Goal attribution without agency cues: the perception of ‘pure reason’ in infancy
Introduction
The currently influential domain-specificity approach holds that early cognitive development can be usefully characterised in terms of independent domain-specific systems of knowledge (e.g. Hirschfeld & Gelman, 1994). In this view the interpretation of and reasoning about the behaviour and properties of entities belonging to a given domain is governed by a set of core principles (Carey & Spelke, 1994) specific to that domain. Concrete proposals about domain-specific organisation of knowledge in cognitive development have been advanced for fields such as language (Chomsky, 1980), physics (Spelke, 1990), biology (Carey, 1985), number (Wynn, 1992), or psychology (Premack, 1990, Leslie, 1994).
One central question that arises within the domain-specificity approach concerns the nature of the selectional processes involved in the identification of the entities that belong to a given domain (cf. Carey & Spelke, 1994). How do people decide, for example, when it is appropriate (or useful) to invoke psychological principles (as opposed to, say, principles of naive physics) to explain or predict diverse phenomena they observe? In this paper we focus on this issue as it relates to the developmental origins of naive psychological understanding and examine how infants and young children come to identify the scope of this particular domain of knowledge.
Naive theory of mind has evolved to represent and interpret people's behaviour in terms of intentional causal mental states such as beliefs and desires (Dennett, 1987, Fodor, 1987, Fodor, 1992, Leslie, 1994), and so, arguably, the proper domain of our naive psychology is human action and the intentional mind states that are causally related to it. Yet, people readily attribute intentions, beliefs, and purposes to immaterial imaginary entities (gods, spirits, etc.), to human artefacts (such as computers), to natural phenomena (such as tornadoes) or to abstract processes (such as natural evolution). A well-known early demonstration of the ‘promiscuity’ of mentalistic explanations is the classical study of Heider and Simmel (1944) in which adult human subjects spontaneously provided rich anthropomorphic interpretations in terms of desires, intentions and beliefs for the animated behavioural interactions of geometric shapes. Similarly, Piaget's work on animistic and finalistic explanations in childhood demonstrated that even young children tend to interpret the behaviour of non-human entities (such as clouds) in terms of attributed intentions and beliefs. The usual explanation for such a tendency to apply psychological concepts to non-human entities holds that it reflects an overextension of an interpretational strategy which has originally developed for the explanation of human intentional actions (Piaget, 1929).
This general view implies that at its ontogenetically earliest emergence naive psychological reasoning is applied exclusively to its proper domain, that is, to the intentional actions of human beings. The initial categorisation of an object as human is thought to be based on the detection of perceptual and/or behavioural features characteristic of human beings and is seen as a precondition for applying the principles of psychological reasoning to interpret its behaviour. In so far as a non-human entity exhibits similar human-like features or behavioural cues, the psychological explanatory framework may become overextended to it as well.
One recent example of such an approach is provided by Meltzoff (1995), who has reported evidence that after having observed an adult model perform three failed instrumental attempts to achieve a goal, 18-month-old infants would re-enact the intended act (that they had never seen realised), and not the failed attempts. However, when the same (unsuccessful) actions were modelled not by a person but by a mechanical device, the infants failed to produce the target act. Meltzoff draws two major conclusions: by the time infants are 18 months old (a) they ‘are construing behaviour in terms of a psychological framework including goals of acts’ and (b) they ‘are thinking in terms of goals that are connected to people and not to things’ (p. 848).
Based on his extended work on neonatal imitation (Meltzoff and Moore, 1977, Meltzoff and Moore, 1989, Meltzoff and Moore, 1994). Meltzoff argues that people are perceptually identified on the basis of their ability to perform biomechanical bodily movements (such as facial expressions or manual acts) that the infant can map onto his/her innate ‘supramodal body scheme’. This results in the behaviour becoming categorised as a ‘human act’ (Meltzoff and Gopnik, 1993, Meltzoff and Moore, 1994) which is a precondition for attributing a goal to it. Since the behaviour of the mechanical device was not human-like in the above sense, it was categorised as a thing rather than a person, and so the infants did not proceed to infer a goal for the robot's behaviour.
Note, however, that there are alternative ways to explain the lack of re-enactment of the intended act in the case of the mechanical device: for example, it may be that the lack of biomechanical movement in the target act restricts the tendency to imitate the act, rather than restricting the tendency to infer a goal from the pattern of behaviour observed. Alternatively, it may also be the case that infants of that age interpret a behaviour as a ‘failed attempt to achieve a goal’ only if the object belongs to an already familiar object kind whose actions have on previous occasions been successfully interpreted as goal-directed.
The plausibility of such alternative accounts for the Meltzoff finding is strengthened by the results of a recent habituation study (Csibra, Gergely, Brockbank, Bı́ró & Koós 1998) which demonstrated that already at 12 months infants can infer a goal for an incomplete action that is performed by a computer-animated circle which shows no human-like features and whose behaviour is not biomechanical. Similarly, Johnson, Slaughter and Carey (1998) have shown that 12-month-olds would visually follow the spatial orientation of a non-human object that had exhibited contingent reactivity to the infant's behaviour, even in conditions in which the object had no human-like features and its behaviour was not biomechanical. Since this kind of ‘gaze following’ response is often considered to reflect (or at least be a precursor of) understanding referential mental states in others (Baron-Cohen, 1994), the finding suggests that neither human-like facial or bodily features nor biomechanical movement are necessary preconditions for the application of psychological principles to interpret an object's behaviour. In fact, the Johnson et al. (1998) study identifies another candidate as the potential basis for categorising an object as belonging to the domain of naive psychological explanations, namely, the object's ability to show contingent reactivity at a distance (see also Mandler, 1992).
In an earlier study (Gergely, Nádasdy, Csibra & Bı́ró 1995) we provided evidence that, comparably to adult subjects (Heider & Simmel, 1944), one-year-olds can interpret the behaviour of computer animated shapes (exhibiting no human-like features) as a case of goal-directed rational action. (In Expt. 1 below we report a replication of this study with 9- and 6-month-olds.) The infants were habituated to a visual event in which they observed a small circle repeatedly approach and contact a large circle by ‘jumping over’ a rectangular figure separating them (Fig. 1A). Adults tend to interpret this behaviour as an instrumental action to achieve a goal state (that of reaching or contacting the large circle), because the jumping approach is seen as a justifiable action to realise such a goal given the constraints of the reality context (i.e. the presence of an ‘obstacle’ separating the two circles). Whether or not one-year-olds also interpret this event in the same manner was assessed by presenting them, following habituation, with test trials in which the ‘obstacle’ was no longer present (Fig. 1C,D). Infants saw either a novel action (the ‘straight line approach’, see Fig. 1D) which, given that the ‘obstacle’ has now been removed, could be judged as a sensible action towards the same goal state, or the already familiar ‘jumping approach’ (Fig. 1C) which, however, could no longer be justified in the new situation as a rational action to achieve the goal. Since the subjects showed more recovery of attention in the test phase when they saw the familiar (but now non-rational) ‘jumping action’ than when they were presented with the new (but sensible) ‘straight line approach’, Gergely et al. (1995) concluded that the one-year-olds had indeed interpreted the habituation event within a naive psychological framework as a rational action towards a particular goal state. Note again that such an interpretation was generated even though the two-dimensional abstract figures on the computer screen showed no featural resemblance to a human person.
What sort of interpretational system may one-year-olds possess to generate the kind of representation of the habituation event that is implied by the above findings? Elsewhere we argued (Gergely and Csibra, 1997, Csibra and Gergely, 1998) that to interpret such an event as a goal-directed action infants apply what we call the ‘teleological stance’ or a ‘naive theory of rational action’. This early interpretational system establishes a specific explanatory relation among three representational elements: the action, the goal state, and the constraints of physical reality. Such a representational structure constitutes a well-formed teleological interpretation, however, only if it satisfies the principle of rational action which states that an action can be explained by reference to a future goal state if and only if it is seen as the most justifiable action towards that goal state that is available within the constraints of reality.
In this model, the behaviour of the jumping circle can be explained teleologically by reference to its end state (contacting the large circle) if the rectangle separating them is interpreted as an impenetrable solid obstacle; this way the jumping approach can be construed as a rational action leading through the shortest available path to the large circle. Note that the principle of rational action also generates an action prediction for the new situation in which the obstacle is no longer present: the small circle ought to approach its goal through the most justifiable (shortest straight line) route that has now become available. In fact, the one-year-olds seem to have generated such an expectation because they looked longer (experiencing incongruence) when the small circle's behaviour remained unchanged after the removal of the obstacle.
In sum: the studies discussed above seem to indicate (a) that as early as 12 months of age infants can apply a naive psychological framework to interpret the behaviour of objects, and (b) that such an interpretation is not dependent on the detection of human-like features or biomechanical movement indicating personhood. As an alternative to the strict linkage of psychological reasoning to persons several authors have proposed that the initial domain of the application of psychological principles corresponds to the wider ontological category of agents. Agents are objects with the capacity to become the causal source of events and exhibit movement cues such as self-propulsion which may serve as the perceptual basis for their identification. In fact, a number of proposals hypothesised that humans are hard-wired to detect self-propelled movement which triggers the direct perceptual categorisation of the object as agent (Baron-Cohen, 1994, Leslie, 1994, Leslie, 1995, Premack, 1990, Premack and Premack, 1995b). According to these proposals (see also Gergely et al., 1995) the detection of movement cues indicating agency is a necessary precondition for the application of psychological principles (for example, goal attribution) to interpret the behaviour of the object.
The concept of agency, however, has been conceptualised at several different levels. In Leslie's tripartite theory of agency (Leslie, 1994, Leslie, 1995) self-propulsion is interpreted by the first component (the ‘theory of body mechanism’) of his hierarchically organised system in terms of the physical notion of ‘force’ and its detection triggers the categorisation of the object as a ‘physical or mechanical agent’ with an internal and renewable source of energy. This agency concept belongs to the domain of naive physics, however, and thus lies outside the psychological domain of theory of mind. Therefore, in and of itself, the detection of self-initiated movement is not sufficient to provide an intentional interpretation of the object's behaviour. The resulting categorisation of the object as a mechanical agent, however, serves as input for the second stage of Leslie's model (‘theory of mind mechanism, system1’). This component of ‘actional agency’ then assigns goals to the actions of self-propelled agents and interprets their behaviour in terms of ‘teleological causation’.
In contrast, the type of modularist models proposed by Premack (1990) or Baron-Cohen (1994) suggests that the detection of self-propulsion triggers the direct categorisation of the object as an ‘intentional agent’. Intentional agents are self-propelled objects whose actions are caused by intentional mental states, they pursue goals, and they can react contingently to the behaviour of other objects at a distance. In order to assign a particular intention or goal to the object, however, the detection of self-initiated movement is not sufficient, so the module will search for further cues (such as directionality, trajectory, or equifinal outcomes, see Baron-Cohen, 1994, Premack and Premack, 1995b) whose presence will lead to the mandatory assignment of a goal. As Premack and Premack (1995a) put it: ‘When activated by the appropriate movement – that of self-propelled object – the first component in the infant's social system outputs the interpretation intentional. The next interpretation that it outputs is goal-directed. Intentional [based on self-propulsion] is a necessary, though not sufficient, condition for this interpretation.’ (p. 208).
Thus, in both of the above approaches the detection of movement cues such as self-propulsion is a necessary, though not sufficient, condition for setting up an intentional interpretation of the object's behaviour. In fact, the Gergely et al. (1995) and the Johnson et al. (1998) studies both provided empirical support for this idea, because they included control conditions in which the object did display self-initiated movement, but, nevertheless, the infants showed no evidence of having interpreted their behaviour in terms of psychological principles. Neither of these studies could, however, answer the question whether the movement cues of agency are, indeed, necessary for interpreting the behaviour of the object within the explanatory framework of naive psychology. This is so because in the experimental conditions of both studies, where a psychological interpretation was demonstrated, the acting object exhibited self-propelled movement. The central aim of this paper, therefore, is to examine whether movement cues of agency are necessary at all for the application of principles of naive psychology, or whether the processes involved in goal assignment are independent of the perception of the type of causal source of energy (internal vs. external) that drives the object's behaviour.
The paper reports three habituation studies. The first experiment is a replication of Gergely et al. (1995) with 9- and 6-month-old subjects to see how early in infancy the sophisticated ability to interpret behaviour teleologically in terms of the principle of rational action occurs. Experiments 2 and 3 are designed to investigate whether this precocious ability for teleological interpretation is dependent on the presence of movement cues of agency such as self-propulsion.
Section snippets
Subjects
Forty-eight 9-month-old (26 males and 22 females, mean age 37.8 weeks, range 32.6–42.7 weeks) and 24 6-month-old (8 males and 16 females, mean age 26.9 weeks, range 24.3–29.3 weeks) infants participated in the study. An additional 26 nine-month-olds and 29 six-month-olds were also tested but were excluded from the data analysis for the following reasons: fussiness or falling asleep (6 and 12), failing to reach the habituation criterion within 16 trials (7 and 6), experimental error (2 and 2),
Experiment 2
As we have noted above, while the results of Gergely et al. (1995) and Expt. 1 suggest that movement cues of agency, such as self-propulsion, may not be sufficient to trigger the attribution of an intention or goal to an object, they, nevertheless, cannot rule out the possibility that the presence of such cues is a necessary precondition for interpreting behaviour from the teleological stance. Therefore, the aim of Expt. 2 was to test whether the perception of self-initiated movement (or of
Experiment 3
In Expt. 2, each event started with the large object moving from the middle to the left side of the screen. We included this movement in order to demonstrate to the infants in the control condition that there was an alternative route available to approach the target object: i.e. that it was possible to pass through under the bar ‘hanging in the air’. To make the experimental stimuli as similar to the control stimuli as possible, in the experimental condition we have also introduced a comparable
General discussion
The three studies presented here may bring us closer to answering one of the central questions of the domain-specificity approach in cognitive development which concerns the nature of our understanding of the ontological boundaries of a given domain. How do people come to identify the entities that belong to a given domain, or, in other words, how do people decide which domain-specific principles (say, of naive physics vs. naive psychology) to apply when interpreting the observed behaviour of
Acknowledgements
This research was supported by the UK Medical Research Council and by grant #T013613 from the National Research Foundation of Hungary (OTKA). We thank Gábor Újvári for his help in programming the animations used in Expts. 1 and 2. We also thank Renée Baillargeon, Suzie Johnson, Szabolcs Kiss, Jean Mandler, Bruce Pennington, Josef Perner, Csaba Pléh, John S. Watson and two anonymous reviewers for their valuable comments on earlier versions of this paper.
References (49)
Representing the existence and the location of hidden objects: Object permanence in 6- and 8-month-old infants
Cognition
(1986)A theory of the child's theory of mind
Cognition
(1992)- et al.
Teleological reasoning in infancy: The infant's naive theory of rational action
A reply to Premack and Premack. Cognition
(1997) - et al.
Taking the intentional stance at 12 months of age
Cognition
(1995) - et al.
Imitation, memory, and the representation of persons
Infant Behavior and Development
(1994) - et al.
A theory of the role of imitation in the emergence of self
The role of action-initiation in young children's causal explanations of action
Cognitive Development
(1996)- et al.
Social understanding at the end of the first year of life
Developmental Review
(1994) - et al.
Infants' concept of animacy
Cognitive Development
(1996) The infant's theory of self-propelled objects
Cognition
(1990)
Principles of object perception
Cognitive Science
How to build a baby that can read minds: Cognitive mechanisms in mind reading
Cahiers de Psychologie Cognitive/Current Psychology of Cognition
Intentional relations and social understanding
Behavioral and Brain Sciences
Giving meaning to movement: A developmental study
Ecological Psychology
Conceptual Change in Childhood
Domain-specific knowledge and conceptual change
Rules and Representations
Development of joint visual attention in infants
The teleological origins of mentalistic action explanations: A developmental hypothesis
Developmental Science
The Intentional Stance
Psychosemantics
Distinguishing between animates and inanimates: not by motion alone
On the ascription of intentional content
Cahiers de Psychologie Cognitive/Current Psychology of Cognition
Cited by (424)
Infants can use temporary or scant categorical information to individuate objects
2024, Cognitive PsychologySeeing social interactions
2023, Trends in Cognitive SciencesThe pupil collaboration: A multi-lab, multi-method analysis of goal attribution in infants
2023, Infant Behavior and DevelopmentChildren and adults use pragmatic principles to interpret non-linguistic symbols
2023, Journal of Memory and LanguageCommonsense psychology in human infants and machines
2023, CognitionInfants do not use payoff information to infer individual goals in joint-action events
2023, Cognitive Development