also, to our surprise, only attractiveness ratings were influenced by participants’ experience of experience, whereas the eerie impression of “getting goosebumps” depended on more specific factors such as the technology’s predictability. In summary, more pronounced manipulations of proximal and distal threats, as well as new ways of measuring technology aversion might be necessary to advance this fascinating research area. Eventually, design implications derived from this line of research could help to ensure that the potential benefits of autonomous technology are not lost because its users feel threatened, replaced, or hopelessly out of control.

STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 78

5 | Study III

Saving Face in Front of the Computer? Culture and Attributions of Human Likeness Influence Users’ Experience of Automatic Facial Emotion Recognition

Jan-Philipp Stein Peter Ohler

Status: published in the journal Frontiers in Digital Humanities

Formal citation/reference:

Stein, J.-P., & Ohler, P. (2018). Saving face in front of the computer? Culture and attributions

of human likeness influence users’ experience of automatic facial emotion recognition.

Frontiers in Digital Humanities, 7, 18. https://doi.org/10.3389/fdigh.2018.00018

Contribution of the authors:

JAN-PHILIPP STEIN – literature review, study conception, design of stimulus materials,

data analysis, writing the first draft of the paper

PETER OHLER – paper review

STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 79

Abstract

In human-to-human contexts, display rules provide an empirically sound construct to explain intercultural differences in emotional expressivity. A very prominent finding in this regard is that cultures rooted in collectivism—such as China, South Korea, or Japan—uphold norms of

emotional suppression, contrasting with ideals of unfiltered self-expression found in several

Western societies. However, other studies have shown that collectivistic cultures do not

actually disregard the whole spectrum of emotional expression, but simply prefer displays of

socially engaging emotions (e.g., trust, shame) over the more disengaging expressions

favored by the West (e.g., pride, anger). Inspired by the constant advancement of affective

technology, this study investigates if such cultural factors also influence how people

experience being read by emotion-sensitive computers. In a laboratory experiment, we

introduce 47 Chinese and 42 German participants to emotion recognition software, claiming

that it would analyze their facial micro-expressions during a brief cognitive task. As we actually present standardized results (reporting either socially engaging or disengaging emotions), we manipulate participants' impression of having matched or violated culturally established display rules in a between-subject design. First, we observe a main effect of culture on the cardiovascular response to the digital recognition procedure: Whereas Chinese participants quickly return to their initial heart rate, German participants remain longer in an

agitated state. A potential explanation for this—East Asians might be less stressed by

sophisticated technology than people with a Western socialization—concurs with recent

literature, highlighting different human uniqueness concepts across cultural borders. Indeed,

while we find no cultural difference in subjective evaluations of the emotion-sensitive

computer, a mediation analysis reveals a significant indirect effect from culture over

perceived human likeness of the technology to its attractiveness. At the same time, violations

of cultural display rules remain mostly irrelevant for participants' reaction; thus, we argue STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 80 that inter-human norms for appropriate facial expressions might be loosened if faces are read by computers, at least in settings that are not associated with any social consequence.

Keywords: display rules, collectivism, emotional suppression, , AFER, facial emotion recognition, cardiovascular activity, human uniqueness STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 81

5.1 | Introduction

Throughout past millennia, the exchange of emotional expressions has mostly been reserved for the interaction between biological entities. As a result, emotional behavior is typically considered as a core concept of human—or, at the very least, animalistic— communication (Wegner & Gray, 2016). Recent breakthroughs in the field of affective computing, however, have started to contest this domain: Contemporary artificial intelligence not only incorporates emotional recognition algorithms, but may also possess the ability to emulate its own “affective state,” reacting to a user's input in a supposedly emotional way.

Due to these innovations, whole new research fields and industries have emerged all over the globe, including social robotics (e.g., Michaud et al., 2000), agent-based psychotherapy (e.g.,

Oker et al., 2015), and emotionally aware smartphone applications (e.g., Chen et al., 2015). A crucial feature for many of these technologies is automatic facial emotion recognition

(AFER)—a camera-based form of facial analysis that gets more distinguished by the day

(Doerrfeld, 2015; Gunes & Hung, 2016). AFER measures a wide range of movements in the user's facial muscles, including micro-expressions that are nearly undetectable to the human eye, before applying classification systems such as the Facial Action Coding System (Ekman

& Friesen, 1978) to provide accurate interpretations of users' mood or short-term affective response. Depending on the variant of recognition software used, the final result may even offer a simultaneous quantification of several emotions—although the reliability of such systems remains the subject of critical debate.

Affective Technology Versus Human Uniqueness

Apart from the relentless advancement of emotion-sensitive technology itself, the public adoption of affective forms of human-computer interaction (HCI) has not proceeded without obstacles. In fact, recent studies have uncovered strong feelings of discomfort among participants who were presented with emotional computers (Gray & Wegner, 2012) and STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 82 empathic digital agents (Stein & Ohler, 2017a). Cross-national research implies that these effects might depend on cultural factors, as different religions and philosophical mindsets promote the importance of human uniqueness to a varying degree (Kaplan, 2004; Vess et al.,

2012). Whereas most Western civilizations remain embedded in Christian principles of anthropocentrism—a philosophy that puts humans above all other creation—East Asian societies tend to have a less restricted idea of human distinctiveness, which allows the attribution of emotional experience to many different entities (Kazuhiko, 2017; Kitano,

2007). In consequence, Chinese or Japanese users may find it hardly problematic if a machine acquires typically “human” features such as the ability to recognize or express its own feelings; people in the West, on the other hand, are often socialized with a “Frankenstein syndrome” (Kaplan, 2004), therefore considering such technology as a threat to human nature itself (Złotowski et al., 2017). In practice, these arguments—although not entirely unchallenged (Haring et al., 2014)—also offer an explanation for the notably higher acceptance of social robots in countries such as Japan or China (Li et al., 2010; MacDorman et al., 2009; Nomura et al., 2015).

The inclusion of philosophical concepts in empirical HCI studies certainly shows that users' acceptance of emotional technology reaches far beyond questions of programming or basic interface design. Still, due to the novelty and constant advancement of affective systems, many theoretical implications in terms of user experience have not yet been addressed in a sufficient way. A psychological construct that remains particularly under- researched in this regard are emotional display rules, which can be defined as behavioral criteria for the expression of emotions that stem from cultural socialization (Ekman &

Friesen, 1969). While numerous studies have highlighted the importance of these sociocultural norms for human-to-human contexts, conclusive findings on the transferability of such rules to human-computer interactions are virtually absent from the academic STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 83 discourse. The current study strives to contribute to this research gap, exploring whether the customary way of expressing emotions among humans may also affect the perception of emotion-sensitive software.

Emotional Display Rules

So far, psychological literature has not yet provided a decisive answer concerning the universality—or incongruity—of emotional experience across different cultures (Derntl et al.,

2012; Elfenbein & Ambady, 2002; Hwang & Matsumoto, 2015; Matsumoto & Ekman,

1989). Nevertheless, scholars have unanimously acknowledged the importance of emotional display rules as a mediator of people's observable emotional behavior. This means that, while cultural socialization might not necessarily impact the conception of emotional states within the individual, it clearly determines which part of the subjective experience is presented to the environment (Matsumoto et al., 2008a). As a result, emotional display rules yield the power to make people feel accepted (or disregarded) by their cultural in-group, contributing profoundly to an individual's psychological well-being (Ford & Mauss, 2015).

Display rules have been shown to correlate with age (Camras, 1985; Underwood et al., 1992) and gender (Brody, 1997; Kring & Gordon, 1998), as well as several personality characteristics (Matsumoto, 2006). Most of all, however, they are influenced by factors of culture, building upon vastly different conventions, taboos, and expectations across ethnic groups, countries, and whole continents (Matsumoto et al., 2008a; Safdar et al., 2009). To structure these effects on a global level, researchers have frequently utilized the differentiation between individualistic and collectivistic cultures, which remains one of the most prominent frameworks in the field of cross-cultural psychology. In its original interpretation as a bipolar dimension (Hofstede, 1980), collectivism describes a culture's tendency to value interdependence, self-restraint, and in-group cohesion, in contrast to the individualistic emphasis on personal goals and self-expression. Based on these criteria, many STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 84

East Asian cultures have typically been attributed with a strong collectivistic orientation, whereas the United States and several West European countries have been labeled as highly individualistic societies (Hofstede, 2001). At the same time, an increasing number of authors have been contesting Hofstede's dichotomy as a general model of the East and the West (e.g.,

Oyserman et al., 2002; Parker et al., 2009; Takahashi et al., 2002), instead suggesting to split the singular construct into two independent traits (Triandis & Gelfand, 1998). In this evolved form, the IND-COL taxonomy is still used extensively for cultural comparisons, not least in the exploration of societal norms such as emotional display rules.

So far, a large number of studies have provided evidence that emotional suppression constitutes the overarching display rule in many countries with a collectivistic orientation

(Matsumoto et al., 2008b), including China (Davis et al., 2012), South Korea (Kim &

Sherman, 2007), Singapore (Moran et al., 2013), and Japan (Safdar et al., 2009). Unlike more individualism-centered societies in the West, these East Asian cultures have been shown to disregard unfiltered emotional displays, instead promoting the concealment of individual feelings for the sake of the collective social order (Markus & Kitayama, 1991; Matsumoto et al., 2008a). Even more so, cultural scientists have suggested that the according principles trace back as far as ancient Taoist and Confucian tradition (Ho, 1986; King & Bond, 1985) and are thus deeply embedded in the “cultural DNA” of the respective societies. As a result, the corresponding norms are usually internalized at an early age—which also explains why, unlike the adverse effects reported for Western samples, emotional suppression actually increases the subjective well-being (Matsumoto et al., 2004), academic performance (Chen et al., 2009), and psychological functioning (Soto et al., 2011) in members of Chinese or

Japanese culture. Fascinatingly, this positive view on emotional concealment is also reflected by a variety of linguistic nuances, as East Asian languages provide an unmatched number of idioms to describe the act of hiding one's emotions, including the notion of “saving face” STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 85

(Ho, 1976). Indeed, for the daily life of many East Asians, the metaphorical sense of “face”

(as a form of social standing) remains inseparably intertwined with the literal face, as well- functioning regulatory mechanisms are deemed essential to avert public humiliation (Dong et al., 2013; Ho et al., 2004).

Display Rules and Types of Emotion

Apart from the well-replicated main effect of culture on emotional suppression norms, many cross-cultural researchers have directed their attention toward potential interaction effects between display rules, different audiences, and specific types of emotion. For instance, recent findings have shown that members of collectivistic cultures are particularly focused on differentiating between private and workplace contexts as they consider appropriate facial displays (Moran et al., 2013; Wang et al., 2012). According to a large-scale comparison of display rules across 32 countries by Matsumoto et al. (2008a), this might actually extend to general in- and out-group effects: Although participants from collectivistic societies reported less expressivity in general, they actually endorsed negative emotional displays toward strangers much more than participants with an individualistic background.

This observation in turn connects to a growing body of literature about the “appropriateness” of selected emotional states (e.g., Eid & Diener, 2001; Kitayama et al., 2000; Seo, 2011), which has suggested that individualistic cultures strongly favor displays of personal success

(e.g., pride, joy), whereas collectivists prefer emotions that highlight interrelatedness—even if their expression emphasizes personal failure (e.g., guilt, shame). Considering the core principles of both cultural orientations, this actually makes perfect sense: Just as the visible acknowledgment of personal shortcomings highlights the investment in the collective well- being, turning guilt and embarrassment into other-focused emotions (Markus & Kitayama,

1991), displays of pride or happiness mostly serve to express private gain, therefore meshing with a more individualistic philosophy (i.e., ego-focused emotions). Lending further support STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 86 to this conceptualization, recent research has revealed that East Asian participants actually appreciated guilt and shame as themes of “social engagement,” while American individuals preferred “socially disengaging” displays of pride and anger (Boiger et al., 2013; Kitayama et al., 2006).

The Physiological Side of Emotional Expression

Anyone who has ever noticed their heart accelerating in fear or anticipation knows that emotional experience is not just an abstract product of the human psyche, but also heavily intertwined with physiological processes. In this regard, the most important interface of the human body can be found in the autonomic nervous system (ANS), which controls a multitude of unconscious bodily functions and mirrors the affective state of the individual through parameters such as blood pressure or skin conductance level (Kreibig, 2010). At the same time, previous research remains indecisive on the question whether physiological reflections actually offer insight in the quality—other than just the intensity—of emotional experience. While some authors argue that autonomic response patterns may in fact be emotion-specific (e.g., Levenson, 2003) or at least convey emotional valence (Bensafi et al.,

2002; Brosschot & Thayer, 2003), recent research suggests that observable differences in bodily reactions merely deliver insight into underlying motivational systems (Mendes, 2016).

The search for emotion-specific, autonomic reactions is further complicated by the fact that changes within the ANS are not only evoked by emotional states, but also reflect many other mental processes such as acute stress (e.g., Dickerson & Kemeny, 2004), higher levels of concentration (e.g., Wass et al., 2016), and ruminative thoughts (e.g., Ottaviani et al., 2009).

On this account, it has become common practice to interpret increases of autonomic activity primarily as indicators of arousal, which can only be linked to specific emotions in controlled laboratory settings.

STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 87

Returning to this study's main topic of emotional display rules, one may also wonder

how emotional suppression efforts register in terms of physiological activity. However,

findings on this subject have pointed into two directions: Just as some studies indicate a

reduced physiological activation after suppression strategies (Zuckerman et al., 1981), others

have provided potent arguments for the increased physiological cost of emotionally

suppressive behavior (Butler et al., 2003; Gross & Levenson, 1993). A possible dissolution of this dispute might be found in cultural adaptation effects. For instance, an experiment conducted by Butler et al. (2009) has shown that emotion-expressive behavior led to an increase in blood pressure among Asian Americans but to a decrease among European

Americans. Similarly, a recent study from the field of neuroscience has reported that, after viewing unpleasant pictures, Asian Americans showed a much faster decrease of the brain potentials related to emotional processing than US Americans (Murata et al., 2013). As more and more similar findings emerge for various tasks and contexts (e.g., Shen et al., 2004; Zhou

& Bishop, 2012), the bulk of the argues for reduced physiological activity in East Asians as they, consciously and sub-consciously, regulate their emotional displays— an effect that might turn out quite differently for members of other cultures.

The Current Study

Introducing common findings from cross-cultural psychology to the research field of affective HCI, the current study set out to scrutinize how Chinese and German users would react to the impression of being “unmasked” by a computer-based form of emotional recognition. For a comprehensive examination of this response, we investigated both participants' physiological arousal as well as their subjective affinity to AFER software following a (supposedly) automatic reading of their facial emotions.

STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 88

The consulted literature provided us with two reasonable assumptions how the chosen cultural samples might differ in their reaction to the presented technology. On the one hand, it seemed highly likely that Chinese participants, as members of a collectivism-oriented culture, would perceive the AFER procedure as an unpleasant experience, considering that the recognition of facial micro-expressions serves as a substantial—and, compared to human interactions, unprecedented—form of “losing face.” In consequence, we found it logical to assume that individuals with this cultural background would show a more pronounced arousal reaction (meaning either a stronger or longer increase of physiological activity—or both). On the other hand, we pondered that the stronger sense of caution against humanlike technology in the West (e.g., Kaplan, 2004; MacDorman et al., 2009; Nomura et al., 2015) could just as well lead to more anxiety among German individuals, who might consider affective computers as a threat to human uniqueness. In consequence, we decided to juxtapose these contradicting hypotheses:

H1a: The physiological arousal measured after feedback from AFER software will

be more intense among Chinese participants.

H1b: The physiological arousal measured after feedback from AFER software will

be more intense among German participants.

Apart from physiological effects, we were also interested in participants' subjective evaluation of the presented technology. In our interpretation, the arguments that had led to our first set of hypotheses applied just as well to this research focus: If a cultural group would perceive the emotionally aware computer as discomforting and arousing, it seemed highly likely that they would also report a lower affinity to the system in question. As such, we again formulated a set of two conflicting assumptions, matching H1a and H1b: STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 89

H2a: The subjective evaluation of emotion recognition software will be less

favorable among Chinese participants.

H2b: The subjective evaluation of emotion recognition software will be less

favorable among German participants.

Compensating for the exploratory nature of our two-fold hypotheses, we included an additional measure to help us understand why one assumption might overrule the other:

Participants' attribution of human likeness to the presented AFER system. Doing so, we strived to find out whether our cultural groups differed in their perceptions of AFER (non-

)artificiality—and whether these attributions served as a mediator between culture and the eventual affinity to the presented technology.

RQ1: Do Chinese and German participants differ in terms of the human likeness

they ascribe to AFER technology?

RQ2: Is participants' AFER affinity mediated by these human likeness attributions?

Apart from potential main effects of culture, we were also interested if users' reactions were to turn out differently as soon as the digital system indicated a violation of—instead of a match with—cultural display rules from the inter-human context. For interactions with human strangers, East Asians have been shown to favor socially engaging emotions (even negative ones such as shame), while Westerners focus on more ego-focused and exclusively positive expressions (e.g., Boiger et al., 2013; Eid & Diener, 2001; Matsumoto et al., 2008a).

Considering this, we were curious to find out whether AFER systems would count as “just another stranger,” toward whom the discussed display norms would be fully in play. If so, the impression of having shown pride in front of the computer should emerge as unpleasant for

Chinese but not for German participants; the latter should dislike reports of ashamed facial expressions instead. STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 90

H3: If emotion recognition software reports facial displays that contradict culture-

specific display rules, participants will show more physiological arousal.

H4: If emotion recognition software reports facial displays that contradict culture-

specific display rules, participants will evaluate it less favorably.

Our interest in comparing different display rule conditions, however, heralded some methodological challenges. After some initial deliberation, we quickly disregarded the idea of artificially inducing specific facial expressions, because we were highly skeptical of the validity and reliability of this approach—especially since our study was supposed to revolve around rather subtle displays of emotion (e.g., expressions of pride). Of course, the alternative of waiting for participants to express a specific emotion without purposely stimulating it seemed just as impractical. An elimination of the identified problems eventually occurred in the form of a more deceptive approach. By providing participants with a fully standardized, faux result instead of a genuine reading of their facial expressions, we found a unique solution to manipulate display rule violations as needed. As an additional merit, this procedure allowed us to standardize the alleged intensity of emotional expressivity across participants—which would have been impossible otherwise. At the same time, we now had to contain the risk of participants doubting the provided results. For this purpose, our study was adjusted in two ways: First, during the initial presentation of the AFER system, we repeatedly emphasized that the technology would focus on so-called micro-expressions, which are hardly discernible to the human eye; secondly, we decided to have participants fill in a demanding cognitive test during the alleged analysis, thereby distracting them from a conscious monitoring of their own face.

STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 91

5.2 | Method

Participants

Following an a-priori analysis of required sample size with G*Power software (Faul

et al., 2007), we recruited 100 students from two ethnic groups at a German university: 51

participants who self-identified as Chinese (24 female, 27 male) and 49 participants who self-

identified as German (39 female, 10 male). To control for possible acculturation effects

among the Chinese participants who temporarily lived as exchange students in Germany, we

formulated two inclusion criteria: (a) having spent one’s youth in Mainland China, Taiwan,

or Hong Kong, and (b) speaking Chinese (e.g., Mandarin) as current main language.

According to previous research (e.g., Guan, 2007; Yu & Wang, 2011), the community of

Chinese exchange students in Germany remains highly connected to their home culture, yet extremely isolated within the host society. This has been confirmed by reports from our participants and colleagues, so that we consider our sample a valid reference for the experiences of young Chinese students. Moreover, the fact that our final sample consisted exclusively of individuals from Mainland China—without any students from Taiwan or Hong

Kong—lends further support to the homogeneity of this experimental group.

Following a manipulation check, a total of three participants (all from the German

sample) had to be excluded from further analysis, as they had doubted the authenticity of the

presented stimuli. Additionally, the data of seven participants (3 Chinese, 4 German) could

not be used due to technical difficulties in their physiological measurements. Lastly, one

Chinese participant had to quit the experiment early on account of his previously undisclosed

color blindness. Therefore, our final sample included 47 Chinese (age M = 26.1 years,

SD = 2.62) and 42 German students (age M = 22.6 years, SD = 3.93), for a total of 89

participants. As a compensation for his or her time, each participant could choose between €5

or partial course credits. STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 92

Procedure

At the beginning of the experiment, participants were told that the current study

served to explore cultural differences in facial micro-expressions during a cognitive task.

Pointing out the video camera and desktop PC in our laboratory, we explained that the experiment would involve state-of-the-art AFER software, which could “monitor [the user's] face for spontaneous muscle movements”—including “even the tiniest contractions”—and

“calculate a summary of the emotional displays during any given task.” As we were not

interested in participants' actual facial displays, but only in their reaction to a manipulated

recollection of it, we prepared two standardized result sheets, with one claiming that the

software had recognized pride and the other indicating shame in the user's micro-expressions.

This resulted in a 2 × 2 between-subject design as illustrated in Figure 12. Participants were

assigned to one of the two feedback conditions by means of a block randomization procedure.

Figure 12. The study’s between-subject design. Table cells contain each condition’s theoretical implications for cultural display rules.

STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 93

After they had filled out an informed consent form, we equipped participants with an

unobtrusive physiological monitoring wristband, recording their heart rate for the remainder

of the experiment. Following a 1-min baseline measurement (which was conducted while

sitting in silence), we then provided the materials for a short intelligence test and instructed

participants to keep their head directed toward the video camera for the duration of the task.

Under the deceptive impression that their facial displays were being monitored, participants

filled in the test for a timed duration of 3 min. Subsequently, they were instructed to access

the analysis' alleged result on a private screen—a method that we chose not only to prevent

potential feelings of humiliation in front of the study conductor, but also to put all focus on

the concept of being analyzed in a “non-human” context.

As soon as participants finished reading the provided results, they filled in a short questionnaire on their acceptance of the software, as well as a few control variables. Lastly, we took down each participant's e-mail address to provide them with an extensive explanation of the study's goals, its deceptive design, and our findings. Those who did not want to enter their e-mail address were debriefed directly and kindly asked to keep the experiment's true nature a secret until all recruited students had taken part.

Stimulus Design

As materials for the brief intelligence test, we used self-created matrix completion

tasks modeled after the widely-used Raven Progressive Matrices (Raven, Raven, & Court,

2003). In this type of test, participants have to choose one of eight options as the missing tile

of a 3 × 3 symbol matrix. Since the performance in matrix tasks does not depend on language

or factual knowledge, they are considered a culture-faire method of testing; even though the actual performance was not relevant to our hypotheses, we therefore chose this type of

cognitive test for our cross-cultural experiment. STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 94

For the believable introduction of an authentic AFER system, we used the recognition software MultiSense, which is freely available as part of the Virtual Human Toolkit (Hartholt et al., 2013). Among other functions, the software includes face tracking and a basic form of expressivity analysis, visualizing the results via various computer windows (e.g., camera feed with automatically aligned 3D grid). Although we chose not to record any actual facial data from our participants, we briefly showed them a live stream of their face within the

MultiSense environment at the beginning of the experiment in order to foster our deception of genuine facial recognition. Figure 13 depicts the visualization setup as it was presented to the participants on the experimenter's swiveling computer screen. Doing so, we always made sure to show the interface for only a couple of seconds and from a small distance, thus obscuring its technological specifics. Furthermore, to mitigate any privacy concerns, we explicitly stated that the software “worked in real-time” and would not need to save recordings of the participant's face at any time.

Figure 13. MultiSense visualization used for the deceptive narrative of emotion recognition. The main window (bottom right) shows a live feed of the user's face, with a 3D grid automatically fitted to relevant facial points. STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 95

Apart from the introduction of the software itself, the authentic presentation of its

manipulated results was absolutely crucial for the success of the experiment. As such, we

prepared a step-by-step procedure to convey the analysis' outcome in a believable way.

Firstly, we composed a web-based result sheet using both HTML and JavaScript code (see

Figure 14), which was able to dynamically display the metadata of each appointment,

including the participant's individual number, cultural group, and time of measurement. As

center part of the sheet, we prepared a table with fictional parameters for eight affective states

(anger, fear, sadness, happiness, surprise, trust, shame, and pride) in the style of existing

AFER frameworks (Doerrfeld, 2015). Most importantly, one of the table's rows was colored

in bright red, highlighting the according emotion as most prevalent feeling during the

intelligence test. Depending on experimental condition, this highlight was set on either pride

or shame, with all other emotions fixed at medium levels.

Figure 14. Procedure to convey the deceptive result sheet after the experimental task. (A) The login screen hosted on a local web server. (B) Result sheet with fictitious data. The allegedly most prevalent emotion is displayed in red—either shame or pride, depending on condition. STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 96

To explicitly direct participants' attention toward the relevant part of the sheet, the study conductor (who was sitting several feet away) was instructed to always tell them to

“focus on the scores in red, which indicate the predominant emotion during the experiment.”

Lastly, we compiled two additional graphs and added them to our sheet, further increasing the saliency of the relevant facial expressions. The completed result page was then translated from German into Simplified Chinese, with back-translations ensuring the similarity of both versions.

Measures

Cardiovascular activity. We used an Empatica E4 physiological measurement wristband (Empatica Inc., 2017) to achieve unobtrusive monitoring of our participants' heart rate. Although the Empatica E4 is able to measure heart rate and heart rate variability with a frequency of 1 Hz (one data point per second), we averaged the values of 60 s to achieve a meaningful reduction of data. During the experiment, the study conductor marked the exact second in which participants opened the manipulated AFER result sheet as time of stimulus onset. For a theoretically coherent analysis, however, we always added 5 s to this time to account for a basic cognitive processing of the presented stimuli. Since the cardiovascular system is known to respond more slowly to stressors than other physiological indicators, we aligned our procedure with previous studies, which have examined autonomic arousal and recovery during the first few minutes after stimulus onset (e.g., Boer, 2016; Brosschot &

Thayer, 2003; Roberts et al., 2008). As a result, the following measures were obtained for all participants: (1) a 1-minute baseline of resting heart rate, starting shortly after the wristband had been equipped; (2) the average heart rate during the first minute after stimulus presentation, starting with a delay of 5 s; (3) the average heart rate during the subsequent second minute after stimulus presentation. STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 97

Attractiveness and human likeness of the presented technology. Participants rated

their subjective impression of the presented software using the technology-related attractiveness index by Ho and MacDorman (2010), which captures the affective response toward a technological stimulus. Although the questionnaire's five semantic differentials

(e.g., “repulsive—agreeable,” “messy—sleek”; rated on a 7-point scale) were originally designed to address visual features in relation to the “uncanny valley” phenomenon, we found them suitable for an evaluation of our more abstract stimuli as well; in fact, the authors suggest that their index basically pinpoints participants' reaction on an evolutionary avoidance-approach continuum, which strongly correlates to perceptions of interpersonal warmth. Due to our study's cross-cultural nature, we translated the original English items into

German and Simplified Chinese and used back-translations by native speakers to ensure semantic equivalence. Both translations proved to be of acceptable to high internal consistency (Chinese version, α = 0.70; German version, α = 0.81). To establish measurement invariance between both versions, we conducted a series of increasingly restrictive confirmatory factor analyses (CFAs), which is a common procedure to test measures for configural, metric, and scalar invariance. Doing so, partial scalar invariance could be established for our translated attractiveness indices. Table 6 gives an overview of the conducted model comparisons.

For the exploratory investigation of human likeness attributions to the AFER technology, we used Ho and MacDorman’s human likeness index (2010), which assesses the amount of animacy and human nature ascribed to a technology (with artificiality and synthetic nature as other endpoints of the spectrum). Whereas the original version of the measure consists of six semantic differentials (e.g., “human-made—humanlike,” “artificial— lifelike”), we excluded two items that did not apply to our disembodied scenario, namely

“biological movement—mechanical movement” and “without definitive lifespan—mortal.” STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 98

The resulting four item scale was again translated, with internal consistency turning out a bit

lower for the Chinese (α = 0.63) than for the German version (α = 0.79). In our interpretation,

this concurs with the reviewed literature as it suggests a more complex understanding of

human likeness in Chinese tradition. Nevertheless, multi-group CFA testing for measurement invariance again indicated partial scalar invariance between our two translations (see Table 6) so that we still included the measure in the exploratory part of our study.

Manipulation checks. By design, the current study did not focus on participants' real emotional displays; quite the opposite, we aimed at convincing them of a standardized result.

Although various efforts were expended to facilitate this goal (e.g., emphasizing the role of micro-expressions, distracting participants from their face, and keeping the reported affect at a moderate level), we also decided to include some form of measurement to assess the success of our deception. Specifically, we asked participants to rate the accuracy of the software's analysis on a self-developed two item scale (α = .84)—which was then used to identify the cases where had to assume a discrepancy between the provided feedback and a person's self-perception. As a conservative rule, all participants who had filled in the lowest score (1 out of 5) on one or both items were completely excluded from our study. Eventually, this was the case for three participants from the German sub-sample, who reported frequent participation in psychological studies and, as such, might have been especially wary of potential deceptions.

S TUDY Table 6. Multi-group confirmatory factorial analyses to check translated scales for measurement invariance. II I :

“

Constrained to be equal χ² df CFI RMSEA S

Scale Model (Δp) AVING across groups (Δ χ²) (Δdf) (ΔCFI) (ΔRMSEA) attractiveness M1 Configural invariance factor structure 16.128 10 - .947 .117 F

M2 Metric invariance factor structure, loadings (2.814) (4) (.59) (.010) (.028) IN ACE factor structure, loadings, Scalar invariance (19.784) (4) (< .01**) (.136) (.072) M3a intercepts F RONT OF THE THE OF RONT M3b Partial scalar invariance factor structure, loadings, (1.609) (2) (.45) (.003) (.009) (except items 4 and 5) intercepts factor structure, loadings, Structural invariance (0.083) (1) (.77) (.008) (.011) M4 intercepts, means C

human likeness M1 Configural invariance factor structure 3.058 4 - 1.000 .000 OMPUTER M2 Metric invariance factor structure, loadings (1.179) (3) (.76) (.000) (.000) factor structure, loadings, M3a Scalar invariance (9.609) (3) (< .05*) (.055) (.093) ? intercepts ”

M3b Partial scalar invariance factor structure, loadings, (4.712) (2) (.09) (.000) (.000) (except item 4) intercepts factor structure, loadings, Structural invariance (7.329) (1) (< .01**) (.090) (.119) M4 intercepts, means

Notes. According to Chen (2007), ΔCFI < .010 and ΔRMSEA < .015 indicate that the invariance assumption also holds for the more restricted model. *Δp < .05, **Δp < .01. 9 9

STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 100

To gain additional insight into our manipulation's validity, we inquired participants to

rate their own performance in the matrix completion task on a 5-point scale. By connecting

these ratings to the actual test results, we were able to assess whether participants could

actually estimate their accomplishment in a realistic manner—which would have disrupted

our manipulation. Fortunately, our analysis showed that both groups had very little insight

into their true performance.

5.3 | Results

Manipulation Checks

Software accuracy. To check the evaluations of software accuracy for significant

group differences, we calculated a two-way analysis of variance (ANOVA) with the between-

subject factors culture and type of feedback. The procedure yielded no significant main effect

for the latter, F(1,85) = 0.81, p = .81, and no significant interaction between factors, F(1,85)

= 2.51, p = .11. Accordingly, we note that reports of both “proud” and “ashamed” facial

displays were seen as moderately accurate, regardless of cultural background. However, we

did observe a significant main effect of culture, F(1,85) = 5.43, p = .02, ηp² = .06—Chinese

participants generally ascribed higher accuracy to the presented software (M = 3.49,

SD = 0.84) than German participants (M = 3.08, SD = 0.80). While this might somewhat reflect the stronger skepticism of the native-speaking students at our university—who typically take part in more psychological studies than exchange students—our finding could also just emphasize the stronger belief in technological prowess among the Chinese. In any case, we report that both group means manifested slightly above the scale’s midpoint, so that we deem our manipulation acceptably successful.

Actual and perceived test performance. A two-way ANOVA with participants’ actual test results as a dependent variable revealed no significant differences between Chinese and German participants, F(1,85) = 0.32, p = .57. On average, German students solved STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 101

M = 5.76 test items correctly (SD = 1.76), closely matching the performance of the Chinese students (M = 5.60 correct answers, SD = 1.54). However, German participants facilitated

their scores with a higher number of total answers (MGER = 9.83, MCN = 8.17), including more

wrong answers (MGER = 4.07, MCN = 2.57). We further investigated whether the two feedback

groups had, by chance, produced significantly different test results, which may have been

problematic for the believability of the deception. However, this was not the case,

F(1,85) = 2.20, p = .14.

In terms of participants’ own perception of their task performance, another two-way

ANOVA with the factors culture and type of feedback revealed a strong main effect of

culture, F(1, 85) = 11.36, p < .01, ηp² = .12. Indeed, our data show that Chinese participants considered their performance significantly better (M = 3.64, SD = 0.82) than German participants (M = 3.05, SD = 0.83). While there was no notable main effect for type of feedback, an interaction between both factors emerged as marginally significant, F(1, 85) =

4.14, p = .05, albeit with a very small effect size of ηp² = .03. Examining our data pattern, we

observed that only Chinese participants rated their performance significantly higher if the

software had indicated “proud” (M = 3.92, SD = 0.64) instead of “ashamed” facial displays

(M = 3.32, SD = 0.89); for the German participants, self-assessment in both “proud” (M =

3.00, SD = 0.97) and “ashamed” conditions (M = 3.09, SD = 0.68) turned out rather similar.

On account of the effect’s marginal size and significance, however, we advise to interpret this finding with caution.

As our final, but probably most crucial manipulation check, we conducted two separate linear regression analyses to find out if the real test result predicted the self- assessment from participants of both groups. This investigation did not result in significant regression equations, neither for Chinese [F(1, 46) = 1.14, p = .29] nor for German [F(1, 41)

= 0.34, p = .56] students. Hence, we argue that participants had relatively little insight if their STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 102

performance had been good or bad, which certainly supported our manipulation of alleged

micro-expressions.

Cardiovascular Activity

Table 7 shows the means and standard deviations for participants’ heart rate during

the baseline measurement, as well as the two minutes after stimulus presentation. For reasons of clarity, the table also contains the relative differences between subsequent data points. As an additional illustration, the graphs in Figure 15 depict the average heart rate changes in the four experimental groups, compared to their respective baseline values.

Table 7. Descriptive statistics for heart rates and relative heart rate changes between the three measuring points.

heart rate in beats per minute (bpm) baseline 1st minute 2nd minute

culture feedback M (SD) M (SD) ∆ to prev. M (SD) ∆ to prev.

Chinese ashamed 79.35 82.29 + 3.7% 80.15 - 2.6% (11.89) (12.88) (12.97) 77.95 82.17 81.40 proud + 5.4% - 0.9% (7.77) (12.00) (11.07)

total 78.60 82.23 80.82 - 1.7% (9.82) (12.28) + 4.6% (11.88)

German ashamed 76.48 79.06 + 3.4% 79.22 + 0.2% (8.14) (9.00) (7.64)

proud 76.86 80.69 + 5.0% 81.33 + 0.8% (8.78) (9.08) (8.88)

total 76.66 79.84 + 4.1% 80.23 + 0.5% (8.35) (8.97) (8.22)

STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 103

Figure 15. The four experimental groups’ heart rate changes in bpm, compared to their respective baseline value.

To check the acquired data for statistically meaningful differences in physiological

activity, we focused on the heart rate changes during the first (initial arousal) and second

minute (sustained arousal) with two separate analyses of covariance. In both ANCOVA

procedures, we controlled for participants' age and gender by entering them as covariates, as

these variables have been shown to profoundly influence cardiovascular (re-)activity (e.g.,

Carroll et al., 2000).

A 2 (culture) × 2 (type of feedback) ANCOVA with the mean heart rate change between baseline and first minute post-stimulus as dependent variable resulted in no significant main effects for culture [F(1, 83) = 0.05, p = .81], type of feedback [F(1, 83) =

1.39, p = .24], or interaction between both [F(1, 83) = 0.06, p = .80]. In terms of participants' initial cardiovascular response, we therefore conclude that there was no meaningful difference between the selected cultural groups—or that the effect was too small to be detected by the test power achieved with our sample size. The same applies to our STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 104

assumptions about the type of given feedback. Although the data of the Chinese participants

indeed suggest more initial arousal if the AFER result had claimed micro-expressions of

pride (M = +4.22 bpm, SD = 8.56) instead of shame (M = +2.94 bpm, SD = 7.50), the observed effect missed the threshold of statistical significance and should therefore be interpreted cautiously. For the German participants, the influence of the given feedback was even smaller.

However, using the heart rate change between the first and second minute post- stimulus as a dependent variable, another ANCOVA yielded a significant main effect of culture, F(1, 83) = 6.26, p = .01, with a moderate effect size of ηp² = .07. At the same time, no

significant effects were found for type of feedback [F(1, 83) = 1.22, p = .27] or factor

interaction [F(1, 83) = 0.03, p = .87]. Thus, in terms of “sustained” arousal, German

participants remained more agitated after the computerized recognition procedure, even

showing yet another increase in heart rate (M = +0.39 bpm, SD = 5.53). The Chinese group,

on the other hand, reduced their heart rate by M = −1.41 bpm (SD = 3.98) during the second

minute after stimulus presentation.

In summary, our findings offer first evidence in favor of hypothesis H1b over H1a:

Even though their initial response was not stronger per se, we argue that the longer arousal in

the German sub-sample does constitute a more intense physiological reaction to the AFER

feedback. At the same time, hypothesis H3 could not be confirmed by our data, as the

manipulated feedback had no noteworthy effect on the arousal of our participants (if potential

type II errors are dismissed).

Subjective impression of the software

Table 8 gives an overview of the scores yielded from this study's self-report measures.

To check both ratings for statistically significant group differences, we conducted a pair of 2

× 2 ANOVAs, again using culture and type of feedback as between-subject factors. STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 105

Table 8. Descriptive statistics for self-report measures.

human likeness attractiveness culture feedback M (SD) M (SD)

Chinese ashamed 2.82 (0.81) 3.62 (0.54)

proud 3.07 (0.71) 3.78 (0.52)

total 2.95 (0.76) 3.71 (0.53)

German ashamed 2.51 (0.90) 3.63 (0.59)

proud 2.29 (0.78) 3.45 (0.53)

total 2.40 (0.84) 3.54 (0.56)

Notes. Both scales range from 1 to 5.

The first ANOVA focusing on participants' attractiveness ratings did not uncover any significant effects, neither main effects for culture [F(1, 85) = 1.962, p = .17] or type of feedback [F(1, 85) = 0.01, p = .96], nor an interaction between both [F(1, 85) = 2.19, p =

.14]. As we found almost identical attractiveness ratings in all four groups, our results supported neither hypothesis H2a nor H2b: The cultural groups did not differ in their subjective liking of the presented stimuli. As alleged violations of human-to-human display rules hardly influenced this as well, we further reject hypothesis H4.

Focusing on our second subjective measure, a two-way analysis of variance with human likeness as the dependent variable resulted in no significant main effect for type of feedback, F(1, 85) = 0.01, p = .94, and no interaction between feedback and culture, F(1, 85)

= 1.95, p = 0.17. However, we now obtained a significant main effect of culture, F(1, 85) =

10.24, p < .01, with a very large effect size of ηp² = .11. As expected from the literature review, Chinese participants showed a notably higher attribution of humanness to the affective software (M = 2.95, SD = 0.76) than German participants (M = 2.40, SD = 0.84), therefore giving a positive answer to RQ1.

STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 106

Mediation Analysis

Our exploration of culture as a quasi-experimental macro-level predictor had not resulted in a clear empirical effect on technology attractiveness. However, to address our question if human uniqueness concepts might actually stand as a mediator between these two constructs (RQ2), we proceeded to a mediation analysis using the PROCESS macro for IBM

SPSS (Hayes, 2013), using five thousand iterations for bootstrap confidence intervals.

Indeed, the procedure uncovered a significant indirect effect from culture over human likeness attributions to attractiveness ratings, b = −0.09, SE = 0.05, 95% CI [−0.23, −0.01], with the mediator accounting for roughly half of the total effect, PM = 0.55. Figure 16 gives

an overview of all obtained standardized regression coefficients in the mediation analysis;

due to the dummy coding of our cultural groups (0 = “Chinese”, 1 = “German”), the negative

coefficient indicates higher outcomes among the Chinese sample. In particular, our data show

that having a Chinese cultural background significantly predicted higher human likeness

attributions, which in turn predicted higher attractiveness ratings. Apart from this indirect

effect, the direct relationship between culture and technology attractiveness remained

insignificant. RQ2 can therefore be answered positively.

Figure 16. Standardized regression coefficients for the relationship between culture and perceived technology attractiveness as mediated by ascribed technology human likeness. (*p < .05; **p < .01).

STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 107

5.4 | Discussion

To investigate if emotion-sensitive forms of human-computer interaction are

influenced by sociocultural factors, we introduced Chinese and German participants to a

facial recognition system and prepared its results to either match or violate traditional norms

for emotional expressions toward strangers. Measuring cardiovascular parameters as well as

subjective impressions of the AFER software, we addressed both the subconscious and

conscious processing of this new, digitally mediated form of “becoming unmasked.”

Based on an interdisciplinary review of literature, two contrasting assumptions

emerged on how Chinese and German individuals might compare in their experience of

automatic facial emotion analyses. Specifically, we contemplated that either the Western

preference for candid facial expressions or the East Asian tradition to accept “human-like”

qualities in non-human entities as something completely natural would eventually tip the

scales in favor of one of the two groups. Our results suggest the latter. Despite comparable

increases of cardiovascular activity immediately after the presentation of the AFER results,

Chinese participants soon returned to a notably lower heart rate, whereas German participants

lingered in a state of sustained arousal. Accordingly, we report a response pattern that

matches findings by Matsumoto et al. (2009), suggesting that the initial response to a

stimulus might be universal before slightly delayed “cultural influences kick in” (p. 1273).

Considering that complex artificiality is often seen as a threat in Western cultures (Kaplan,

2004; Złotowski et al., 2017), we argue that the arousal of the German students might indeed indicate some form of anxiety, i.e., an autonomic reflection of their subliminal wariness toward the novel technology. At the same time, it has to be noted that changes in cardiovascular activity do not allow a direct interpretation of emotional quality: Being aroused by a stimulus could just as well signal curiosity or positive excitement. Similarly, we deem it possible that the quicker regulation of Chinese participants' heart rate is heavily STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 108 influenced by their proficiency in autonomic emotional suppression, which has been demonstrated by previous experiments (e.g., Shen et al., 2004; Zhou & Bishop, 2012). With these limitations in mind, we suggest that the reported main effect of culture on the physiological arousal evoked by AFER is interpreted conservatively.

At the same time, our investigation of human likeness evaluations and their revealed role as a mediator between culture and the final AFER affinity clearly support the suggested interpretation of our results. Whereas cultural background showed no isolated effect on the attractiveness participants ascribed to the emotionally aware system, we found that our groups differed greatly in how “animate” and “human-like” they considered the presented computer, which in turn predicted the final attractiveness ratings. In our explanation, this implies that culture as a macro-level container for many confounding variables may not suffice to completely explain views on technology—yet be essential in forming people's basic philosophy and worldview (e.g., the idea of what constitutes a human-natured entity), which then interacts with other individual dispositions to form actual attitudes and behaviors. For

HCI developers, this builds toward an unambiguous recommendation: Only by tapping into both cross-cultural and individual-level forms of research, they might eventually settle on the right amount of “human” that customers from different backgrounds would like to rediscover in their technology.

Lastly, we report that different types of feedback by the emotion recognition system had very little impact on participants' experience in our scenario. Empirically, it did not matter whether the computer claimed displays of pride or shame as the most prevalent facial expression, neither for participants' arousal nor for their liking of the software—and regardless of cultural background. Providing an interpretation of these findings, we suggest that the confidential reveal in our experiment (the AFER results were conveyed on a private screen) must have weakened the perceived importance of traditional display rules by a great STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 109 extent. Since a recognition system that is used privately has no bearing on one's social standing or any in-group coherence, people might be completely indifferent about culturally desirable expressions in such a setting. Based on this argument, however, we would expect much stronger effects once the technology was used to trigger meaningful real-life consequences. Considering that psychological counseling, e-learning, and job assessments are all being targeted as application fields for AFER, we expect various use cases in which a much stronger need for “appropriate” facial displays will arise—may it be a virtual classroom or a automatic job selection procedure. As such, we strongly believe that, even though they remained insignificant in our single-user experiment, traditional display rules will instantly find new relevance once emotionally aware systems turn into actual mediators of interdependence, social standing, or financial success.

Limitations

Our results are limited in their generalizability due to the use of a convenience sample that consisted exclusively of students ranging between 18 and 37 years. Similarly, we note a slightly uneven distribution of female and male participants, especially in the German group.

Although we tried to control for these factors (e.g., by including them as covariates or additional predictors in our statistical tests), different results might emerge if other samples, for instance children or elderly participants, were to experience the provided scenario. This also applies to participants' level of education: Since our student sample represents only a small part of the socio-economic spectrum, we consider it highly likely that other findings would be obtained with participants pursuing other occupations.

In regard to the cultural comparison conducted in this study, we note that both the

Chinese and the German group consisted of individuals from different parts of the respective country, which potentially underestimates regional influences. However, in light of the long cultural distance between both nations, we still think that our study achieved an insightful STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 110

juxtaposition of cultural differences in the perception and experience of affective technology.

Nevertheless, we think that future studies might benefit from explicitly asking participants

about their reliance on cultural values. Qualitative methods such as semi-structured interviews might be particularly useful in this regard, promising a less ambiguous understanding of how users' socialization has contributed to their conceptualization of emotional appropriateness, preference for collectivistic values, and, consequently, their reaction to AFER analyses.

Conclusion

From an early age, most people are socialized to adjust their emotional output as soon as they interact with other perceptive entities—that is, typically, with other humans. Yet, due to breakthroughs in AFER technology, computers are now also able to read emotions from the human face, thereby entering the world of emotional communication as an exciting new intermediary.

At first, we pondered the possibility that common display rules might simply be transferred one-to-one to the interaction with emotion-sensitive technology. However, based on our empirical efforts, we have come to the conclusion that AFER systems do not provoke the same concerns about “appropriate” facial displays that are common among humans—at least as long as they possess only limited influence on other outcomes. In isolated interactions with “benevolent” affective computers, people simply have no incentive to be anxious about a certain kind of result. For the near future, however, we predict that real-life applications of

AFER will hardly stay as inconsequential or innocuous as our experimental scenario. Clearly, the technology has not been designed as a “single-player gimmick” but as a method of collecting data in numerous contexts. In current times, it seems all but far-fetched to envision autonomous cars, medical robots, or automatic job interview systems whose emotional perceptiveness all but determines crucial outcomes for the humans involved; and the STUDY III: “SAVING FACE IN FRONT OF THE COMPUTER?” 111 establishment of such procedures will surely have users turn back to their human-to-human norms of behavior. If our study is any indication, it will depend both on cultural and individual factors whether society appreciates this development—or faces it with an underlying anxiety.

GENERAL DISCUSSION 112

6 | General Discussion

“Besen, Besen, seid’s gewesen” (engl. “broomstick, broomstick, to the corner”)—with

this simple phrase from the sorcerer’s mouth, the horrors conjured by his apprentice can be

stopped in Goethe’s world-famous ballad (1798, pp. 36–37). Proceeding from fiction to

empirical reality, the findings of my research serve as evidence that this might be exactly the

level of control people desire when interacting with the autonomous technology of the 21st

century. As participants’ aversion was significantly related to their situational control and

threat experience in a naturalistic VR scenario, I have little doubt that the successful adoption

of elaborate neural network and deep learning systems will clearly depend on perceptions of

controllability. In practice, developers might argue that this contradicts their economic goals—after all, ‘always-on’ devices promise bigger collections of valuable data, and smartphone assistants that learn how to support their users’ daily routines by accessing all

kinds of information promise to make their owners dependent on them. But on account of my

obtained findings, I would argue that the inclination to deprive customers of their power of

decision only makes sense in the short run. As soon as people notice their lack of control,

they will feel more like the helpless apprentice than the powerful sorcerer; and my research

suggests that this will make them lose a lot of desire to engage with the technology in

question.

Admittedly, the findings described in the previous chapters may not completely

warrant pessimism of this gravity. In the face of my self-developed AI stimuli, the number of participants who were explicitly uncomfortable turned out rather low, whereas many others expressed their curiosity about the technology presented in the studies. Of course, this relative ‘harmlessness’ of the experiments was fully intended—research needs to stay ethical and refrain from exposing human subjects to unnecessary stress or trauma (German

Psychological Society, 2016). However, in my interpretation, this does not take away from GENERAL DISCUSSION 113

the verdict that reducing people’s impression of being in control led to a noticeable change of

their comfort in two subsequent experiments. In Study 1, it was already enough to change the perception of the exact same character from ‘human’ to ‘autonomous agent’ in order to evoke feelings of strange and spine-tingling eeriness; in the second VR study, individual perceptions of control and predictability turned out as strong predictors of users’ threat and attractiveness evaluations. If anything, the fact that these relationships could already be uncovered in rather benevolent and harmless settings speaks even louder to the importance of the demonstrated effects.

Taken together, I therefore suggest that my work adds up to quite irrefutable recommendations for future developments of autonomous (and, especially, emotion- sensitive) technology. Considering that “feeling threatened” will hardly work as a sales pitch for innovative products, I suggest that designers of digital systems turn controllability into a crucial mantra for new inventions. What might seem like a paradox at first—the idea of autonomous technology is to make it more self-controlled after all—could potentially be resolved by preparing different levels of decisional power: a ‘hierarchy of control’, so to speak. In all probability, the neural network that autonomously learns about its environment in the most astonishing way will instantly lose much of its threatening nature if it regularly asks for permission, hence reserving the ultimate power for the human in charge. Without a doubt, this will only become more important as soon as physical bodies enter the equation.

Thinking of robots with hydraulic extremities, which inevitably enable them to hurt observers, I suggest that programmers equip their machines with constant reaffirmations whether an intended action is acceptable (e.g., “May I approach you?” or “May I read your mood?”). The robotic politeness might seem overtly artificial or ridiculous at first, but taking my empirical insight as a starting point, it might actually be the wiser choice to boost users’ situational control by all means available. In the end, this could even include what has been GENERAL DISCUSSION 114 labeled a ‘kill switch’ by dystopian fiction: Taking the option to immediately terminate all processing and/or movement and embedding it so deep into the system’s artificial brain that no malfunction could potentially prevent its use. Inspired by my literature review, the conducted experiments, and the many conversations with my participants, I have come under the impression that supplying users with this tool could be highly beneficial to avoid their disregard of a technology. At the same time, it has to be considered that the concept of a kill switch becomes much more complicated as soon as artificial intelligence acquires a noticeable form of self-awareness—because from an ethical perspective, this would turn the metaphoric use of the word ‘killing’ into an actual termination of life (McMahan, 2002). As long as digital systems do not show convincing signs of a phenomenal consciousness, however, I suggest that the implementation of the discussed override functions remains much more innocuous and should be considered by developers.

Of course, all of the discussed measures only address one half of the developed Model of Autonomous Technology Threat, namely its proximal facet. Although the second component (regarding concepts of human identity) turned out less influential than expected in my second study, I think that a combination of my various results does lend support to the validity of human uniqueness attributions as an antecedent of the acceptance of emotion- sensitive AI. More precisely, I observed that interacting with embodied empathic systems turned the perception of human qualities into something negative, whereas the impression of a disembodied AI with the ability to recognize emotions seemed to benefit from higher levels of anthropomorphism. For the theorized threat model, this essentially suggests an interaction effect of proximal and distal factors; on the other hand, it has to be noted that my studies only provide initial evidence in this direction. For truly conclusive statements, much more research is needed.

GENERAL DISCUSSION 115

6.1 | A Tentative “Uncanny Valley of Mind”

In its introduction, my dissertation has demonstrated how the novel concept of an

Uncanny Valley of Mind could describe people’s acceptance of autonomous and highly intelligent technologies as a function of attributed mental capacities. Although the number of experiments conducted for this thesis hardly suffice to provide a comprehensive ‘topography’ of the hypothesized phenomenon, my results at least allow for some adjustments to the initial theory.

Figure 17. Modified concept of the Uncanny Valley of Mind.

By integrating the three presented studies, I suggest a modified pattern in which entities with high physical presence indeed fall into the proposed valley, whereas technology with low physicality actually seems to avoid the assumed user aversion—at least at the investigated level of mental competence. At the same time, I suppose that even non- embodied systems will ultimately reach a threshold of ‘attributed mind’ that comes a little too close to human experience. The alternative scenario, which would see non-physical AI becoming only more accepted as it successively reaches complete equivalence to the human mind, just contradicts too much of the reviewed literature. In consequence, I presume that the GENERAL DISCUSSION 116

Uncanny Valley of Mind for this type of entity will eventually arrive, albeit significantly later

than for AI with physical embodiments. Regardless of any new assumptions, however, I have

to acknowledge that this level of mental competence was not yet addressed by my empirical

work, so that I have designed the graph of the tentative Uncanny Valley of Mind to end in

ambiguous slopes (Fig. 17), allowing for both possibilities.

6.2 | Limitations and Future Work

Although I have expended my best efforts to turn this dissertation into an integral and

methodologically sound contribution to the emerging field of human-autonomous-technology

interaction, several limitations have to be noted about the conducted work. Foremost, I would

like to point out that my three studies each focused on a different form of AI embodiment, yet

did not involve the comparison of such variants within a single experiment. While one might

instinctively want to proceed with comparing the obtained data (e.g., the ‘attractiveness’

scores) across studies, this should be considered as problematic. Not only would a

comparison between different experimental designs ignore the role of potentially numerous

confounding variables, it is also impeded by the fact that the third study had to disregard one

of the used UV scales due to its intercultural obscurity. For the continuation of the documented research, this means that AI embodiment should definitely be turned into an experimental factor within a singular follow-up study. By manipulating the design of an empathic AI’s body within the same scenario, a more robust examination of interaction effects will be achieved—which in turn may help tremendously to advance the mapping of the suggested Uncanny Valley of Mind. Most of all, I suggest that these follow-up projects should also include tangible embodiments to increase the significance of the proposed influence of situational perceptions. Even though the high immersion of my VR experiments arguably resulted in participants reacting to the presented agents as if they were physically GENERAL DISCUSSION 117

present, the introduction of robotic entities would surely be beneficial for a deeper

understanding of the theorized relationships.

Another approach that could yield fascinating insights concerning mind-body

interactions would be to differentiate not only between different quantities of physicality (i.e.,

merely walking along a spectrum between bodiless interfaces and humanoid robots), but also

to focus on different qualities of artificial bodies—thus melting more traditional aspects of

UV research into the proposed theory. For instance, previous investigations on robot design

have indicated that aesthetic factors such as perceptions of cuteness might offer a powerful

protective mechanism against users’ aversion to human-like replicas (Dzieza, 2014; Hanson,

2006). According to Rosenthal-von der Pütten and Krämer (2014), this effect may be explained by the evolutionary biological phenomenon of the baby-scheme, which connects certain visual stimuli to attributions of a submissive and non-threatening nature. On the one hand, a possible resilience on account of this effect matches my developed Model of

Autonomous Technology Threat perfectly; on the other hand, it has to be considered that the perceived cuteness of entities (animals in particular) also increases the level of mental

competence ascribed to them (Wegner & Gray, 2016). Following this line of thought, it could also be possible that an entity that already approaches the Uncanny Valley of Mind through its observable behavior is actually pushed into truly discomforting territory by its distinctly

‘adorable’ appearance. Due to the emergence of two contradictory assumptions, future

experiments that compare AI embodiments of different human likeness, cuteness, and general

aesthetic appeal should yield especially interesting results.

Apart from the mentioned problem of impeded comparability, the conducted studies have been subject to several other methodological shortcomings. Most of all, all three

experiments included deceptive elements and followed a Wizard-of-Oz approach, which was chosen for practical reasons (standardization of AI behavior and affordability). Although I GENERAL DISCUSSION 118

made sure to include straightforward manipulation checks and conscientiously excluded

participants at the first sign of disbelief, it would certainly be beneficial to complement these scenarios with more truthful settings. In my reflection on this thesis, I cannot rule out that

some participants doubted the presented technologies or answered according to social

desirability—despite my best efforts to provide highly authentic stimuli.

Lastly, it should be noted that all empirical work reported in this dissertation made use

of convenience sampling methods, which revolved around recruiting students at the local

university. As such, the collected samples present only a small fracture of the general

society—not only in numbers, but also concerning distributions of age, socio-economic status, religion, and cultural background. Across all three experiments, no participant exceeded the age of 38, and questions on other demographic data usually resulted in the same answers: A high level of education, an atheistic worldview, and German descent (apart from

Study 3, in which half the sample consisted of self-identified Chinese individuals). Based on this limitation, the described effects are clearly restricted in their generalizability. Just as

elderly people have been shown to be much more skeptical of novel technology

(Buckingham, 2013), I would expect the data of fundamentalist Christian or Muslim samples

to involve much more hostility towards emotion-sensitive AI than indicated by my mostly

atheistic participants. In the end, this might be especially relevant for the developed distal

factor in my Model of Autonomous Technology Threat—assuming that other samples would

offer much more variance in this regard, the influence of attitudinal factors has probably been

underestimated by the results described in this dissertation.

6.3 | Concluding Remarks

Artificial intelligence is en route to become one of the most defining inventions in

human history. Even if one disregards the more extreme scenarios that have been theorized—

e.g., the idea of a technological singularity, after which intelligent systems enter an GENERAL DISCUSSION 119 unstoppable mode of exponential self-enhancement—the relentless improvement of digital minds undoubtedly heralds rather dramatic changes for human society. Somewhat disconcertingly, even technology enthusiasts and leading AI developers have expressed their doubts whether the entered path might really lead to the improvement, or rather the downfall, of human civilization. A prominent ambassador for a more cautious perspective is Elon

Musk, who is regarded as one of the most successful technological entrepreneurs of the 21st century; Musk has recently warned that we might actually be assembling “an immortal dictator from which we can never escape” (Paine, 2018). In contrast to this, more optimistic industry analysts predict that AI applications such as autonomous driving or cancer detection are bound to save millions of lives and should therefore be pursued without hesitation (e.g.,

Kalra & Groves, 2017). Complicating this discussion further is the fact that all outcomes of the imminent technological revolution not only rest upon the benevolence of the emerging autonomous systems, but also on the motives of the (human) companies behind them. Even if the elaborate digital minds of the near to distant future continue to obey their human creators—either voluntarily or because they are bound by behavioral restrictions similar to

Asimov’s Three Laws of Robotics4 (1950)—they are still the product of an industry with its own motives and economic goals.

In conclusion, one might say that a lot of uncertainty surrounds the imminent technological era; in my opinion, this translates into a very important task for social scientists. Most of all, developers and political decision-makers have to be given valid information about the needs and worries of the broader public to guide them in their quest for innovation—which is a task that only psychological research may be able to achieve. Even if the human mind interacting with an AI system may initially constitute a ‘black box’ as

4 In his short story “Runaround”, science fiction author Isaac Asimov introduced three fundamental principles that could be imposed on intelligent robots to avoid conflict: The machines should (1) never injure humans, even if it happens through inaction, (2) always obey orders as long as the first law remains intact, and (3) protect their own existence if it does not conflict with the other two principles. GENERAL DISCUSSION 120

obscure as the digital entity itself, concentrated academic efforts will surely manage to shed

some light on the desires most common among human society. Assuming that our civilization is indeed approaching a crossroads between continued man-machine coexistence and potential conflict, it would be only appropriate if scientific insight was used to decide how much of our perceived human uniqueness is abandoned. Surely, every lost differentia specifica could be compensated by potentially marvelous gains; yet, the consequences of AI advancement are simply too grave to just belong to an elite group of entrepreneurs. REFERENCES 121

References

Adee, S. (2015, May 13). Say hello to machines that read your emotions to make you happy. New Scientist. Retrieved from https://www.newscientist.com/article/mg22630212- 900-say-hello-to-machines-that-read-your-emotions-to-make-you-happy/

Adolphs, R. (1999). Social cognition and the human brain. Trends in Cognitive Sciences, 3(12), 469–479. doi:10.1016/S1364-6613(99)01399-6

Allison, A. (2006). Millennial monsters: Japanese toys and the global imagination. Berkeley: University of California Press.

Appel, M., Weber, S., Krause, S., & Mara, M. (2016). On the eeriness of service robots with emotional capabilities. In Proceedings of the Eleventh ACM/IEEE International Conference on Human Robot Interaction, Christchurch, New Zealand (pp. 411–412). doi:10.1109/HRI.2016.7451781

Århem, K. (2016). Southeast Asian animism in context. In K. Århem & G. Sprenger (Eds.), Animism in Southeast Asia (pp. 3–30). Abingdon: Routledge.

Asimov, I. (1950). I, robot. New York City: Gnome Press.

Azar, B. (2005). How mimicry begat culture. Monitor on Psychology, 36(9), 54.

Bailenson, J. N., Blascovich, J., Beall, A. C., & Loomis, J. M. (2003). Interpersonal distance in immersive virtual environments. Personality and Social Psychology Bulletin, 29(7), 819–833. doi:10.1177/0146167203253270

Banks, J. (2018). A perceived moral agency scale: development and validation of a metric for humans and social machines. Computers in Human Behavior, 90, 363–371. doi.org/10.1016/j.chb.2018.08.028

Barnes, B. (2011, March 15). Many culprits in fall of a family film. New York Times. Retrieved from http://www.nytimes.com/2011/03/15/business/media/15mars.html

REFERENCES 122

Barrett, J. L., & Johnson, A. H. (2003). The role of control in attributing intentional agency to inanimate objects. Journal of Cognition and Culture, 3(3), 208–217. doi:10.1163/156853703322336634

Bartneck, C., Kanda, T., Ishiguro, H., & Hagita, N. (2009). My robotic doppelgänger – A critical look at the uncanny valley theory. In Proceedings of the 18th IEEE International Symposium on Robot and Human Interactive Communication, Toyama, Japan (pp. 269–276). doi:10.1109/ROMAN.2009.5326351

Bartneck, C., Verbunt, M., Mubin, O., & Al Mahmud, A. (2007). To kill a mockingbird robot. In Proceedings of the 2007 ACM/IEEE International Conference on Human- Robot Interaction, Arlington, USA (pp. 81–87). doi:10.1145/1228716.1228728

Bastian, B., Loughnan, S., Haslam, N., & Radke, H. R. (2011). Don’t mind meat? The denial of mind to animals used for human consumption. Personality and Social Psychology Bulletin, 38(2), 247–256. doi:10.1177/0146167211424291

Benner, K., & Wingfield, N. (2016, January 29). Apple sets its sights on virtual reality. New York Times. Retrieved from http://www.nytimes.com/2016/01/30/technology/apple- sets-its-sights-on-virtual-reality.html

Bennett, C. C., & Šabanović, S. (2015). The effects of culture and context on perceptions of robotic facial expressions. Interaction Studies, 16(2), 272–302. doi: 10.1075/is.16.2.11ben

Bensafi, M., Rouby, C., Farget, V., Bertrand, B., Vigouroux, M., & Holley, A. (2002). Influence of affective and cognitive judgments on autonomic parameters during inhalation of pleasant and unpleasant odors in humans. Neuroscience Letters, 319(3), 162–166. doi:10.1016/S0304-3940(01)02572-1

Biocca, F. (1997). The cyborg’s dilemma: Embodiment in virtual environments. Journal of Computer-Mediated Communication, 3(2). doi:10.1111/j.1083-6101.1997.tb00070.x

Boer, D. P. (2016). The Wiley handbook on the theories, assessment and treatment of sexual offending. Hoboken: John Wiley & Sons.

REFERENCES 123

Boiger, M., Mesquita, B., Uchida, Y., & Barrett, L. F. (2013). Condoned or condemned: the situational affordance of anger and shame in Japan and the US. Personality and Social Psychology Bulletin, 39(4), 540–553. doi:10.1177/0146167213478201

Borody, W. A. (2013). The Japanese roboticist Masahiro Mori's Buddhist inspired concept of “The Uncanny Valley”. Journal of Evolution and Technology, 23(1), 31–44.

Broadbent, E., Kumar, V., Li, X., Sollers, J., Stafford, R. Q., MacDonald, B. A., & Wegner, D. M. (2013). Robots with display screens: A robot with a more humanlike face display is perceived to have more mind and a better personality. PLOS One, 8, e72589. doi:10.1371/journal.pone.0072589

Brody, L. R. (1997). Gender and emotion: Beyond stereotypes. Journal of Social Issues, 53(2), 369–394. doi:10.1111/j.1540-4560.1997.tb02448.x

Brosschot, J. F., & Thayer, J. F. (2003). Heart rate response is longer after negative emotions than after positive emotions. International Journal of Psychophysiology, 50, 181–187. doi:10.1016/S0167-8760(03)00146-6

Buckingham, D. (2013). Is there a digital generation? In D. Buckingham & R. Willett (Eds.), Digital generations: Children, young people, and new media (pp. 1–18). New York: Routledge.

Burger, J. M., & Cooper, H. M. (1979). The desirability of control. Motivation and Emotion, 3(4), 381–393. doi:10.1007/BF00994052

Burke, S. (2017, October 12). Google admits its new smart speaker was eavesdropping on users. CNN. Retrieved from http://money.cnn.com/2017/10/11/technology/google- home-mini-security-flaw/

Butler, E. A., Egloff, B., Wilhelm, F. H., Smith, N. C., Erickson, E. A., & Gross, J. J. (2003). The social consequences of expressive suppression. Emotion, 3(1), 48–67. doi:10.1037/1528-3542.3.1.48

REFERENCES 124

Butler, E. A., Lee, T. L., & Gross, J. J. (2009). Does expressing your emotions raise or lower your blood pressure? The answer depends on cultural context. Journal of Cross- Cultural Psychology, 40, 510–517. doi:10.1177/0022022109332845

Call, J., & Tomasello, M. (2008). Does the chimpanzee have a theory of mind? Thirty years later. Trends in Cognitive Sciences, 12(5), 187–192. doi:10.1016/j.tics.2008.02.010

Camras, L. (1985). Socialization of affect communication. In M. Lewis & C. Saarni (Eds.), The socialization of emotions (pp. 141–160). New York: Plenum Press.

Carnegie Mellon University Graphics Lab (2015). CMU Graphics Lab Motion Capture . Retrieved from http://mocap.cs.cmu.edu

Carroll, D., Harrison, L. K., Johnston, D. W., Ford, G., Hunt, K, Der, G., & West, P. (2010). Cardiovascular reactions to psychological stress: the influence of demographic variables. Journal of Epidemiology and Community Health, 54, 876–877. doi:10.1136/jech.54.11.876

Cartmill, M. (1990). Human uniqueness and theoretical content in paleoanthropology. International Journal of Primatology, 11(3), 173–192. doi:10.1007/BF02192868

Cesario, J. (2014). Priming, replication, and the hardest science. Perspectives on Psychological Science, 9(1), 40–48. doi:10.1177/1745691613513470

Cheetham, M., Pavlovic, I., Jordan, N., Suter, P., & Jancke, L. (2013). Category processing and the human likeness dimension of the uncanny valley hypothesis: eye-tracking data. Frontiers in Psychology, 4, 108. doi:10.3389/fpsyg.2013.00108

Chekroud, A. M., Everett, J. A., Bridge, H., & Hewstone, M. (2014). A review of neuroimaging studies of race-related prejudice: Does amygdala response reflect threat? Frontiers in Human Neuroscience, 8. doi:10.3389/fnhum.2014.00179

Chen, F. F. (2007). Sensitivity of Goodness of Fit indexes to lack of measurement invariance. Structural Equation Modeling: A Multidisciplinary Journal, 14(3), 464–504. doi:10.1080/10705510701301834

REFERENCES 125

Chen, M., Zhang, Y., Li, Y., Mao, S., & Leung, V. C. M. (2015). EMC: Emotion-aware mobile cloud computing in 5G. IEEE Network, 29(2), 32–38. doi:10.1109/MNET.2015.7064900

Chen, X., Chen, H., Li, D., & Wang, L. (2009). Early childhood behavioral inhibition and social and school adjustment in Chinese children: A 5-year longitudinal study. Child Development, 80(6), 1692–1704. doi:10.1111/j.1467-8624.2009.01362.x

Cohen, H., & Lefebvre, C. (2017). Handbook of categorization in cognitive science. Cambridge: Elsevier.

Cohen, J. (1988). Statistical power analysis for the behavioral sciences. Hillsdale: Erlbaum.

Connolly, A. C., Sha, L., Guntupalli, J. S., Oosterhof, N., Halchenko, Y. O., Nastase, S. A., ... Haxby, J. V. (2016). How the human brain represents perceived dangerousness or “predacity” of animals. The Journal of Neuroscience, 36(19), 5373–5384. doi:10.1523/jneurosci.3395-15.2016

Creed, C., Beale, R., & Cowan, B. (2014). The impact of an embodied agent's emotional expressions over multiple interactions. Interacting with Computers, 27(2), 172–188. doi:10.1093/iwc/iwt064

Davis, E., Greenberger, E., Charles, S., Chen, C., Zhao, L., & Dong, Q. (2012). Emotion experience and regulation in China and the United States: How do culture and gender shape emotion responding? International Journal of Psychology, 47(3), 230–239. doi:10.1080/00207594.2011.626043

Derntl, B., Habel, U., Robinson, S., Windischberger, C., Kryspin-Exner, I., Gur, R. C., & Moser, E. (2012). Culture but not gender modulates amygdala activities during explicit emotion recognition. BMC Neuroscience, 13(1), 54. doi:10.1186/1471-2202- 13-54

Descartes, R., & Cottingham, J. (1986). Meditations on first philosophy: With selections from the Objections and Replies. Cambridgeshire: Cambridge University Press.

REFERENCES 126

Descola, P. (1994). The society of nature: A native ecology in Amazonia. New York: Cambridge University Press.

Dickerson, S. S., & Kemeny, M. E. (2004). Acute stressors and cortisol responses: A theoretical integration and synthesis of laboratory research. Psychological Bulletin, 130(3), 355–391. doi:10.1037/0033-2909.130.3.355

Doerrfeld, B. (2015, December 31). 20+ emotion recognition APIs that will leave you impressed, and concerned [Blog post]. Retrieved from http://nordicapis.com/20- emotion-recognition-apis-that-will-leave-you-impressed-and-concerned/

Dong, P., Huang, X. I., & Wyer, R. S. (2013). The illusion of saving face: How people symbolically cope with embarrassment. Psychological Science, 24(10), 2005–2012. doi:10.1177/0956797613482946

Dzieza, J. (2014, August 5). Why robots are getting cuter: Rise of the adorable machines. The Verge. Retrieved from https://www.theverge.com/2014/8/5/5970779/rise-of-the- adorable-machines

Eid, M., & Diener, E. (2001). Norms for experiencing emotions in different cultures: Inter- and intranational differences. Journal of Personality and Social Psychology, 81(5), 869–885. doi:10.1037//0022-3514.81.5.869

Ekman, P., & Friesen, W. V. (1969). The repertoire of nonverbal behavior: Categories, origins, usage, and coding. Semiotica, 1, 49–98. doi: 10.1515/semi.1969.1.1.49

Ekman, P., & Friesen, W. V. (1978). Facial Action Coding System. Palo Alto: Consulting Psychologists Press.

Elfenbein, H. A., & Ambady, N. (2002). On the universality and cultural specifity of emotion recognition: A meta-analysis. Psychological Bulletin, 128(2), 203–235.

Epley, N., Waytz, A., Akalis, S., & Cacioppo, J. T. (2008). When we need a human: Motivational determinants of anthropomorphism. Social Cognition, 26(2), 143–155. doi:10.1037/0033-2909.128.2.203

REFERENCES 127

Epley, N., Waytz, A., & Cacioppo, J. T. (2007). On seeing human: a three-factor theory of anthropomorphism. Psychological Review, 114, 864–886. doi:10.1037/0033- 295X.114.4.864

Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods, 39, 175–191. doi:10.3758/BF03193146

Ferrari, F., Paladino, M. P., & Jetten, J. (2016). Blurring human-machine distinctions: Anthropomorphic appearance in social robots as a threat to human distinctiveness. International Journal of Social Robotics, 8(2), 287–302. doi:10.1007/s12369-016- 0338-y

Fessler, D. M., Holbrook, C., & Snyder, J. K. (2012). Weapons make the man (larger): Formidability is represented as size and strength in humans. PLOS One, 7(4): e32751. doi:10.1371/journal.pone.0032751

Ford, B. Q., & Mauss, I. B. (2015). Culture and emotion regulation. Current Opinion in Psychology, 3, 1–5. doi:10.1016/j.copsyc.2014.12.004

Fossataro, C., Sambo, C. F., Garbarini, F., & Iannetti, G. D. (2016). Interpersonal interactions and empathy modulate perception of threat and defensive responses. Scientific Reports, 6, 19353. doi:10.1038/srep19353

Fox, J., Ahn, S. J., Janssen, J., Yeykelis, L., Segovia, K., & Bailenson, J. N. (2014). Avatars versus agents: A meta-analysis quantifying the effect of agency. Human-Computer Interaction, 30(5), 401–432. doi:10.1080/07370024.2014.921494

Franck, R., & Iannaccone, L. R. (2014). Religious decline in the 20th century West: Testing alternative explanations. Public Choice, 159(3–4), 385–414. doi:10.1007/s11127-013- 0103-9

Fraune, M. R., Nishiwaki, Y., Šabanović, S., Smith, E. R., & Okada, M. (2017). Threatening flocks and mindful snowflakes: How group entitativity affects perceptions of robots.

REFERENCES 128

In Proceedings of the 2017 ACM/IEEE International Conference on Human-Robot Interaction, Vienna, Austria (pp. 205–213). doi:10.1145/2909824.3020248

Frijda, N. H., & Mesquita, B. (1994). The social roles and functions of emotions. In H. R. Markus & S. Kitayama (Eds.), Emotion and culture (pp.51–87). New York: American Psychological Association.

Fukuda, H., & Ueda, K. (2010). Interaction with a moving object affects one's perception of its animacy. International Journal of Social Robotics, 2(2), 187–193. doi:10.1007/s12369-010-0045-z

Fuller, M. (2014). The concept of the soul: Some scientific and religious perspectives. In M. Fuller (Ed.), The concept of the soul: Scientific and religious perspectives (pp. 1–4). Newcastle upon Tyne: Cambridge Scholars Publishing.

Gallagher, H. L., & Frith, C. D. (2003). Functional imaging of ‘theory of mind’. Trends in Cognitive Sciences, 7(2), 77–83. doi:10.1016/S1364-6613(02)00025-6

Gallagher, S., & Zahavi, D. (2008). The phenomenological mind: An introduction to and cognitive science. New York: Routledge.

Gao T., & Scholl B. J. (2011). Chasing vs. stalking: Interrupting the perception of animacy. Journal of Experimental Psychology: Human Perception & Performance, 37, 669– 684. doi:10.1037/a0020735.

Gee, F. C., Browne, W. N., & Kawamura, K. (2005). Uncanny valley revisited. In Proceedings of the 14th IEEE International Workshop on Robot and Human Interactive Communication, Nashville, USA (pp. 151–157). doi:10.1109/ROMAN.2005.1513772

German Psychological Society (2016). Berufsethische Richtlinien des BDP und der DGPs. Retrieved from https://www.dgps.de/index.php?id=85

Goethe, J. W. (1798). Der Zauberlehrling [The Sorcerer’s Apprentice]. In F. Schiller (Ed.), Musen-Almanach für das Jahr 1798 [Muses’ Almanac for the year 1798] (pp. 32–37). Tübingen: J. G. Cotta.

REFERENCES 129

Gray, A. J. (2010). Whatever happened to the soul? Some theological implications of neuroscience. Mental Health, Religion & Culture, 13(6), 637–648. doi:10.1080/13674676.2010.488424

Gray, H. M., Gray, K., & Wegner, D. M. (2007). Dimensions of mind perception. Science, 315, 619. doi:10.1126/science.1134475

Gray, K., & Schein, C. (2012). Two minds vs. two philosophies: Mind perception defines morality and dissolves the debate between deontology and utilitarianism. Review of Philosophy and Psychology, 3(3), 405–423. doi:10.1007/s13164-012-0112-5

Gray, K., & Wegner, D. M. (2012). Feeling robots and human zombies: Mind perception and the uncanny valley. Cognition, 125, 125–130. doi:10.1016/j.cognition.2012.06.007

Green, R. D., MacDorman, K. F., Ho, C. -C., & Vasudevan, S. K. (2008). Sensitivity to the proportions of faces that vary in human likeness. Computers in Human Behavior, 24(5), 2456–2474. doi:10.1016/j.chb.2008.02.019

Grinbaum, A. (2015). Uncanny valley explained by Girard’s theory. IEEE Robotics & Automation Magazine, 22(1), 152–150. doi:10.1109/MRA.2014.2385568

Gross, J. J., & Levenson, R. W. (1993). Emotional suppression: Physiology, self-report, and expressive behavior. Journal of Personality and Social Psychology, 64, 970–986. doi:10.1037/0022-3514.64.6.970

Guadagno, R. E., Blascovich, J., Bailenson, J. N., & McCall, C. (2007). Virtual humans and persuasion: The effects of agency and behavioral realism. Media Psychology, 10, 1– 22.

Guan, H. (2007). Anpassung und Integration der chinesischen Studierenden in Deutschland (Doctoral dissertation). Retrieved from http://webdoc.sub.gwdg.de/ebook/dissts/Bremen/Guan2007.pdf

Gunes, H., & Hung, H. (2016). Is automatic facial expression recognition of emotions coming to a dead end? The rise of the new kids on the block. Image and Vision Computing, 55, 6–8. doi:10.1016/j.imavis.2016.03.013

REFERENCES 130

Hanson, D. (2006). Exploring the aesthetic range for humanoid robots. In Proceedings of the ICCS/CogSci-2006 Long Symposium: Toward Social Mechanisms of Android Science, Vancouver, Canada (pp. 16–20).

Hare, B., Call, J., Agnetta, B., & Tomasello, M. (2000). Chimpanzees know what conspecifics do and do not see. Animal Behaviour, 59(4), 771–785. doi:10.1006/anbe.1999.1377

Haring, K. S., Mougenot, C., Ono, F., & Watanabe, K. (2014). Cultural differences in perception and attitude towards robots. International Journal of Affective Engineering, 13(3), 149–157. doi:10.5057/ijae.13.149

Hartholt, A., Traum, D., Marsella, S. C., Shapiro, A., Stratou, G., Leuski, A., ... Gratch, J. (2013). All together now: Introducing the Virtual Human Toolkit. In R. Aylett, B. Krenn, C. Pelachaud, & H. Shimodaira (Eds.), Proceedings of the 13th International Conference on Intelligent Virtual Agents, Edinburgh, UK (pp. 368–361).

Haslam, N. (2006). Dehumanization: An integrative review. Personality and Social Psychology Review, 10(3), 252–264. doi:10.1207/s15327957pspr1003/4

Haslam, N., Loughnan, S., Kashima, Y., & Bain, P. (2009). Attributing and denying humanness to others. European Review of Social Psychology, 19, 55–85. doi:10.1080/10463280801981645

Hassabis, D., Kumaran, D., Summerfield, C., & Botvinick, M. (2017). Neuroscience-Inspired Artificial Intelligence. Neuron, 95(2), 245-258. doi:10.1016/j.neuron.2017.06.011

Hauer, T. (2018). Society and the Second Age of machines: Algorithms versus ethics. Society, 1–7. doi:10.1007/s12115-018-0221-6

Heyes, C. (2015). Animal mindreading: What's the problem? Psychonomic Bulletin & Review, 22(2), 313–327. doi:10.3758/s13423-014-0704-4

Ho, C.-C., & MacDorman, K. F. (2010). Revisiting the uncanny valley theory: Developing and validating an alternative to the Godspeed indices. Computers in Human Behavior, 26, 1508–1518. doi:10.1016/j.chb.2010.05.015

REFERENCES 131

Ho, C.-C., MacDorman, K. F., & Pramono, Z. A. D. (2008). Human emotion and the uncanny valley: A GLM, MDS, and ISOMAP analysis of robot video ratings. In Proceedings of the 3rd ACM/IEEE International Conference on Human-Robot Interaction, Amsterdam, Netherlands (pp. 169–176). doi:10.1145/1349822.1349845

Ho, D. Y. F. (1976). On the concept of face. American Journal of Sociology, 81(4), 867–884. doi:10.1086/226145

Ho, D. Y. F. (1986). Chinese patterns of socialization: A critical review. In M. H. Bond (Ed.), The psychology of the Chinese people (pp. 1–37). Hong Kong: Oxford University Press.

Ho, D. Y. F., Fu, W., & Ng, S. M. (2004). Guilt, shame and embarrassment: Revelations of face and self. Culture & Psychology, 10(1), 64–84. doi:10.1177/1354067X04044166

Hofstede, G. (1980). Culture's consequences: International differences in work related values. Beverly Hills: Sage.

Hofstede, G. (2001). Culture's consequences: Comparing values, behaviors, institutions and organizations across nations. Thousand Oaks: Sage Publications.

Hong, H., Yamins, D. L., Majaj, N. J., & DiCarlo, J. J. (2016). Explicit information for category-orthogonal object properties increases along the ventral stream. Nature Neuroscience, 19(4), 613–622. doi:10.1038/nn.4247

Horowitz, M. J., Duff, D. F., & Stratton, L. O. (1964). Body-buffer zone. Exploration of personal space. Archives of General Psychiatry, 11(6), 651–656. doi:10.1001/archpsyc.1964.01720300081010

Horton, H. (2016, March 24). Microsoft deletes ‘teen girl’ AI after it became a Hitler-loving sex robot within 24 hours. The Telegraph. Retrieved from http://www.telegraph.co.uk/technology/2016/03/24/microsofts-teen-girl-ai-turns-into- a-hitler-loving-sex-robot-wit/

REFERENCES 132

Hwang H., & Matsumoto D. (2015). Evidence for the universality of facial expressions of emotion. In M. Mandal & A. Awasthi (Eds.), Understanding facial expressions in communication (pp. 41–56). New Delhi: Springer.

Iacoboni, M. (2009). Imitation, empathy, and mirror neurons. Annual Review of Psychology, 60, 653–670. doi:10.1146/annurev.psych.60.110707.163604

Jentsch, E. (1906). Zur Psychologie des Unheimlichen [On the psychology of the uncanny]. Psychiatrisch-neurologische Wochenschrift, 22, 195–205.

Kalra, N. & Groves, D. G. (2017). The enemy of good: Estimating the cost of waiting for nearly perfect automated vehicles (Report No. RR2150). Santa Monica: RAND Corporation.

Kang, M. (2009). The ambivalent power of the robot. Antennae, 1(9), 47–58.

Kaplan, F. (2004). Who is afraid of the humanoid? Investigating cultural differences in the acceptance of robots. International Journal of Humanoid Robotics, 1(3), 465–480. doi:10.1142/S0219843604000289

Katagiri, Y., Nass, C., & Takeuchi, Y. (2001). Cross-cultural studies of the computers are social actors paradigm: The case of reciprocity. In M. J. Smith, G. Salvendy, D. Harris, & R. Koubek (Eds.), Usability evaluation and interface design: Cognitive engineering, intelligent agents, and virtual reality (pp. 1558–1562). Mahwah: Lawrence Erlbaum Associates.

Kazuhiko, K. (2017). An introduction to yōkai culture: Monsters, ghosts, and outsiders in Japanese history. Tokyo: Japan Publishing Industry Foundation for Culture.

Kennedy, R. S., Lane, N. E., Berbaum, K. S. & Lilienthal, M. G. (1993). Simulator Sickness Questionnaire: An enhanced method for quantifying simulator sickness. The International Journal of Aviation Psychology, 3(3), 203–220. doi:10.1207/s15327108ijap0303_3

REFERENCES 133

Kim, H. S., & Sherman, D. K. (2007). “Express yourself”: Culture and the effect of self- expression on choice. Journal of Personality and Social Psychology, 92(1), 1–11. doi:10.1037/0022-3514.92.1.1

Kim, S., & McGill, A. L. (2011). Gaming with Mr. Slot or gaming the slot machine? Power, anthropomorphism, and risk perception. Journal of Consumer Research, 38(1), 94– 107. doi:10.1086/658148

King, A. Y. C., & Bond, M. H. (1985). The Confucian paradigm of man: A sociological view. In W. S. Tseng & D. Y. H. Wu (Eds.), Chinese culture and mental health (pp. 29–45). San Diego: Academic Press.

Kitano, N. (2007). Animism, Rinri, modernization; the base of Japanese robotics. Paper presented at the 2007 IEEE International Conference on Robotics and Automation (ICRA’07), Rome, Italy. Retrieved from http://www.roboethics.org/icra2007/

Kitayama, S., Markus, H. R., & Kurokawa, M. (2000). Culture, emotion, and well-being: Good feelings in Japan and the United States. Cognition & Emotion, 14, 93–124. doi:10.1080/026999300379003

Kitayama, S., Mesquita, B., & Karasawa, M. (2006). Cultural affordances and emotional experience: Socially engaging and disengaging emotions in Japan and the United States. Journal of Personality and Social Psychology, 91(5), 890–903. doi:10.1037/0022-3514.91.5.890

Klahn, A. L., Klinkenberg, I. A., Notzon, S., Arolt, V., Pantev, C., Zwanzger, P., & Junghöfer, M. (2016). Prepare for scare—Impact of threat predictability on affective visual processing in spider phobia. Behavioural Brain Research, 307, 84–91. doi:10.1016/j.bbr.2016.03.045

Knobe, J., & Prinz, J. (2008). about consciousness: Experimental studies. Phenomenology and the Cognitive Sciences, 7, 67– 83. doi:10.1007/s11097-007- 9066-y

REFERENCES 134

Krämer, N. C., Lucas, G., Schmitt, L., & Gratch, J. (2018). Social snacking with a virtual agent—On the interrelation of need to belong and effects of social responsiveness when interacting with artificial entities. International Journal of Human-Computer Studies, 109, 112–121. doi:10.1016/j.ijhcs.2017.09.001

Kreibig, S. D. (2010). Autonomic nervous system activity in emotion: A review. Biological Psychology, 84, 394–421. doi:10.1016/j.biopsycho.2010.03.010

Kring, A. M., & Gordon, A. H. (1998). Sex differences in emotion: Expression, experience, and physiology. Journal of Personality and Social Psychology, 74, 686–703. doi:10.1037/0022–3514.74.3.686

Kwak, S. S., Kim, J. S., & Choi, J. J. (2017). The effects of organism- versus object-based robot design approaches on the consumer acceptance of domestic robots. International Journal of Social Robotics, 9(3), 359–377. doi:10.1007/s12369-016- 0388-1

Leary, M. R., Kelly, K. M., Cottrell, C. A., & Schreindorfer, L. S. (2013). Construct validity of the need to belong scale: Mapping the nomological network. Journal of Personality Assessment, 95(6), 610–624. doi:10.1080/00223891.2013.819511

LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521, 436–444. doi:10.1038/nature14539

Leotti, L. A., Iyengar, S. S., & Ochsner, K. N. (2010). Born to choose: The origin and value of the need for control. Trends in Cognitive Sciences, 14(10), 457–463. doi:10.1016/j.tics.2010.08.001

Levenson, R. W. (2003). Autonomic specificity and emotion. In R. J. Davidson, K. R. Scherer, & H. H. Goldsmith (Eds.), Handbook of affective sciences (pp. 212–224). New York: Oxford University Press.

Li, D., Rau, P. L. P., & Li, Y. (2010). A cross-cultural study: Effect of robot appearance and task. International Journal of Social Robotics, 2, 175–186. doi:10.1007/s12369-010- 0056-9

REFERENCES 135

Lim, S., & Reeves, B. (2010). Computer agents versus avatars: Responses to interactive game characters controlled by a computer or other player. International Journal of Human- Computer Studies, 68, 57–68. doi:10.1016/j.ijhcs.2009.09.008

Lim, Y. M., & Aylett, R. (2007). Feel the difference: A guide with attitude! In Proceedings of the 7th International Conference on Intelligent Virtual Agents, Paris, France (pp. 317–330). doi:10.1007/978-3-540-74997-4_29

Lin, P. (2016). Why ethics matters for autonomous cars. In M. Maurer, J. C. Gerdes, B. Lenz, & H. Winner (Eds.), Autonomous Driving (pp. 69–85). Heidelberg: Springer.

Litjens, G., Sánchez, C. I., Timofeeva, N., Hermsen, M., Nagtegaal, I., Kovacs, I., ... Laak, J. V. (2016). Deep learning as a tool for increased accuracy and efficiency of histopathological diagnosis. Scientific Reports, 6(1). doi:10.1038/srep26286

Liu, B. & Sundar, S. S. (2018). Should machines express sympathy and empathy? Experiments with a health advice chatbot. Cyberpsychology, Behavior, and Social Networking, 21(10). doi:10.1089/cyber.2018.0110

MacDonald, D. J., & Standing, L. G. (2002). Does self-serving bias cancel the Barnum effect? Social Behavior and Personality, 30(6), 625–630. https://doi.org/ doi:10.2224/sbp.2002.30.6.625

MacDorman, K. F., & Entezari, S. (2015). Individual differences predict sensitivity to the uncanny valley. Interaction Studies, 16(2), 141–172. doi:10.1075/is.16.2.01mac

MacDorman, K. F., & Ishiguro, H. (2006). The uncanny advantage of using androids in social and cognitive science research. Interaction Studies, 7(3), 297–337. doi:10.1075/is.7.3.03mac

MacDorman, K. F., Vasudevan, S. K., & Ho, C.-C. (2009). Does Japan really have robot mania? Comparing attitudes by implicit and explicit measures. AI & Society, 23(4), 485–510. doi:10.1007/s00146-008-0181-2

REFERENCES 136

Markus, H. R., & Kitayama, S. (1991). Culture and the self: Implications for cognition, emotion, and motivation. Psychological Review, 98(2), 224–253. doi:10.1037/0033- 295X.98.2.224

Martin, B., & Hanington, B. (2012). Universal methods of design (pp. 204–205). Beverly: Rockport Publishers.

Matsumoto, D. (2006). Are cultural differences in emotion regulation mediated by personality traits? Journal of Cross-Cultural Psychology, 37(4), 421–437. doi:10.1177/0022022106288478

Matsumoto, D., & Ekman, P. (1989). American-Japanese cultural differences in intensity ratings of facial expressions of emotion. Motivation and Emotion, 13(2), 143–157. doi:10.1007/BF00992959

Matsumoto, D., LeRoux, J. A., Bernhard, R., & Gray, H. (2004). Personality and behavioral correlates of intercultural adjustment potential. International Journal of Intercultural Relations, 28, 281–309.

Matsumoto, D., Willingham, B., & Olide, A. (2009). Sequential dynamics of culturally moderated facial expressions of emotion. Psychological Science, 20(10), 1269–1274. doi:10.1111/j.1467-9280.2009.02438.x

Matsumoto, D., Yoo, S. H., Fontaine, J., Anguas-Wong, A. M., Arriola, M., Ataca, B., ... Grossi, E. (2008a). Mapping expressive differences around the world: The relationship between emotional display rules and Individualism v. collectivism. Journal of Cross-Cultural Psychology, 39(1), 55–74. doi:10.1177/0022022107311854

Matsumoto, D., Yoo, S. H., & Nakagawa, S. (2008b). Culture, emotion regulation, and adjustment. Journal of Personality and Social Psychology, 94(6), 925–937. doi:10.1037/0022-3514.94.6.925

McCroskey, J. C. (1982). An introduction to rhetorical communication. Englewood Cliffs: Prentice-Hall.

REFERENCES 137

McCulloch, W. S., & Pitts, W. (1943). A logical calculus of the ideas immanent in nervous activity. The Bulletin of Mathematical Biophysics, 5(4), 115–133. doi:10.1007/bf02478259

McMahan, J. (2002). The ethics of killing: Problems at the margins of life. New York: Oxford University Press.

Meehl, P. (1956). Wanted—a good cookbook. American Psychologist, 11(6), 263–272. doi:10.1037/h0044164

Mendes, W. B. (2016). Emotion and the autonomic nervous system. In L. F. Barrett, M. Lewis, & J. Haviland-Jones (Eds.), Handbook of emotions (pp. 166–181). New York: Guilford Publications.

Michaud, F., Pirjanian, P., Audet, J., & Létourneau, D. (2000) Artificial emotion and social robotics. In L. E. Parker, G. Bekey, & J. Barhen (Eds.), Distributed Autonomous Robotic Systems 4 (pp. 121–130). Tokyo: Springer.

Michotte, A. (1963). The perception of . New York: Basic Books.

Mitchell, R. W., Thompson, N. S., & Miles, H. L. (1997). Anthropomorphism, anecdotes, and animals. Albany: State University of New York Press.

Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., ... Hassabis, D. (2015). Human-level control through deep . Nature, 518(7540), 529–533. doi:10.1038/nature14236

Moore, R. K. (2012). A Bayesian explanation of the ‘uncanny valley’ effect and related psychological phenomena. Scientific Reports, 2(864), 1–5. doi:10.1038/srep00864

Moosa, M. M., & Ud-Dean, M. (2010). Danger avoidance: An evolutionary explanation of uncanny valley. Biological Theory, 5, 12–14. doi:10.1162/BIOT_a_00016

Moran, C. M., Diefendorff, J. M., & Greguras, G. J. (2013). Understanding emotional display rules at work and outside of work: The effects of country and gender. Motivation and Emotion, 37, 323–334. doi:10.1007/s11031-012-9301-x

REFERENCES 138

Morewedge, C. K., Preston, J., & Wegner, D. (2007). Timescale bias in the attribution of mind. Journal of Personality and Social Psychology, 93(1), 1–11. doi:10.1037/0022- 3514.93.1.1

Mori, M. (1970). The uncanny valley. Energy, 7(4), 33–35.

Murata, A., Moser, J. S., & Kitayama, S. (2013). Culture shapes electrocortical responses during emotion suppression. Social Cognitive and Affective Neuroscience, 8(5), 595– 601. doi:10.1093/scan/nss036

Nass, C. I., Lombard, M., Henriksen, L., & Steuer, J. (1995). Anthropocentrism and computers. Behaviour & Information Technology, 14(4), 229–238. doi:10.1080/01449299508914636

Nass, C. & Moon, Y. (2000). Machines and mindlessness: Social responses to computers. Journal of Social Issues, 56(1), 81–103. doi:10.1111/0022-4537.00153

Ngai, E. W. T., Tao, S. S. C., & Moon, K. K. L. (2015). Social media research: Theories, constructs, and conceptual frameworks. International Journal of Information Management, 35, 33–44. doi:10.1016/j.ijinfomgt.2014.09.004

Nomura, T., Syrdal, D. S., & Dautenhahn, K. (2015). Differences on social acceptance of humanoid robots between Japan and the UK. In M. Salem, P. Baxter, & K. Dautenhahn (Eds.), Proceedings of the 4th International Symposium on New Frontiers in Human-Robot Interaction, Canterbury, UK (pp. 115–120).

Nordhoff, S., de Winter, J., Kyriakidis, M., van Arem, B., & Happee, R. (2018). Acceptance of driverless vehicles: results from a large cross-national questionnaire study. Journal of Advanced Transportation, 1–22. doi:10.1155/2018/5382192

Ojalehto, B. L., & Medin, D. L. (2015). Perspectives on culture and concepts. Annual Review of Psychology, 66, 249–275. doi:10.1146/annurev-psych-010814-015120

REFERENCES 139

Oker, A., Prigent, E., Courgeon, M., Eyharabide, V., Urbach, M., Bazin, N. ..., Brunet-Gouet, E. (2015). How and why affective and reactive virtual agents will bring new insights on social cognitive disorders in schizophrenia? An illustration with a virtual card game paradigm. Frontiers in Human Neuroscience, 9, 133. doi:10.3389/fnhum.2015.00133

Ottaviani, C., Shapiro, D., Davydov, D. M., Goldstein, I. B., & Mills, P. J. (2009). The autonomic phenotype of rumination. International Journal of Psychophysiology, 72(3), 267–275. doi:10.1016/j.ijpsycho.2008.12.014

Oyserman, D., Coon, H. M., & Kemmelmeier, M. (2002). Rethinking individualism and collectivism: Evaluation of theoretical assumptions and meta-analyses. Psychological Bulletin, 128(1), 3–72. doi:10.1037/0033-2909.128.1.3

Pagel, M. (2012). Evolution: Adapted to culture. Nature, 482, 297–299. doi:10.1038/482297a

Paine, C. (Director). (2018, April 06). Do You Trust This Computer? [Video file]. Retrieved from http://doyoutrustthiscomputer.org/

Panksepp, J. (2011). The basic emotional circuits of mammalian brains: Do animals have affective lives? Neuroscience and Biobehavioral Reviews, 35, 1791–1804. doi:10.1016/j.neubiorev.2011.08.003

Parker, R., S., Haytko., D. L., & Hermans, C. M. (2009). Individualism and collectivism: Reconsidering old assumptions. Journal of International Business Research, 8(1), 127.

Patel, H., & MacDorman, K. F. (2015). Sending an avatar to do a human's job: Compliance with authority persists despite the uncanny valley. Presence, 24(1), 1–23. doi:10.1162/PRES_a_00212

Powers, K. E., Worsham, A. L., Freeman, J. B., Wheatley, T., & Heatherton, T. F. (2014). Social connection modulates perceptions of animacy. Psychology Science, 25(10), 1943–1948. doi:10.1177/0956797614547706

REFERENCES 140

Premack, D. & Woodruff, G. (1978). Does the chimpanzee have a theory of mind? Behavioral and Brain Sciences, 1(4), 515–526. doi:10.1017/S0140525X00076512

Ramey, C. H. (2005). The uncanny valley of similarities concerning abortion, baldness, heaps of sand, and humanlike robots. In Proceedings of Views of the Uncanny Valley Workshop: IEEE-RAS International Conference on Humanoid Robots, Tsubaka, Japan (pp. 8–13).

Rapee, R. M. (1997). Perceived threat and perceived control as predictors of the degree of fear in physical and social situations. Journal of Anxiety Disorders, 11(5), 445–461. doi:10.1016/S0887-6185(97)00022-4

Raven, J., Raven, J. C., & Court, J. H. (2003). Manual for Raven's Progressive Matrices and Vocabulary Scales. San Antonio: Harcourt Assessment.

Reeves, B., & Nass, C. I. (1996). The media equation: How people treat computers, television, and new media like real people and places. Stanford: CSLI Publications.

Roberts, N. A., Levenson, R. W., & Gross, J. J. (2008). Cardiovascular costs of emotion suppression cross ethnic lines. International Journal of Psychophysiology, 80, 82–87. doi:10.1016/j.ijpsycho.2008.06.003

Rosenblatt, F. (1958). The perceptron: a probabilistic model for information storage and organization in the brain. Psychological Review, 65(6), 386–408. doi:10.1037/h0042519

Rosenthal-von der Pütten, A. M., & Krämer, N. C. (2014). How design characteristics of robots determine evaluation and uncanny valley related responses. Computers in Human Behavior, 36, 422–439. doi: 10.1016/j.chb.2014.03.066

Roubroeks, M., Ham, J., & Midden, C. (2011). When artificial social agents try to persuade people: The role of social agency on the occurrence of psychological reactance. International Journal of Social Robotics, 3(2), 155–165. doi:10.1007/s12369-010- 0088-1

REFERENCES 141

Rustemli, A. (1988). The effects of personal space invasion on impressions and decisions. The Journal of Psychology, 122(2), 113–118. doi:10.1080/00223980.1988.9712696

Saenz, A. (2010, August 10). We live in a jungle of artificial intelligence that will spawn sentience. Singularity Hub. Retrieved from https://singularityhub.com/2010/08/10/we- live-in-a-jungle-of-artificial-intelligence-that-will-spawn-sentience/

Safdar, S., Friedlmeier, W., Matsumoto, D., Yoo, S. H., Kwantes, C. T., Kakai, H., & Shigemasu, E. (2009). Variations of emotional display rules within and across cultures: A comparison between Canada, USA, and Japan. Canadian Journal of Behavioural Science, 41(1), 1–10. doi:10.1037/a0014387

Saygin, A. P., Chaminade, T., Ishiguro, H., Driver, J., & Frith, C. (2012). The thing that should not be: predictive coding and the uncanny valley in perceiving human and humanoid robot actions. Social Cognitive and Affective Neuroscience, 7(4), 413–422. doi:10.1093/scan/nsr025

Schaller, M., & Park, J. H. (2011). The behavioral immune system (and why it matters). Current Directions in Psychological Science, 20(2), 99–103. doi:10.1177/0963721411402596

Searle, J. R. (1998). How to study consciousness scientifically. Philosophical Transactions of the Royal Society B: Biological Sciences, 353(1377), 1935–1942. doi:10.1098/rstb.1998.0346

Seo, Y. S. (2011). Individualism, collectivism, and client expression of different emotions: their relations to perceived counselor effectiveness. Asia Pacific Education Review, 12(2), 251–261. doi:10.1007/s12564-010-9136-7

Seyama, J., & Nagayama, R. S. (2007). The uncanny valley: Effect of realism on the impression of artificial human faces. Presence, 16(4), 337–351. doi:10.1162/pres.16.4.337

Shen, B.-J., Stroud, L. R., & Niaura, R. (2004). Ethnic differences in cardiovascular responses to laboratory stress: A comparison between Asian and White Americans.

REFERENCES 142

International Journal of Behavioral Medicine, 11(3), 181–186. doi:10.1207/s15327558ijbm1103_7

Shimada, M., Minato, T., Itakura, S., & Ishiguro, H. (2018). Uncanny valley of androids and the lateral inhibition hypothesis. In H. Ishiguro & F. Dalla Libera (Eds.), Geminoid Studies (pp. 137–155). Singapore: Springer.

Shumaker, R. W., & Swartz, B. (2002). When traditional methodologies fail: Cognitive studies of great apes. In M. Bekoff, C. Allen, & G. Burghardt (Eds.), The cognitive animal: Empirical and theoretical perspectives on animal cognition (pp. 335–343). Cambridge: MIT Press.

Sorokowska, A., Sorokowsi, P., Hilpert, P., Cantarero, K., Frackowiak, T., Ahmadi, K., ... Pierce, J. D. (2017). Preferred interpersonal distances: A global comparison. Journal of Cross-Cultural Psychology, 48(4), 557–592. doi:10.1177/0022022117698039

Soto, J. A., Perez, C. R., Kim, Y.-H., Lee, E. A., & Minnick, M. R. (2011). Is expressive suppression always associated with poorer psychological functioning? A cross- cultural comparison between European Americans and Hong Kong Chinese. Emotion, 11(6), 1450–1455. doi:10.1037/a002334

Spaulding, S. (2013). Mirror neurons and social cognition. Mind and Language, 28(2), 233– 257. doi:10.1111/mila.12017

Steck, A., & Steck, B. (2016). Brain and mind: Subjective experience and scientific . New York: Springer.

Stein, J.-P., Liebold, B., & Ohler, P. (2017). Between threat and control. Linking situational control and human distinctiveness concerns to virtual agents’ uncanniness. Paper presented at the 67th Annual Conference of the International Communication Association, May 25–29, San Diego, USA.

Stein, J.-P., Liebold, B., & Ohler, P. (2019). Stay back, clever thing! Linking situational control and human uniqueness concerns to the aversion against autonomous

REFERENCES 143

technology. Computers in Human Behavior, 95, 73–82. doi:10.1016/j.chb.2019.01.021

Stein, J.-P., & Ohler, P. (2017a). Venturing into the uncanny valley of mind—The influence of mind attribution on the acceptance of human-like characters in a virtual reality setting. Cognition, 160, 43–50. doi:10.1016/j.cognition.2016.12.010

Stein, J.-P., & Ohler, P. (2017b). Distal and proximal paths leading into the uncanny valley of mind. Talk held at the 10th Conference of the Media Psychology Division of the German Society for Psychology (DGPs), September 6-8, Landau, Germany.

Stein, J.-P., & Ohler, P. (2018a). Saving face in front of the computer? Culture and attributions of human likenes influence users’ experience of automatic facial emotion recognition. Frontiers in Digital Humanities, 7, 18. https://doi.org/10.3389/fdigh.2018.00018

Stein, J.-P., & Ohler, P. (2018b). Saving face in front of the computer? Culture influences cardiovascular reactions to facial emotion recognition software. Talk held at the 68th Annual Conference of the International Communication Association, May 24–28, Prague, Czech Republic.

Stevens, J. (1999). Intermediate statistics: A modern approach. London: Erlbaum.

Strube, M. J., & Werner, C. (1984). Personal space claims as a function of interpersonal threat: The mediating role of need for control. Journal of Nonverbal Behavior, 8(3), 195–209. doi:10.1007/BF00987291

Suddendorf, T. (2013). The gap: The science of what separates us from other animals. New York: Basic Books.

Sundar, S. S., Waddell, T. F., & Jung, E. H. (2016). The Hollywood robot syndrome: Media effects on older adults’ attitudes toward robots and adoption intentions. Paper presented at the 11th Annual ACM/IEEE International Conference on HumanRobot Interaction, Christchurch, New Zealand.

REFERENCES 144

Takahashi, K., Ohara, N., Antonucci, T. C., & Akiyama, H. (2002). Commonalities and differences in close relationships among the Americans and Japanese: A comparison by the individualism/collectivism concept. International Journal of Behavioral Development, 26(5), 453–465. doi:10.1080/01650250143000418

Tam, K.-P., Lee, S.-L., & Chao, M. M. (2013). Saving Mr. Nature: Anthropomorphism enhances connectedness to and protectiveness towards nature. Journal of Experimental Social Psychology, 49, 514–521. doi:10.1016/j.jesp.2013.02.001

Tinwell, A. (2014). Applying psychological plausibility to the uncanny valley phenomenon. In M. Grimshaw (Ed.), The Oxford handbook of virtuality (pp. 173–186). Oxford: University of Oxford Press.

Tinwell, A., Abdel Nabi, D., & Charlton, J. (2013). Perception of psychopathy and the uncanny valley in virtual characters. Computers in Human Behavior, 29, 1617–1625. doi:10.1016/j.chb.2013.01.008

Tinwell, A., & Grimshaw, M. (2009). Bridging the uncanny: An impossible traverse? In Proceedings of the 13th International MindTrek Conference: Everyday Life in the Ubiquitous Era, Tampere, Finland (pp. 66–73). doi:10.1145/1621841.1621855

Tinwell, A., & Sloan, R. J. S. (2014). Children's perception of uncanny human-like virtual characters. Computers in Human Behavior, 36, 286–296. doi:10.1016/j.chb.2014.03.073

Tomonaga, M., & Uwano, Y. (2010). Bottlenose dolphins’ (Tursiops truncatus) Theory of Mind as demonstrated by responses to their trainers’ attentional states. International Journal of Comparative Psychology, 23(3), 386–400.

Tremoulet, P. D., & Feldman, J. (2006). The influence of spatial context and the role of in the interpretation of animacy from motion. Perception & Psychophysics, 68, 1047–1058. doi:10.3758/BF03193364

REFERENCES 145

Triandis, H. C., & Gelfand, M. J. (1998). Converging measurement of horizontal and vertical individualism and collectivism. Journal of Personality and Social Psychology, 1, 118–128. doi:10.1037/0022-3514.74.1.118

Turkle, S. (1984). The second self: Computers and the human spirit. New York: Simon & Schuster.

Underwood, M. K., Coie, J. D., & Herbsman, C. R. (1992). Display rules for anger and aggression in school-age childen. Child Development, 63, 366–380.

Unity [Computer software]. (2015). Retrieved from https://unity3d.com/

Urgen, B. A., Kutas, M., & Saygin, A. P. (2018). Uncanny valley as a window into predictive processing in the social brain. Neuropsychologica, 114, 181–185. doi:10.1016/j.neuropsychologia.2018.04.027

Vaes, J., Paladino, M. P., Castelli, L., Leyens, J. P., & Giovanazzi, A. (2003). On the behavioral consequences of infra-humanization: The implicit role of uniquely human emotions in intergroup relations. Journal of Personality and Social Psychology, 85(6), 1016–1034. doi:10.1037/0022-3514.85.6.1016

Varela, F. J., Rosch, E., & Thompson, E. (2016). The embodied mind: Cognitive science and human experience. Cambridge: MIT Press.

Vess, M., Arndt, J., & Cox, C. R. (2012). Faith and nature: The effect of death-relevant cognitions on the relationship between religious fundamentalism and connectedness to nature. Social Psychological and Personality Science, 3(3), 333–340. doi:10.1177/1948550611420303

Vogeley, K., & Bente, G. (2010). “Artificial humans”: Psychology and neuroscience perspectives on embodiment and nonverbal communication. Neural Networks, 23, 1077–1090. doi:10.1016/j.neunet.2010.06.003 von der Pütten, A., Krämer, N. C., & Gratch, J. (2010). How our personalities shape our interactions with virtual characters. Implications for research and development. In J.

REFERENCES 146

Allbeck, N. Badler, T. Bickmore, C. Pelachaud, & A. Safonova (Eds.), Intelligent Virtual Agents (pp. 208–221). Berlin: Springer.

Wang, D., Khosla, A., Gargeya, R., Irshad, H., & Beck, A. H. (2016). Deep learning for identifying metastatic breast cancer. arXiv preprint arXiv:1606.05718.

Wang, H., Zhou, Z., Li, Y., Chen, Z., Lu, P., Wang, W., ... Yu, L. (2017). Comparison of machine learning methods for classifying mediastinal lymph node metastasis of non- small cell lung cancer from 18F-FDG PET/CT images. EJNMMI Research, 7(1). doi:10.1186/s13550-017-0260-9

Wang, S., Lilienfeld, S. O., & Rochat, P. (2015). The uncanny valley: Existence and explanations. Review of General Psychology, 19(4), 393–407. doi:0.1037/gpr0000056

Wang, S., & Rochat, P. (2017). Human perception of animacy in light of the uncanny valley phenomenon. Perception, 46(12), 1386-1411. doi:10.1177/0301006617722742

Wang, Y., Liu, X., & Cui, L. (2012). Influential factors of emotional display rules in Chinese adolescents. Psychology, 3(6), 504–506. doi:10.4236/psych.2012.36072

Wass, S., De Barbaro, K., & Clackson, K. (2015). Learning and the autonomic nervous system: Understanding interactions between stress, concentration and learning during early childhood. Frontiers in Neuroscience. doi:10.3389/conf.fnins.2016.92.00015

Waytz, A., Gray, K., Epley, N., & Wegner, D. M. (2010). Causes and consequences of mind perception. Trends in Cognitive Sciences, 14, 383–388. doi:10.1016/j.tics.2010.05.006

Waytz, A., Heafner, J., & Epley, N. (2014). The mind in the machine: Anthropomorphism increases trust in an autonomous vehicle. Journal of Experimental Social Psychology, 52, 113–117. doi:10.1016/j.jesp.2014.01.005

Waytz, A. & Norton, M. I. (2014). Botsourcing and outsourcing: Robot, British, Chinese, and German workers are for thinking – not feeling – jobs. Emotion, 14, 434–444. doi:10.1037/a0036054

REFERENCES 147

Wegner, D. M., & Gray, K. (2016). The mind club: Who thinks, what feels, and why it matters. New York: Viking.

Weibel, D., Wissmath, B., Habegger, S. Steiner, Y., & Groner, R. (2008). Playing online games against computer versus human controlled opponents: effects on presence, flow, and enjoyment. Computers in Human Behaviour, 24, 2274–2291. doi:10.1016/j.chb.2007.11.002

Yamada, Y., Kawabe, T., & Ihaya, K. (2013). Categorization difficulty is associated with negative evaluation in the “uncanny valley” phenomenon. Japanese Psychological Research, 55, 20–32. doi:10.1111/j.1468-5884.2012.00538.x

Yamakage, M. (2012). The essence of Shinto: Japan’s spiritual heart. Tokyo: Kodansha International.

Young, K. L., & Carpenter, C. (2018). Does science fiction affect political fact? Yes and no: A survey experiment on “Killer Robots”. International Studies Quarterly, 62(3), 562– 576. doi:10.1093/isq/sqy028

Young, L., & Waytz, A. (2013). Mind attribution is for morality. In S. Baron-Cohen, M. Lombardo, & H. Tager-Flusberg (Eds.), Understanding other minds (pp. 93–103). Oxford: Oxford University Press. doi:10.1093/acprof:oso/9780199692972.003.0006

Yu, W., & Wang, W. (2011). An investigation into the acculturation strategies of Chinese students in Germany. Intercultural Communication Studies, 20(2), 190–210.

Zhou, T. & Bishop, G. D. (2012). Culture moderates the cardiovascular consequences of anger regulation strategy. International Journal of Psychophysiology, 86, 291–298. doi:10.1016/j.ijpsycho.2012.10.010

Złotowski, J., Sumioka, H., Nishio, S., Glas, D. F., Bartneck, C., & Ishiguro, H. (2015). Persistence of the uncanny valley: The influence of repeated interactions and a robot’s attitude on its perception. Frontiers in Psychology, 6, 883. doi:10.3389/fpsyg.2015.00883

REFERENCES 148

Złotowski, J., Yogeeswaran, K., & Bartneck, C. (2017). Can we control it? Autonomous robots threaten human identity, uniqueness, safety, and resources. International Journal of Human-Computer Studies, 100, 48–54. doi:j.ijhcs.2016.12.008

Zuckerberg, M. (2014). Acquiring Oculus Rift. Retrieved from https://www.facebook.com/zuck/posts/10101319050523971

Zuckerman, M., Klorman, R., Larrance, D. T., & Spiegel, N. H. (1981). Facial, autonomic, and subjective components of emotion: The facial feedback hypothesis versus the externalizer–internalizer distinction. Journal of Personality and Social Psychology, 41, 929–944. doi:10.1037/0022-3514.41.5.929

CURRICULUM VITAE 149

JAN-PHILIPP|STEIN

* 15. Juni 1988 in Mannheim

[email protected] www.jpstein.de

AKADEMISCHER WERDEGANG

04|2015 — 04|2019 DFG-Graduiertenkolleg „CrossWorlds“, Technische Universität Chemnitz Angestrebte Promotion (Dr. rer. nat.) im Fach Psychologie

Wissenschaftliche Betreuung: Prof. Dr. Peter Ohler Dissertationsschrift: “Exploring the Uncanny Valley of Mind. An Investigation of Human Uniqueness Concepts, Control Perceptions, and Threat Experience in the Face of Emotion-Sensitive Artificial Intelligence”

10|2006 — 08|2012 Technische Universität Dresden Diplom im Fach Psychologie

Studienschwerpunkte: Klinische Psychologie, Arbeitspsychologie Diplomarbeit: „Selbstdarstellung auf Facebook: Einflüsse des Geschlechts und narzisstischer Tendenzen“ Bewertung: sehr gut

LEHRPRAXIS

Lehrveranstaltungen Technische Universität Chemnitz

Seminar „Methoden der quantitativen Sozialforschung“ (2015, 2017) Seminar „Wissenschaftliches Schreiben“ (2016, 2017) Seminar „Kognitive und emotionale Verarbeitung virtueller Welten“ (2017) Praxisprojekt „Forschungsvertiefung: Medienpsychologie“ (2016) Übung „Scripting für Kommunikationswissenschaftler“ (2017)

Julius-Maximilians-Universität Würzburg

Seminar „Psychologie der Online- und Mobilkommunikation I“ (2018) Seminar „Psychologie der Online- und Mobilkommunikation II“ (2018, 2019) Seminar „Wissenschaftliches Arbeiten“ (2018) Vorlesung „Grundlagen der Medienkommunikation: Sozialpsychologie“ (2019)

Betreute Neun Bachelor- und zwei Masterarbeiten

Abschlussarbeiten Studiengänge: Medienkommunikation, Medien- und Instruktionspsychologie

CURRICULUM VITAE 150

Themenfelder: Stereotype, Social Media und Körperwahrnehmung, Parasoziale Interaktion auf Youtube, Cyberbullying, Figurenanalyse im Animationsfilm BERUFLICHE ERFAHRUNGEN

seit 08|2018 Lehrstuhl für Kommunikationspsychologie und Neue Medien, Julius-Maximilians-Universität Würzburg Wissenschaftliche Mitarbeit

Forschungsschwerpunkte: Social Robotics, Social Media, Medien und Selbst

04|2015 — 07|2018 Professur Medienpsychologie, Technische Universität Chemnitz Wissenschaftliche Mitarbeit

Forschungsschwerpunkte: Virtuelle Realität, digitale Agenten, kulturvergleichende Psychologie, Social Media

2003 — 2015 freiberuflich Digitales Design

Gestaltung von Web- und Printmedien

02|2013 — 07|2013 Psychosoziale Beratungsstelle, Studentenwerk Dresden Psychologische Beratung

Durchführung von Einzelclearings und Beratungsreihen für Studierende

10|2009 — 03|2012 Institutsambulanz und Tagesklinik, TU Dresden Studentische Mitarbeit

Versuchsleitung in der Sozialphobie-Forschung Dreh und Post-Produktion filmischer Lehrsequenzen

04|2011 — 06|2011 Jugendstrafanstalt Berlin Praktikum

Psychologische Begleitung des Sozialtherapeutischen Strafvollzuges

11|2010 — 02|2011 Institut für Klinische Psychologie und Psychotherapie, TU Dresden Studentische Mitarbeit

Mitarbeit in der Studie „E@T - Intervention bei Anorexia Nervosa“ Rekrutierung von Versuchspersonen und Datenverarbeitung

CURRICULUM VITAE 151

AUSZEICHNUNGEN & ENGAGEMENT

Auszeichnungen Top Five Paper (Kategorie „Mass Communication”) des 69th Annual Congress of the International Communication Association 2019

Stipendium der Studienstiftung des deutschen Volkes 2006 – 2008

Deutscher Multimediapreis „MB21" 2005, 2006

Kommissionsarbeit Mitglied der Berufungskommission „Prädiktive Verhaltensanalyse“ (W3), Technische Universität Chemnitz 2018

Ad-Hoc-Reviews Cyberpsychology, Behavior, and Social Networking Journal for Articles in Support of the Null Hypothesis Telematics & Informatics

Annual Congress of the International Communication Association Tagung der DGPs-Fachgruppe „Medienpsychologie“

Mitgliedschaften International Communication Association (ICA) seit 2016

Deutsche Gesellschaft für Psychologie (DGPs) seit 2019

BERICHTERSTATTUNG

Online-Medien Science/AAAS (diskutiert Stein & Ohler, 2017) http://www.sciencemag.org/news/2017/03/beware-emotional-robots-giving- feelings-artificial-beings-could-backfire-study-suggests

DigitalTrends (diskutiert Stein & Ohler, 2017) http://www.digitaltrends.com/cool-tech/uncanny-valley-mind/

LIST OF PUBLICATIONS 152

PUBLIKATIONEN

Journalartikel Appel, M., Krisch, N., Stein, J.-P., & Weber, S. (2019). Smartphone zombies! (peer-reviewed) Pedestrians’ distracted walking as a function of their fear of missing out. Journal of Environmental Psychology.

Stein, J.-P., Liebold, B., & Ohler, P. (2019). Stay back, clever thing! Linking situational control and human uniqueness concerns to the aversion against autonomous technology. Computers in Human Behavior, 95, 73–82. https://doi.org/10.1016/j.chb.2019.01.021

Koban, K., Stein, J.-P., Eckhardt, V., & Ohler, P. (2018). Quid pro quo in Web 2.0. Connecting personality traits and Facebook usage intensity to uncivil commenting intentions in public online discussions. Computers in Human Behavior, 79, 9–18. https://doi.org/10.1016/j.chb.2017.10.015.

Stein, J.-P., Lu, X., & Ohler, P. (2018). Mutual perceptions of Chinese and German students at a German university: Stereotypes, media influence, and evidence for a negative contact hypothesis. Compare. https://doi.org/10.1080/03057925.2018.1477579

Stein, J.-P., & Ohler, P. (2018). Saving face in front of the computer? Culture and attributions of human likeness influence users’ experience of automatic facial emotion recognition. Frontiers in Digital Humanities, 7, 18. https://doi.org/10.3389/fdigh.2018.00018

Stein, J.-P., & Ohler, P. (2018). Uncanny…but convincing? Inconsistency between a virtual agent’s facial proportions and vocal realism reduces its credibility and attractiveness, but not its persuasive success. Interacting With Computers. https://doi.org/10.1093/iwc/iwy023

Stein, J.-P., & Ohler, P. (2017). Venturing into the uncanny valley of mind—The influence of mind attribution on the acceptance of human-like characters in a virtual reality setting. Cognition, 160, 43–50. https://doi.org/10.1016/j.cognition.2016.12.010

Buchkapitel Stein, J.-P. (2017). Cloud Strife. In J. Banks, R. Meija, & A. Adams (Eds.), 100 Greatest Characters (pp. 39–40). Lanham, USA: Rowman & Littlefield.

Konferenzbeiträge Stein, J.-P., & Krause, E. (2019). Every (Insta-)gram counts? An investigation of Instagram’s first-, second-, and third-order cultivation effects on users’ body image. Paper to be presented at the 69th Annual Conference of the International Communication Association, May 24–28, Washington, USA.

Stein, J.-P., Koban, K. & Joos, S. (2019). Worth the effort? Comparing viewers’ identification, parasocial interaction, immersion, and enjoyment of different YouTube vlog production styles and topics. Paper to be presented at the 69th Annual Conference of the International Communication Association, May 24–28, Washington, USA.

Stein, J.-P., Lu, X., & Ohler, P. (2018). Mutual perceptions of Chinese and German students at a German university: Stereotypes, media influence, and a worrisome twist on the contact hypothesis. Paper presented at the 68th Annual Conference of the International Communication Association, May 24–28, Prague, Czech Republic.

LIST OF PUBLICATIONS 153

Stein, J.-P., & Ohler, P. (2018). Mismatching the realism of a persuasive agent’s voice and face reduces its credibility and attractiveness, but not its persuasive success. Paper presented at the 68th Annual Conference of the International Communication Association, May 24–28, Prague, Czech Republic.

Stein, J.-P., & Ohler, P. (2018). Saving face in front of the computer? Culture influences cardiovascular reactions to facial emotion recognition software. Paper presented at the 68th Annual Conference of the International Communication Association, May 24–28, Prague, Czech Republic.

Etzold, B., & Stein, J.-P. (2017). Der User zwischen Anwendung und Exponat – Smart Devices und Tabletop im Museumskontext. Talk held at the 47th Annual Conference of the German Computer Science Society, September 25-29, Chemnitz.

Gawin, W., Stein, J.-P., Fries, U., Gaudel, J., Zwingmann, K., & Voelcker-Rehage, C. (2017). The effect of physiological arousal on the quiet eye of elite badminton players. Paper presented at the 22nd Annual Congress of the European College of Sport Science, July 5-8, MetropolisRuhr, Germany.

Stein, J.-P., & Ohler, P. (2017). Distal and proximal paths leading into the uncanny valley of mind. Paper presented at the 10th Conference of the Media Psychology Division of the German Society for Psychology (DGPs), September 6- 8, Landau.

Stein, J.-P., Liebold, B., & Ohler, P. (2017). Between threat and control. Linking situational control and human distinctiveness concerns to virtual agents’ uncanniness. Paper presented at the 67th Annual Conference of the International Communication Association, May 25–29, San Diego, USA.

Koban, K., Stein, J.-P., & Eckhardt, V. (2017). Haters gonna hate. Connecting personality traits and usage intensity to dysfunctional commenting in Facebook discussions. Paper presented at the 67th Annual Conference of the International Communication Association, May 25–29, San Diego, USA.

Stein, J.-P., Lu, X., & Ohler, P. (2017). Industrious East, practical West? Connecting stereotypes among German and Chinese students to their perceptions of native speaking media. Paper presented at the 59th Conference of Experimental Psychologists (TeaP), March 26-29, Dresden, Germany.

AFFIDAVIT 154

Selbstständigkeitserklärung (Affidavit)

Name, Vorname: Stein, Jan-Philipp geboren am: 15.06.1988 geboren in: Mannheim

Hiermit erkläre ich, dass ich die vorliegende Arbeit mit dem Titel

Exploring the Uncanny Valley of Mind. An Investigation of Human Uniqueness Concepts, Control Perceptions, and Threat Experience in the Face of Emotion-Sensitive Artificial Intelligence selbstständig angefertigt habe. Es wurden nur die in der Arbeit ausdrücklich benannten Quellen und Hilfsmittel benutzt.

Die vorliegende Arbeit ist frei von Plagiaten. Alle Ausführungen, die wörtlich oder inhaltlich aus anderen Schriften entnommen sind, habe ich als solche kenntlich gemacht. Ich stimme zu, dass diese Arbeit auf Plagiate überprüft werden darf.

Chemnitz, den 30.04.2018

______

(Jan-Philipp Stein)