6 Beyond the object-dependence view: the important differences between tracking intentional agents and non intentional objects

The evidence reviewed so far is compatible with what we called the object-dependence view of agent tracking. In the terminology of the object-file theory, the conclusion would be that object files may, on occasion, be put to use in tracking intentional agents; agent files are thus sometimes reduced to object files. However, there are reasons for thinking that, while we do track intentional agents by tracking their bodies, we should resist the stronger, deflationary view. For there appear to be important differences between tracking physical objects and tracking intentional agents.

The variety of ontological theories of personal identity 

First, the distinction between object and agent tracking is required by the ontological theories of personal/self identity which are not consistent with the thesis of the co-instantiation expressed in section 5, which states that a human agent is co-instantiated with his/her material human body. The conflict would heighten if one replaces ‘agent’ by ‘identical person’ in the thesis of the co-instantiation. The latter version of the thesis is rejected by a number of metaphysics for which the criteria of personal identity are not reducible to objecthood/bodiliness criteria, but pertain to other criteria such as psychological continuity (e.g., Shoemaker, 1997) or the understanding specific to shared intentionality and social folkpsychology (e.g., Ames et al., 2001; Tomasello, Carpenter, Call, Behne, & Moll, in press). For instance, Shoemaker (1997) defends a functionalist analysis of personal identity which rests on conditions of psychological continuity that persist in spite of dramatic bodily transformations such as brain transplantations or teleportation. On the credit of an object-dependence view of personal identity, is worth noting however that the compatibility of Shoemaker’s analysis with materialism remains subject to caution (van Inwagen, 1997). In addition, a number of metaphyscians defend ontological analyses of personal identity which seem consistent with the thesis of the co-instantiation for personal identity. For instance van Inwagen (1990) and Merricks (2001)⁶ uphold that each person is a material object and persists over time because this person is identical to the biological organism he/she is (strictly). In the present article, our general use of ‘agent’ instead of ‘person’ is an attempt to remain as neutral as possible with respect to the puzzles of personal identity. Our claim focuses on agency and can remain true with a variety on different ontologies of personal identity.

Argument related to means of direct perceptual anchoring into agency (agent indexing) 

Apart from the previous ontological considerations, arguments can also be found with respect to an examination of the perceptual indexing of agents. Even though this leaves intact the thesis of body tracking (section 5), there are important differences in the properties that can trigger and maintain singular agent perceptions – i.e. the opening/indexing and maintenance of ‘agent files’ –, and not just in the descriptive information carried out by the singular perceptual representation. One argument supporting this claim is that specific mechanisms seem to anchor the mind on properties that are usually co-instantiated with agency, and that appear as direct means for perceptually keeping track of agents (i.e. physical objects that instantiate agency properties).

For instance, humans can efficiently detect and track biological motions which are specific to agents. A tradition dating from the studies by G. Johansson (1973; 1975) showed how moving organisms can be detected purely from motion information. Johansson hypothesized that people would be able to perceive the movement of the human body from just the motion of the body’s joints. To test this hypothesis, he filmed an actor in the dark with small lights attached to his joints (ankles, knees, hips, shoulders, elbows, and wrists) so that nothing was visible except the lights. When the actor was seated motionless in a chair, observers perceived a meaningless configuration of points, rather like a constellation of stars; nobody perceived the lights to be connected to a human agent. But within fractions of a second after the actor began to move, first standing up and then walking, he was immediately and unmistakably perceived as a human agent in motion. In other further studies, other researchers have found that observers are able to discriminate between male and female walkers who have lights placed just on their ankles, knees, and hips (Cutting, Proffitt, & Kozlowski, 1978) – see also Troje (2002) and the simulation of this phenomenon at http://www.journalofvision.org/2/5/2/genderclass.html.

Also, there is evidence that certain types of motion lead to the attribution of specific types of intentional states (e.g. Gergely, Nadasdy, Czibra, & Biro, 1995; Heider & Simmel, 1944; Jacob & Jeannerod, 2003: pp. 222-6). Heider and Simmel (1944) showed human adults an animation on a screen involving three geometric objects: a small circle, a small triangle and a large triangle moving in the vicinity of a very large non-moving square. Unlike the perception of static display of the three geometric objects, the perception of the kinetic structure of the patterns of motion of the objects conveyed psychological and even social information about the objects. Subjects used highly intentional verbs to describe the behavior of the triangles such as they ‘chased’, ‘attacked’, ‘caressed’ or ‘comforted’ the circle. In the same tradition, in reply to Atran (1998), Todd & López (1998: p. 592) describe experiments similar to the famous previous ones: ‘We have explored a simple visual cue-based algorithm for judging intention from motion in just such instances (Blythe, Miller, & Todd, 1996). We had participants generate motions of two moving “bugs” on a computer screen, corresponding to simple intentional categories including pursuit, evasion, fighting, courtship, and play. Other participants were later able to categorize the intentions of the “bugs” with high accuracy from their trajectories alone. This study supports the notion that animate intention can be determined using only a few simple spatiotemporal cues (which include, from trajectory analysis, relative heading, relative distance, relative velocity, and vorticity or “loopiness”). Knowing the intention (as opposed to the general intentionality that Atran mentions) of another organism can trigger the appropriate domain-specific mechanism for response, including species-level categorization and recall of relevant traits.’

Moreover, in the visual domain, there is evidence that the visual system does not rely on the same resources for recognizing non-face objects as opposed to faces (e.g. Grill-Spector, Knouf, & Kanwisher, 2004). This might suggest that the presence of a face – an agent specific cue – can contribute to a specific file indexing, an agent-based indexing. This could be the case not only in vision, but also in auditory perception and cross-modal integration. It seems plausible, for instance, that noises that bear the specific signature of human or animal behavior trigger the opening of an agent file – i.e. we can detect and track a person just by hearing a sound that bears the specific signature of a human body (voice, footsteps). This is the case when one is detecting a sound related to the phonological part of the body (e.g. Handel, 1995).

Argument from parsimony for the direct access to agency 

The empirical data drawn from the study of early motor and sensory skills seem to support the idea that there may even be two distinct kinds of files. What would be the use of such a distinction from a computational perspective – as introduced by Ballard et al. (1997) and Pylyshyn (2001)?  It is often necessary to direct the focus of attention specifically toward agents, as opposed to mere non intentional objects, for instance in order to achieve collaborative actions (e.g. shared intentionality in word learning or collective sports), or to evaluate judgments involving intentional relations. If the deflation view is true, the only difference between object and agent files would be in terms of the descriptive content they encode. This would imply that there would have to be a (descriptive- or content-driven) search for intentional properties in object files, which seems computationally costly. As opposed to this approach, one can speculate that the addressing mechanism of each file was typed as object or as agent; however, such a search would be made easier from a computational perspective. For instance, one might speculate that files could be sorted by a non conceptual ‘tag’ signaling directly that it is a ‘(mere) object file’ or an ‘agent file’; such a tag would avoid the need to access the descriptive content of a file so as to target one of the two kinds of entity for any attentional or motor routine (e.g. routines such as ‘escaping from a predator’, ‘seeking for help’ or ‘searching for a team member’).

Rationality and agency 

Evidence for/against rationality is also relevant for tracking intentional agents, but not objects (which, since they are not intentional agents at all, are neither rational nor irrational). Thus, e.g., if someone consistently violates obvious norms of reasoning (e.g., modus ponens), it becomes difficult to even make sense of them as intentional agents [e.g.,  Dennett (1969; 1971), Davidson (1980; 1984) ]; though of course, it may be not difficult at all to continue to track their bodies. Similarly, when someone consistently exhibits markedly different patterns of behavior/reasoning in different domains (cf. discussions of  ‘compartmentalized thinking’, and/or ‘false consciousness’), even if these differences do not correlate to marked differences in features of their body, we sometimes find it necessary to suppose, in effect, that there are two agents there (less dramatically, that markedly different psychological states and traits characterize the agent in the two domains) even though there may be seamless continuity in his/her single body. This assumption is frequently made about people supposed to be ‘possessed’ as depicted in ghost movies or stories; it is manifested in a pathological way in syndromes such as the Capgras syndrome where the patient believes that his/her near ones have been replaced by substitutes occupying the same body. In short, considerations of a conspecific’s (ir-)rationality is a criterion relevant to the opening, maintenance, and updating of agent files, but it is not relevant at all to (non intentional) object files; and perceived discontinuities or dramatic differences in an agent’s intentional states themselves can lead us to ‘split’ or seriously amend the relevant agent file, even though the corresponding object file (used in tracking his/her body) remains intact, and does not undergo comparable changes in its contents.

7  Conclusions

Assuming a realist ontology of the material world and an object-dependent epistemology, we have examined the possibility of tracking not only physical objects but intentional agents. Our contention, referred to as the object-dependence view, is that tracking intentional agents requires the same kind of processes used in tracking physical objects. Furthermore, we postulated, in a manner consistent with the findings of experiments conducted on visual perception, that it is possible to draw a parallel between the classes of properties that characterize and produce agent and object files (indexing, preservation and encoded properties or cues). Given the similar – though not, we have argued, identical – anchoring, content and architecture of (non intentional) object files and agent files, the exact relation between them remains an open question which we have only begun to explore. However, at least the following seems plausible: (1) we perceptually and cognitively keep track of intentional agents, (2) object files and agent files are distinguishable in important ways, but nevertheless (3) agent tracking exploits the resources of object tracking by anchoring the agent file on perceptible features of agents’ bodies (thesis of body tracking).

8   Possible directions for future research

Our object-dependence view suggests directions for further research. A number of classical experiments have been carried out on object perception and multiple-object tracking in vision, in various conditions including conditions of target occlusions. It would be interesting to see what the results would be in analogous experiments in the case of visual multiple-agent tracking. What sorts of changes (in motion? behavior? intention?) are/are not relevant to subjects for maintaining the same agent file rather than opening a new one? Can one produce in an experimental display cases where persons-as-bodies and persons-as-agents ‘come apart’ (as when ‘the soul rises to heaven’, or ‘body-switching’ cases). Presently, the need first lies not in answering these questions but in raising them.

However, there is plenty of anecdotal evidence we could cite. For example, we have no difficulty in conceiving of ‘body-switching’ – an agent simply occupying a different body. For a ‘fictional-indexical’ descriptions of this, see Dennett (1978b); for a metaphysical conception, see the examples given by Parfit (1971) and Shoemaker (1997). This is not incompatible with our view. In fact, it might support it because the object-dependence can be exquisitely discovered in the ways one conceives of or depicts the ‘traveling souls’. For, depictions of souls leaving a particular body often take the form of ghostly bodies leaving the body of flesh-and-blood. This is not surprising, for how else could souls be depicted – or, we ask, how else could they be tracked? Is it not by depicting human bodies that painters have been representing the souls of ‘The Blessed’ in Paradise and of ‘The Damned’ in Hell? See for instance ‘The Last Judgement’ (1431) by Fra Angelico (http://www.abcgallery.com/A/angelico/angelico39.html, and the details of the blessed (http://www.abcgallery.com/A/angelico/angelico41.html) and the damned (http://www.abcgallery.com/A/angelico/angelico42.html) in the picture.

We do have some trouble making sense of cases of fission/fusion (where one agent becomes two, or two become one – a good illustration for this could be found in the view of Lewis (1976). Here too, the cases might support our position: fission/fusion are puzzling, not merely because they involve splitting/joining minds, but because our usual means of tracking agents (via their bodies) is frustrated – the one-body/one-soul correlation is disrupted.

Acknowledgments: We would like to thank for their comments on earlier versions of this paper Erika Marchetto, Nivedita Gangopadhyay and the members of the ‘agent tracking group’ of CNRS Thematic School ‘Reference to objects’ (http://www.objectreference.org/). Part of this research has been founded by the Enactive Network.