Conceptualization and Validation of a Novel Protocol for Investigating the Uncanny Valley Megan K. Strait Abstract Tufts University Loosely based on principles of similarity-attraction, robots Medford, MA 02155, USA [email protected] intended for social contexts are being designed with in- creasing human similarity to facilitate their reception by and communication with human interactants. However, the ob- servation of an uncanny valley – the phenomenon in which certain humanlike entities provoke dislike instead of liking – has lead some to caution against this practice. Substantial evidence supports both of these contrasting perspectives on the design of social technologies. Yet, owing to both empirical and theoretical inconsistencies, the relationship between anthropomorphic design and people’s liking of the technology remains poorly understood. arXiv:1808.03977v1 [cs.HC] 12 Aug 2018 Here we present three studies which investigate people’s explicit ratings of and behavior towards a large sample of real-world robots. The results show a profound “valley effect” on people’s willingness to interact with humanlike robots, thus highlighting the formidable design challenge the uncanny valley poses for social robotics. In addition AUTHOR BIO: Megan Strait is a PhD candidate in the Cognitive Science Program at to advancing uncanny valley theory, Studies 2 and 3 con- Tufts University. She received her BA in computer science from Wellesley College in tribute and validate a novel laboratory task for objectively 2010 and her MSc from Tufts University in 2014. Her graduate research is centered in the areas of human-robot and human-computer interaction. She is especially measuring people’s perceptions of humanlike robots. interested in the uncanny valley and its implications for the design and use of humanlike robots. Her dissertation (Understanding the Uncanny: The Effects of Human Similarity on Aversion towards Humanlike Robots) advances uncanny valley Author Keywords theory and introduces a novel protocol for evaluating human-robot interactions. Uncanny valley theory, social robotics; human-robot interac- tion; emotion; emotion regulation Introduction similarity-attraction theory, it is expected that such robots Loosely based on similarity-attraction theory [10], tech- offer more natural and effective interactions by capitalizing nologies intended for social contexts are being designed on traits which are more familiar and intuitive to people. with increasing human similarity to facilitate their recep- tion by and communication with people. This anthropomor- This perspective is not unfounded. In fact, it is rather well- phization of social technologies represents a particularly supported by a large empirical base. Humanlike robots are powerful mechanism for facilitating human-computer and perceived as more thoughtful (e.g., [4], intelligent (e.g., [7]), human-robot interactions. Specifically, increasing the hu- and importantly, likeable (e.g., [9, 47]) relative to their less man similarity of a nonhuman entity can elicit more positive humanlike counterparts. People also report greater comfort responding from its human interactants, which in turn, leads in their presence (e.g., [50]) and are more receptive to a to positive social outcomes such as improved rapport in robot interlocutor and compliant on collaborative tasks the application domains such as education, collaboration, and greater the similarity of the robot (e.g., [2, 43]). therapy (e.g., [3]). The Uncanny Valley For example, explicit humanlike cues such as a humanlike Others have suggested a more nuanced relationship be- face presented on a computer screen (as compared with a tween human similarity and people’s liking of anthropomor- text-based computer) lead people to respond more pos- phic entities. In particular, the emergence of increasingly itively to and feel more relaxed with the computer (e.g., humanlike robots and other artificial entities brought to light [52]), personified interfaces help users engage in tasks a competing phenomenon: the uncanny valley [39]. The (e.g., [29]), and humanlike web content (socially-rich text uncanny valley refers to the observation of certain entities and picture elements) increaseses perceptions of useful- – often those with a highly humanlike appearance – pro- ness, trust and enjoyment of shopping websites, leading to voking significant discomfort, instead of affinity as would be more favourable consumer attitudes (e.g., [19]). Similarly, predicted by similarity-attraction theory. equipping a robot with humanlike attributes facilitates the Over the course of nearly five decades since Masahiro formation of rapport with, empathic responding towards, Mori’s formal introduction of the uncanny valley theory into and positive appraisals of it (e.g., [9, 47, 50]). scientific discourse, empirical inquiries have compiled sub- The Anthropomorphization of Social Robots stantial evidence of its existence (for a review, see [28]). Within the human-robot interaction community, this has Specifically, people tend to rate highly humanlike entities as lead to a pervasive assumption that people’s liking of so- eerie and more unnerving than less humanlike instances cial robots is a monotonically-increasing function of human (e.g., [33]). Such dislike appears to also manifest in infants similarity (i.e., greater human similarity is always better). [31, 35] and even other primates [53], suggesting the gen- This assumption is reflected by the sheer number of engi- eral phenomenon is relatively pervasive. neering efforts towards developing humanlike robots (see Yet, uncanny valley theory continues to be critically ques- Figure 1: Exemplars of a human Table1) to the instantiation of a new field of study ( an- (top) and two robots of varying tioned due to various inconsistencies in and shortcomings droid science) devoted to this topic [21]. Consistent with human similarity (center, bottom). of empirical probes (e.g., [5, 28, 60]). For example, peo- ple respond negatively towards some but not all instances model of emotion regulation [15] as theoretical grounding. Affetto [23] of highly humanlike agents (e.g., [48]). Similarly for cer- The motivations for doing so follow from the literature on Aila [30] tain humanlike attributes, while some find their application avoidant behavior, defining it as a person’s unwillingness to Asimo [49] provokes dislike (e.g., [45]), others find the exact opposite experience negative emotions and their desire to change BARTHOC [16] (e.g., [58]). It thus remains difficult to decide which perspec- the form or frequency of situations giving rise to those expe- Baxter [12] tive to take in the design of these social technologies. riences (e.g., [11]) – which is an example of emotion regula- CB2 [37] tion through situation selection. Emiew [20] At the root of the above issues lay several gaps in the lit- Geminoid series [40] erature. Most notably, the community lacks a consistent The implications of avoidant behavior is particularly im- HRP [26] methodology for investigating the uncanny valley and its portant to human-computer and human-robot interaction HRP-4C [27] effects. Moreover, albeit due to practical limitations, the lit- given a primary aim of design is to facilitate interaction. HUBO series [44] erature largely draws on subjective assessment and on the Thus, while increasing a robot’s human similarity can effect iCub [36] comparison of very few agents. Given the high variability of positive social outcomes, it remains crucial to understand Kaspar [8] both subjective measures and the appearance of real-world when/why a design effects negative responding. Hence, Kobian [59] robots, it may serve to explain at least some of the inconsis- the purpose of Studies 2–3 was to investigate whether the Kojiro [38] tencies between studies using different robots. uncanny valley presents a serious consideration for human- Enon [25] robot interaction via more objective assessment of its im- Flobi [32] Present Research pact on people’s behavior. That is, we wanted to determine MDS [6] In the following sections, we present three experimental in- whether highly humanlike robots can be so emotionally mo- Nao [14] vestigations of people’s perceptions of a large sample of tivating that they evoke avoidant behavior. 1 REEM [56] real-world robots. Here we showed participants a series Repliee series [34] of pictures depicting humans and robots of varying human In Studies 2–3, we again presented the series of pictures Robonaut [1] similarity (low, moderate, and high). We collected partici- depicting humans and robots of varying human similarity, Robovie [22] pants’ subjective ratings of the agents’ appearances on two but with the addition of an option to press a button to re- Saya [18] dimensions: human similarity as a manipulation check, and move the picture if the participant wished to stop looking at Telenoid [41] eeriness to determine whether the set of humanlike robots it. In addition to the subjective ratings of the agents’ appear- Twendy-One [24] refelcts an uncanny valley. ances, we collected the percentage of button presses to Valkyrie [46] measure the frequency of attempts to end encounters with To move towards more objective assessment, Studies 2 Wabian [42] the various agents as an index of avoidant behavior. and 3 additionally propose and validate a novel protocol Wakamaru [51] for measuring people’s behavior towards humanlike robots. Study 1 Table 1: Exemplars of humanlike Specifically, Studies 2–3 investigate the link between the Based