Past exam of the zoology course of the University of Cambridge / 2025 / Paper zm5 / 5

Non-human animals show evidence for several components of social-state tracking, but the evidence does not establish a uniform, human-like theory of mind. Theory of mind should be decomposed into sensitivity to goals, visual access, knowledge and belief, rather than treated as a single ability that an animal either possesses or lacks. The central evidential question is whether behaviour requires representing another's mental state, or can be explained by learned relationships among observable actions, contexts and outcomes.

Goal attribution has support from experiments distinguishing similar outcomes produced by different actions. Chimpanzees responded differently to a person who was unwilling to transfer food and one who was willing but unable: they persisted differently and left earlier in the unwilling condition. ^[ref] Holding the absence of food transfer broadly constant prevents the explanation that any unrewarded trial produces the same reaction. The contrast supports sensitivity to the organization or apparent purpose of actions. However, differences in movement and affect remain observable cues, so the finding need not imply a fully explicit concept of intention. Generalization to new obstacles and action sequences is a stronger test than repeating familiar signals.

Visual perspective taking concerns another individual having a different view. When dominant and subordinate chimpanzees competed for food, subordinates preferentially obtained food to which they had better visual access. Giving the subordinate a headstart forced its choice before the dominant's approach could reveal the answer, reducing a straightforward movement-following explanation. ^[ref] Such controls support tracking of another's visual access. They do not show that the subject represents every aspect of how the scene looks to the other animal; access to an object is a narrower problem than its appearance from a different viewpoint.

The distinction extends from seeing to knowing. An animal can potentially remember whether another witnessed hiding even after neither currently sees the food. Controlled competitive tasks found chimpanzees sensitive to knowledge versus ignorance, yet not successful in closely related false-belief conditions. ^[ref] Knowledge attribution therefore has stronger support in those tasks than an unrestricted ability to predict action from mistaken beliefs. Negative performance is not proof of incapacity: memory, inhibitory demands, motivation and task comprehension may limit the response. But a positive result for knowledge should not be relabelled a positive result for belief simply because both are social.

Bird studies provide evidence beyond primates. Ravens that had experienced a peephole altered cache-protection behaviour when a potential observer could be heard but not seen. The use of recorded sounds removed an actual competitor's gaze and movements as immediate guides, and the manipulation of available visual access made the test more informative than comparison of merely occupied and empty rooms. ^[ref] This supports generalization from access conditions to a possible observer. It remains possible to model performance through learned relations between openings, sounds and risk; novel transfer tests are needed to distinguish how abstractly the bird represents “being seen”.

Experience with one's own actions can also change responses to others. Scrub jays with previous pilfering experience showed stronger re-caching responses after having been observed caching. ^[ref] One interpretation is projection of the bird's own pilfering experience onto a potential competitor. Another is acquisition of a more general rule about cache vulnerability. The finding makes prior experience a causal candidate and suggests a relation between self and other, but it does not establish that the bird consciously imagines another individual's intentions. Observed and unobserved caching, prior experience and the opportunity to relocate food all need to be distinguished.

False-belief understanding sets a particularly demanding criterion: the subject must predict an agent using information that the subject knows is wrong. In great apes, an anticipatory looking paradigm found looks towards the location where an agent falsely believed an object remained, even when the subject had seen the object's removal. ^[ref] This bypasses some demands of an active food-choice response. Yet gaze towards a previously attended location can also arise from action prediction, attentional cues or associations with where an actor last saw an object. Anticipation is therefore evidence about prediction, whose representational basis needs further controls.

A subsequent experiment manipulated the ape's own experience of a barrier that looked the same from a distance but was either opaque or translucent. All subjects then watched the same test actions. Those with the opaque experience were more likely to anticipate the actor searching according to the outdated location. ^[ref] Because the visible test behaviour was held constant, an explanation relying only on a difference in the actor's immediate behaviour was weakened. The result supports transfer from self-experience to another's informational situation. It still leaves questions about the breadth of the represented belief: a limited model of visual registration and action expectation may succeed without the flexible propositional reasoning characteristic of human language.

Taken together, positive anticipatory-looking results and negative active-choice results need not be forced into a single yes/no answer. The tasks measure different outputs and impose different demands. Stronger evidence would show the same individuals generalizing across novel agents, objects and routes of access, distinguishing true belief, false belief and ignorance while physical events, attention and rewards are controlled. Eye-tracking analyses should specify regions and timing in advance, report the subjects who produce no usable anticipatory look, and avoid inferring population-wide competence solely from a selected subset of trials. Replication and uncertainty around effects matter especially with small captive samples.

The relation to self-representation is similarly componential. Using one's experience of a transparent barrier to predict another's access suggests some correspondence between self and other. It does not demonstrate an autobiographical self or reflection on one's own beliefs. Metacognition, such as selective information seeking when knowledge is missing, concerns monitoring one's own processing and is not equivalent to attributing knowledge to somebody else. Mirror self-recognition concerns one's body and is neither a necessary consequence nor a sufficient demonstration of belief attribution. These distinctions prevent a visually based test from becoming a universal scale of mentality.

The defensible conclusion is a mosaic: animals can track goals and access, some experiments support knowledge attribution and belief-sensitive predictions, and the most elaborate interpretation remains contested. The experiments constrain possible representations of self and others; their value lies in progressively excluding alternatives, rather than naming a mental state after any apparently intelligent social response.