Reading wild minds: A computational assay of Theory of Mind sophistication across seven primate species

Abstract

Theory of Mind (ToM), i.e. the ability to understand others' mental states, endows humans with highly adaptive social skills such as teaching or deceiving. Candidate evolutionary explanations have been proposed for the unique sophistication of human ToM among primates. For example, the Machiavellian intelligence hypothesis states that the increasing complexity of social networks may have induced a demand for sophisticated ToM. This type of scenario ignores neurocognitive constraints that may eventually be crucial limiting factors for ToM evolution. In contradistinction, the cognitive scaffolding hypothesis asserts that a species' opportunity to develop sophisticated ToM is mostly determined by its general cognitive capacity (on which ToM is scaffolded). However, the actual relationships between ToM sophistication and either brain volume (a proxy for general cognitive capacity) or social group size (a proxy for social network complexity) are unclear. Here, we let 39 individuals sampled from seven non-human primate species (lemurs, macaques, mangabeys, orangutans, gorillas and chimpanzees) engage in simple dyadic games against artificial ToM players (via a familiar human caregiver). Using computational analyses of primates' choice sequences, we found that the probability of exhibiting a ToM-compatible learning style is mainly driven by species' brain volume (rather than by social group size). Moreover, primates' social cognitive sophistication culminates in a precursor form of ToM, which still falls short of human fully-developed ToM abilities.

Fig 1. Sociobiological features of tested non-human primates species.

On this graph, the social group size (x-axis) and ECV (y-axis) are shown for each species on a log-log scale. Note that reported species' group sizes exhibit substantial variability across ethological field studies. In this work, we have chosen to rely on the average group size from a series of more than a hundred published studies (ensuing standard deviations are depicted as horizontal bars on the graph). We refer the interested reader to S2 Text for more details. Overall, there is no significant statistical correlation between group size and ECV (r = -0.37, p = 0.41) across these species. The position of the human species is shown for comparison purposes. The graphical inset also shows the relationship between group size and ECV, this time across the 130 primate species reported in [91]. Species investigated in this work are depicted in red.

Fig 2. Experimental protocol showing the three basic phases of the game.

A: the experimenter hiding the food in one hand out of individual’s view (inside the brown paper box visible in two following pictures), B: the individual choosing one hand by pointing or touching it, C: the individual getting the food reward if choosing the correct hand. Photo credits C. Trapanese.

Fig 3. Behavioural performance results.

A: net rate of correct answer (y-axis) is shown as a function of species (x-axis) and opponent condition (RB: blue, 0-ToM: green and 1-ToM: red). Errorbars depict standard error of the mean. B: condition-specific performance pattern averaged across species. C: simulated performance pattern of 0-ToM. D: simulated performance pattern of cooperative 1-ToM. E: performance pattern of human adults [49]. Graphs B to E use the same colour coding as in A.

Fig 4. Volterra decomposition of primates’ trial-by-trial choices sequences.

The magnitude of Volterra kernels (y-axis) is plotted as a function of species (x-axis) and opponent condition (same colour coding as in Fig 3). A: weight of the opponent’s actions (imitative tendency). B: weight of one’s own actions (perseverative tendency).

Fig 5. Volterra decompositions of non-mentalizing and cooperative mentalizing artificial agents, as well as human adults [49] performing the same task.

A: competitive 0-ToM, B: cooperative 1-ToM, C: human adults. Same format as Fig 4.

Fig 6. Bayesian model comparison results.

A: the average probability pToM of exhibiting a ToM-compatible learning style (±standard error) is plotted for each species, in ascending order. The red dotted line corresponds to chance discrimination (pToM = 0.5). Note that orangutans' ToM sophistication reaches pToM = 0.51±0.14 if we exclude one individual that shows characteristic signs of Down syndrome (see Discussion section). B: the probability of exhibiting a ToM-compatible learning style (pToM, y-axis) is plotted as a function of group size (x-axis). The red plain line indicates the best-fitting linear regression, and the red dotted lines depict the corresponding 95% confidence interval. C: the probability of exhibiting a ToM-compatible learning style (pToM, y-axis) is plotted as a function of ECV (x-axis).

Fig 7. Estimated frequencies of learning styles.

The posterior mean of model frequency (y-axis) is shown for each learning style (x-axis), among species with large brains (A) and small brains (B). Note that the median-split on ECV actually separates apes from prosimians and monkeys, which is consistent with primates’ phylogenic relationships (see S1 Text). The colour code indicates the type of learning style (blue: no-ToM, red: competitive ToM, violet: cooperative ToM). Errorbars indicate posterior standard deviations. For comparison purposes, chance level (10%) is indicated using black dotted lines.