DOCSLIB.ORG

MIT Encyclopedia of the Cognitive Sciences

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
MIT Encyclopedia of the Cognitive Sciences Introspection 419 Trevarthen, C. (1994). Infant semiosis. In W. Noth, Ed., Origins of and hence whether introspection, properly so-called, exists. Semiosis. Berlin: Mouton de Gruyter, pp. 219–252. According to Gilbert Ryle (1949) and William Lyons Trevarthen, C. (1995). Contracts of mutual understanding: negoti- (1986), what we loosely describe as attending to current ating meaning and moral sentiments with infants. In P. Wohl- perceptions is really just perceiving in an attentive manner. muth, Ed., The Crisis of Text: Issues in the Constitution of But perceiving attentively itself sometimes involves attend- Authority. San Diego: University of San Diego School of Law, Journal of Contemporary Legal Issues, Spring 1995, vol. 6: ing to the perceiving, as when one is explicitly aware of 373–407. visually concentrating on something. Moreover, when we report what mental states we are in, those reports express higher-order mental representations of the states we report; Intonation Ryle’s denial that remarks such as “I am in pain” are liter- ally about one’s mental states is groundless. See PROSODY AND INTONATION; PROSODY AND INTONA- It is often held that introspection involves some “inner TION, PROCESSING ISSUES sense” by which we perceive our own mental states. The seemingly spontaneous and unmediated character of per- ceiving generally would then explain why introspection Introspection itself seems spontaneous and immediate. This model could, in addition, appeal to mechanisms of perceptual attention to Introspection is a process by which people come to be atten- explain how we come to focus attentively on our concurrent tively conscious of mental states they are currently in. This mental states. focused CONSCIOUSNESS of one’s concurrent mental states is But introspection cannot be a form of perceiving. Per- distinct from the relatively casual, fleeting, diffuse way we ception invariably involves sensory qualities, and no quali- are ordinarily conscious of many of our mental states. ties ever occur in introspection other than those of the “Introspection” is occasionally applied to both ways of sensations and perceptions we introspect; the introspecting being conscious of one’s mental states (e.g., Armstrong itself produces no additional qualities. Moreover, speech 1968/1993), but is most often used, as in what follows, for acts generally express not perceptions, but thoughts and the attentive way only. other intentional states (see INTENTIONALITY). So intro- Introspection involves both the mental states intro- spective reports express intentional states about the mental spected and some mental representation of those very states states we introspect, and introspective representations of (as suggested by the etymology, from the Latin spicere concurrent mental states involve assertive intentional states, “look” and intra “within”; looking involves mental repre- or thoughts. Introspection is deliberate and attentive sentations of what is seen). Because it involves higher- because these higher-order intentional states are themselves order mental representations of introspected states, intro- attentive and deliberate. And our introspecting seems spon- spection is a kind of conscious METAREPRESENTATION or taneous and unmediated presumably because we remain METACOGNITION. unaware of any mental processes that might lead to these WILHELM WUNDT (1911/1912) held that introspection higher-order intentional states. Introspection consists in provides an experimental method for psychology, and relied conscious, attentively focused, higher-order thoughts about on it in setting up, in 1879 in Leipzig, the first experimental our concurrent mental states. psychology laboratory. Some challenged this introspection- Despite Comte’s claim that attention cannot be divided, ist method, following Auguste Comte’s (1830–42) denial people can with a little effort attend to more than one thing. that a single mind can be both the agent and object of intro- And attentive consciousness of concurrent mental states spection. This, Comte had held, would divide attention could in any case occur whenever the target mental state between the act and object of introspecting, which he was not itself an attentive state. thought impossible. These concerns led WILLIAM JAMES A related concern is that attending to concurrent mental (1890) and others to propound instead a method of immedi- states may distort their character. But it is unclear why that ate retrospection. should happen, inasmuch as attention does not generally Introspectionist psychology foundered mainly not for alter the properties of its object. Introspection itself cannot these reasons, but because results from different introspec- show that distortion occurs, because even if it seems to, that tionist laboratories frequently conflicted. Still, experimental appearance might be due not to the distorting effect of intro- procedures in psychology continue to rely on subjects’ spection, but to introspection’s making us aware of more of access to their current mental states, though the theoretical a state’s properties or of a different range of properties. Sim- warrant for this reliance is seldom discussed. ilarly for the idea that introspective attention might actually The phenomenological movement in philosophy, pio- bring the introspected state into existence (Hill 1991: chap. neered by Wundt’s contemporary Edmund Husserl (1913/ 5). That may well happen, but it may instead be that, when 1980), held that introspection, by “bracketing” consciousness that seems to happen, introspection simply makes one newly from its object, enables us to describe and analyze conscious- aware of a state that already existed. ness, and thereby solve many traditional philosophical prob- Work by John H. Flavell (1993) has raised doubt about lems. This methodology encountered difficulties similar to whether children five and younger have introspective access those that faced introspectionist psychology. to their mental states. Four- and five-year-olds describe Some have questioned whether higher-order mental rep- themselves and others as thinking, feeling, and experienc- resentations of concurrent mental states ever actually occur ing. But they also describe people while awake as going for 420 Introspection significant periods without thinking or feeling anything Flavell, J. H. (1993). Young children’s understanding of thinking whatever. Doubtless these children themselves have, when and consciousness. Current Directions in Psychological Sci- awake, normal streams of consciousness. But they seem not ence 2(2): 40–43. to think of themselves in that way and, hence, not to intro- Hill, C. S. (1991). Sensations: A Defense of Type Materialism. spect their streams of consciousness. Flavell also reports Cambridge: Cambridge University Press. Husserl, E. (1913/1980). Ideas Pertaining to a Pure Phenomenol- that these children determine what people attend to and ogy and to a Phenomenological Philosophy, vol. 1. Trans. T. E. think about solely on the basis of behavioral cues and envi- Klein and W. E. Pohl. The Hague and Boston: M. Nijhoff. ronmental stimulation. So perhaps their inability to intro- James, W. (1890). The Principles of Psychology. 2 vols. New York: spect results from their simply not conceiving of thoughts Henry Holt. and experiences as states that are sometimes conscious (see Lashley, K. S. (1958). Cerebral organization and behavior. In H. C. THEORY OF MIND). Solomon, S. Cobb, and W. Penfield, Eds., The Brain and Some have held that introspective access to one’s men- Human Behavior, vol. 36. Association for Research in Nervous tal states cannot be erroneous or, at least, that it overrides and Mental Diseases, Research Publications. Baltimore: Will- all other evidence (see SELF-KNOWLEDGE). RENÉ DES- iams and Wilkins, pp. 1–18. Lyons, W. (1986). The Disappearance of Introspection. Cam- CARTES (1641/1984) famously noted that one cannot, when bridge, MA: MIT Press/Bradford Books. thinking, doubt that one is thinking. But this hardly shows Nisbett, R. E., and T. DeCamp Wilson. (1977). Telling more than that when one is thinking one always knows one is, much we can know: verbal reports on mental processes. Psychologi- less that one is invariably right about which thoughts one cal Review 84 (3): 231–259. has. In a similar spirit, Sydney Shoemaker (1996) has Ryle, G. (1949). The Concept of Mind. London: Hutchinson. urged that when one has a belief one always knows one Shoemaker, S. (1996). The First-Person Perspective and Other does, because a rational person’s believing something itself Essays. Cambridge: Cambridge University Press. involves cognitive dispositions that constitute that person’s Wundt, W. (1911/1912). An Introduction to Psychology. Trans. knowing about the belief. But the relevant rationality often Rudolf Pintner. London: George Allen and Unwin. fails to accompany our beliefs and other first-order mental states. Further Readings Indeed, psychological research reveals many such Brentano, F. (1874/1973). Psychology from an Empirical Stand- lapses of rationality. In addition to the misrepresentations point. O. Kraus, Ed., L. L. McAlister, Eng. ed. Trans. A. C. of one’s own mental states discovered by SIGMUND FREUD, Rancurello, D. B. Terrell, and L. L. McAlister. London: Rout- other work (e.g., Nisbett and Wilson 1977) shows that ledge and Kegan Paul. introspective judgments frequently result from confabula- Broad, C. D. (1925). The Mind and Its Place in Nature. London: tion. People literally invent mental states to explain
Load more

Recommended publications
  • Theory of Mind As Inverse Reinforcement Learning. Current
    Available online at www.sciencedirect.com ScienceDirect Theory of mind as inverse reinforcement learning 1,2 Julian Jara-Ettinger We review the idea that Theory of Mind—our ability to reason think or want based on how they behave. Research in about other people’s mental states—can be formalized as cognitive science suggests that we infer mental states by inverse reinforcement learning. Under this framework, thinking of other people as utility maximizers: constantly expectations about how mental states produce behavior are acting to maximize the rewards they obtain while mini- captured in a reinforcement learning (RL) model. Predicting mizing the costs that they incur [3–5]. Using this assump- other people’s actions is achieved by simulating a RL model tion, even children can infer other people’s preferences with the hypothesized beliefs and desires, while mental-state [6,7], knowledge [8,9], and moral standing [10–12]. inference is achieved by inverting this model. Although many advances in inverse reinforcement learning (IRL) did not have Theory of mind as inverse reinforcement human Theory of Mind in mind, here we focus on what they learning reveal when conceptualized as cognitive theories. We discuss Computationally, our intuitions about how other minds landmark successes of IRL, and key challenges in building work can be formalized using frameworks developed in a human-like Theory of Mind. classical area of AI: model-based reinforcement learning 1 Addresses (hereafter reinforcement learning or RL) . RL problems 1 Department of Psychology, Yale University, New Haven, CT, United focus on how to combine a world model with a reward States function to produce a sequence of actions, called a policy, 2 Department of Computer Science, Yale University, New Haven, CT, that maximizes agents’ rewards while minimizing their United States costs.
    [Show full text]
  • The Role of Language in Theory of Mind Development
    Top Lang Disorders Vol. 34, No. 4, pp. 313–328 Copyright c 2014 Wolters Kluwer Health | Lippincott Williams & Wilkins The Role of Language in Theory of Mind Development Jill G. de Villiers and Peter A. de Villiers Various arguments are reviewed about the claim that language development is critically connected to the development of theory of mind. The different theories of how language could help in this process of development are explored. A brief account is provided of the controversy over the capacities of infants to read others’ false beliefs. Then the empirical literature on the steps in theory of mind development is summarized, considering studies on both typically developing and various language-delayed children. Suggestions are made for intervention by speech language pathologists to enhance the child’s access to understanding the minds of others. Key words: conversation, false beliefs, false complements, language, narrative, theory of mind N THIS ARTICLE, we review arguments on interactions between language and ToM I and evidence for the claim that children’s in both typically developing and language- language acquisition is crucially connected to delayed children. We end with suggestions their development of a theory of mind (ToM). for using language interventions to facilitate Theory of mind refers to the child’s ability to children’s understanding of their own mental understand that other people have minds, and states as well as the minds of others. those minds contain beliefs, knowledge, de- sires, and emotions that may be different from THEORIES OF THE RELATIONSHIP those of the child. If the child can figure out BETWEEN LANGUAGE AND ToM those contents, then other people’s behavior will begin to make sense and, therefore, be Many writers have made a convincing case predictable and explainable.
    [Show full text]
  • Consciousness and Theory of Mind: a Common Theory?
    Consciousness and Theory of Mind: a Common Theory? Keywords: Consciousness, Higher-Order Theories, Theory of Mind, Mind-reading, Metacognition. 1 Introduction Many of our mental states are phenomenally conscious. It feels a certain way, or to borrow Nagel's expression, there is something it is like to be in these states. Examples of phenomenally conscious states are those one undergoes while looking at the ocean or at a red apple, drinking a glass of scotch or a tomato juice, smelling coee or the perfume of a lover, listening to the radio or a symphonic concert, or feeling pain or hunger. A theory of consciousness has to explain the distinctive properties that phe- nomenally conscious states have and other kind of states lack. Higher-Order Representational (HOR) theories1 attempt to provide such an explanation. Ac- coding to these theories, phenomenally conscious states are those that are the objects of some kind of higher-order process or representation. There is some- thing higher-order, a meta-state, in the case of phenomenal conscious mental states, which is lacking in the case of other kind of states. According to these theories, consciousness depends on our Theory of Mind. A Theory of Mind, henceforth ToM, is the ability of humans to identify their own mental states and attribute mental states dierent from their owns to others. Such an ability can, at least conceptually, be decomposed into another two: mindreading and metacognition. Human beings are able to entertain representations of other people men- tal states thanks to the mindreading ability. We attribute beliefs, perceptions, feelings or desires to other people and predict and explain their behaviour ac- cordingly.
    [Show full text]
  • Theory of Mind for a Humanoid Robot
    Theory of Mind for a Humanoid Robot Brian Scassellati MIT Artificial Intelligence Lab 545 Technology Square – Room 938 Cambridge, MA 02139 USA [email protected] http://www.ai.mit.edu/people/scaz/ Abstract. If we are to build human-like robots that can interact naturally with people, our robots must know not only about the properties of objects but also the properties of animate agents in the world. One of the fundamental social skills for humans is the attribution of beliefs, goals, and desires to other people. This set of skills has often been called a “theory of mind.” This paper presents the theories of Leslie [27] and Baron-Cohen [2] on the development of theory of mind in human children and discusses the potential application of both of these theories to building robots with similar capabilities. Initial implementation details and basic skills (such as finding faces and eyes and distinguishing animate from inanimate stimuli) are introduced. I further speculate on the usefulness of a robotic implementation in evaluating and comparing these two models. 1 Introduction Human social dynamics rely upon the ability to correctly attribute beliefs, goals, and percepts to other people. This set of metarepresentational abilities, which have been collectively called a “theory of mind” or the ability to “mentalize”, allows us to understand the actions and expressions of others within an intentional or goal-directed framework (what Dennett [15] has called the intentional stance). The recognition that other individuals have knowl- edge, perceptions, and intentions that differ from our own is a critical step in a child’s development and is believed to be instrumental in self-recognition, in providing a perceptual grounding during language learning, and possibly in the development of imaginative and creative play [9].
    [Show full text]
  • Mind Perception Daniel R. Ames Malia F. Mason Columbia
    Mind Perception Daniel R. Ames Malia F. Mason Columbia University To appear in The Sage Handbook of Social Cognition, S. Fiske and N. Macrae (Eds.) Please do not cite or circulate without permission Contact: Daniel Ames Columbia Business School 707 Uris Hall 3022 Broadway New York, NY 10027 [email protected] 2 What will they think of next? The contemporary colloquial meaning of this phrase often stems from wonder over some new technological marvel, but we use it here in a wholly literal sense as our starting point. For millions of years, members of our evolving species have gazed at one another and wondered: what are they thinking right now … and what will they think of next? The interest people take in each other’s minds is more than idle curiosity. Two of the defining features of our species are our behavioral flexibility—an enormously wide repertoire of actions with an exquisitely complicated and sometimes non-obvious connection to immediate contexts— and our tendency to live together. As a result, people spend a terrific amount of time in close company with conspecifics doing potentially surprising and bewildering things. Most of us resist giving up on human society and embracing the life of a hermit. Instead, most perceivers proceed quite happily to explain and predict others’ actions by invoking invisible qualities such as beliefs, desires, intentions, and feelings and ascribing them without conclusive proof to others. People cannot read one another’s minds. And yet somehow, many times each day, most people encounter other individuals and “go mental,” as it were, adopting what is sometimes called an intentional stance, treating the individuals around them as if they were guided by unseen and unseeable mental states (Dennett, 1987).
    [Show full text]
  • Machine Theory of Mind
    Machine Theory of Mind Neil C. Rabinowitz∗ Frank Perbet H. Francis Song DeepMind DeepMind DeepMind [email protected] [email protected] [email protected] Chiyuan Zhang S. M. Ali Eslami Matthew Botvinick Google Brain DeepMind DeepMind [email protected] [email protected] [email protected] Abstract 1. Introduction Theory of mind (ToM; Premack & Woodruff, For all the excitement surrounding deep learning and deep 1978) broadly refers to humans’ ability to rep- reinforcement learning at present, there is a concern from resent the mental states of others, including their some quarters that our understanding of these systems is desires, beliefs, and intentions. We propose to lagging behind. Neural networks are regularly described train a machine to build such models too. We de- as opaque, uninterpretable black-boxes. Even if we have sign a Theory of Mind neural network – a ToM- a complete description of their weights, it’s hard to get a net – which uses meta-learning to build models handle on what patterns they’re exploiting, and where they of the agents it encounters, from observations might go wrong. As artificial agents enter the human world, of their behaviour alone. Through this process, the demand that we be able to understand them is growing it acquires a strong prior model for agents’ be- louder. haviour, as well as the ability to bootstrap to Let us stop and ask: what does it actually mean to “un- richer predictions about agents’ characteristics derstand” another agent? As humans, we face this chal- and mental states using only a small number of lenge every day, as we engage with other humans whose behavioural observations.
    [Show full text]
  • Theory of Mind by Margaret Alic Updated on Dec 23, 2009 From
    Theory of Mind By Margaret Alic Updated on Dec 23, 2009 from http://www.education.com/reference/article/theory-of-mind/ Theory of mind (ToM) is the intuitive understanding of one's own and other people's minds or mental states— including thoughts, beliefs, perceptions, knowledge, intentions, desires, and emotions—and of how those mental states influence behavior. Sometimes called intuitive psychology, folk psychology, or even mind-reading, ToM is an innate human ability. The understanding that others have mental states different from one's own makes it possible to infer what others are thinking and to predict their behavior. This ability to recognize one's own state of mind and those of others is central to human consciousness. The study of ToM and identification of the skills comprising ToM is a rapidly changing the field of developmental psychology. THE THEORY OF TOM The word theory in ToM refers to a person's tacit belief rather than to theory in the scientific sense. However, ToM is similar to a scientific theory in that it enables one to interpret and predict another's actions by evaluating their thoughts, beliefs, desires, or emotions. ToM also qualifies as a theory in the sense that the representations of reality generated by ToM are not directly observable. The ability to develop a ToM is a cognitive ability that develops by degrees from infancy through childhood and adolescence on into adulthood. It forms a basis for a child's acquisition of language and the development of appropriate social behavior and skills. ToM includes the ability to understand that the representations generated in one's mind do not necessarily reflect the real world and that other people may hold different representations.
    [Show full text]
  • Theory of Mind (Neural Basis)
    Theory of Mind (Neural Basis) Rebecca Saxe Department of Brain and Cognitive Sciences, MIT In Press at: Encyclopedia of Consciousness The externally observable components of human actions carry only a tiny fraction of the information that matters. Human observers are vastly more interested in perceiving or inferring the mental states - the beliefs, desires and intentions - that lie behind the observable shell. If a person checks her watch, is she uncertain about the time, late for an appointment, or bored with the conversation? If a person shoots his friend on a hunting trip, did he intend revenge or just mistake his friend for a partridge? The mechanism people use to infer and reason about another person’s states of mind is called a ‘Theory of Mind’ (ToM). One of the most striking discoveries of recent human cognitive neuroscience is that there is a group of brain regions in human cortex that selectively and specifically underlie this mechanism. The importance of mental states for behavior prediction is especially clear when the person doing the behavior is misinformed: that is, when the actor has a false belief. False beliefs therefore figure heavily in the study of ToM. The original “False Belief task” was developed for use with pre-school aged children. In the basic design, a child watches while a puppet places an object in location A. The puppet leaves the scene and the object is transferred to location B. The puppet returns and the child is asked to predict where the puppet will look for the object. Three-year-olds say the puppet will look in location B, where the object actually is; older children say the puppet will look in location A, where the puppet last saw the object.
    [Show full text]
  • What Does the Study of Autism Tell Us About the Craft of Folk Psychology?
    See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/224930550 What does the study of autism tell us about the craft of folk psychology? CHAPTER · JANUARY 2008 CITATION READS 1 67 2 AUTHORS, INCLUDING: Richard Griffin Tufts University 11 PUBLICATIONS 286 CITATIONS SEE PROFILE Available from: Richard Griffin Retrieved on: 31 January 2016 Griffin, R. & Dennett, D.C. (2008). What does the study of autism tell us about the craft of folk attribution of "intentions" was at the basis of folk psychology has seen something of a psychology? In T. Striano & V. Reid (Eds.) Social Cognition: Development, Neuroscience, and resurgence. In the 1980s and 1990s, the child-development literature focused largely on Autism (pp. 254-280). Wiley-Blackwell. what Bennett (1978) called the "rockbottom issue" of mental-state-attribution - beliefs about beliefs - publishing hundreds of experiments on the topic (Wellman, Cross & Watson, 2001). Yet, with technical advances in probing the knowledge of infants and the ascendance of neuroimaging, the focus on belief has waned, and has shifted to the What Does the Study of Autism Tell us about the putatively "simpler" mental states of intentions, goals, and desires (Johnson, 2000). Craft of Folk Psychology? According to a recent New York Times article entitled "Cells that read minds" (Blakeslee 2006), Premack and Woodruff's initial claim appears to have been vindicated. The article touts the properties, or at least certain claims about the properties, of "mirror" Richard Griffin and Daniel C. Dennett neurons, first discovered in the premotor cortex of macaque monkeys (Gallese, Fadiga, Fogassi, & Rizzolatti, 1996).
    [Show full text]
  • The You-Turn in Philosophy of Mind: on the Significance of Experiences That Aren’T Mine
    THE YOU-TURN IN PHILOSOPHY OF MIND: ON THE SIGNIFICANCE OF EXPERIENCES THAT AREN’T MINE A DISSERTATION SUBMITTED TO THE GRADUATE DIVISION OF THE UNIVERSITY OF HAWAI‘I AT MĀNOA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY IN PHILOSOPHY AUGUST 2018 By Joshua Stoll Dissertation Committee: Arindam Chakrabarti, Chairperson Vrinda Dalmiya Ronald Bontekoe Joseph Tanke Jesse Knutson Ashley Maynard Acknowledgements I could not have written this dissertation without the help and support of others, including my family, my friends, and my professors at the UH Mānoa Department of Philosophy. My parents Luci and Howie Stoll have always been there to unwaveringly support my endeavors. They have been a great source of love, inspiration, and guidance. I owe them everything I have become. My brother Adam and his wife Lisa and their kids have also been a great source of love, inspiration, and guidance. And I’m thankful that my nieces and nephew really got me moving on this project by playfully asking me every Thanksgiving (perhaps with some prodding from my parents) if I have finished it yet. In addition, I’d like to thank my fellow graduate student colleague-friends in the department, in particular: Matt Izor, Elyse Byrnes, Ben Zenk, Brandon Underwood, Jane Allred, and Joel Label. More particularly, out of this group, I’d like to thank Sydney Morrow for being a close friend and great officemate, as well as Ian Nicolay, for being a close friend and great roommate. Our camaraderie and support of each other, the community we formed, truly made me feel at home here in Hawai‘i.
    [Show full text]
  • Artificial Intelligence and Its Applications in Theory of Mind
    المجلة اﻷكاديمية لﻷبحاث والنشر العلمي | اﻹصدار الرابع | تأريخ اﻹصدار: 9102-8-5 ISSN: 2706-6495 Artificial Intelligence and its Applications in Theory of Mind Hanan Mohammed al-qahtani King Fahd University Email: [email protected] Abstract The first use of the phrase “Artificial Intelligence” in 1956 is attributed to John McCarthy of the University of Massachusetts. It is a fairly new field which involves programming computers to play games, comprehend and respond to natural language, reproduce neural networks like those of humans and exhibit sensitivity (Hear, see, move, and react to the environment) like humans (this branch is termed robotics). First we must define Artificial Intelligence. A number of proposed definitions would be considered here. Keywords: Artificial Intelligence, Theory of Mind, AI www.ajrsp.com 1 المجلة اﻷكاديمية لﻷبحاث والنشر العلمي | اﻹصدار الرابع | تأريخ اﻹصدار: 9102-8-5 ISSN: 2706-6495 Introduction The first use of the phrase “Artificial Intelligence” in 1956 is attributed to John McCarthy of the University of Massachusetts. It is a fairly new field which involves programming computers to play games, comprehend and respond to natural language, reproduce neural networks like those of humans and exhibit sensitivity (Hear, see, move, and react to the environment) like humans (this branch is termed robotics). First we must define Artificial Intelligence. A number of proposed definitions would be considered here. It is the science and engineering of making intelligent machines, especially intelligent computer programs. The ability of a computer or other machine to perform actions thought to require intelligence… An intelligent machine would be more flexible than a computer and would engage in the kind of "thinking" that people actually do.
    [Show full text]
  • Machine Theory of Mind
    Machine Theory of Mind Neil C. Rabinowitz 1 Frank Perbet 1 H. Francis Song 1 Chiyuan Zhang 2 Matthew Botvinick 1 Abstract of others, such as their desires and beliefs. This ability is typically described as our Theory of Mind (Premack Theory of mind (ToM) broadly refers to humans’ & Woodruff, 1978). While we may also leverage our ability to represent the mental states of others, own minds to simulate others’ (e.g. Gordon, 1986; Gallese including their desires, beliefs, and intentions. & Goldman, 1998), our ultimate human understanding of We design a Theory of Mind neural network – other agents is not measured by a correspondence between a ToMnet – which uses meta-learning to build our models and the mechanistic ground truth, but instead such models of the agents it encounters. The by how much they enable prediction and planning. ToMnet learns a strong prior model for agents’ future behaviour, and, using only a small num- In this paper, we take inspiration from human Theory of ber of behavioural observations, can bootstrap Mind, and seek to build a system which learns to model to richer predictions about agents’ characteris- other agents. We describe this as a Machine Theory of tics and mental states. We apply the ToMnet Mind. Our goal is not to assert a generative model of to agents behaving in simple gridworld environ- agents’ behaviour and an algorithm to invert it. Rather, we ments, showing that it learns to model random, focus on the problem of how an observer could learn au- algorithmic, and deep RL agents from varied tonomously how to model other agents using limited data populations, and that it passes classic ToM tasks (Botvinick et al., 2017).
    [Show full text]