What do we know about the contents of the cognitions of other agents or conspecifics, either immediately present, or ones we have interacted with in the recent past in our environment? One important suggestion deriving from philosophy and cognitive science is that we and others engage in mental simulation. Mental simulation refers to the capacity to represent the likely intentions or likely set of behaviours that another person may have engaged in. This is an elastic concept, for it embraces classic cognitive science concepts, such as representation and inference, but it does so specifically with reference to the behaviour of another. If we take a situation involving conflict, for example, we can infer from the tempo, tone, and prosody of the speech of another, their state of mind and the suite of intentions that they are likely to form. But we go further than that. What we intend to do is based in part on the behaviours of the other, in order to infer what the other person is likely to do next. There are many other elements involved, including components involving signalling from the other’s behaviour, which are useful for making predictions about what it is they are likely to do. But simulation goes further than this. It requires the observer to make an estimate of what it is that the observed is likely to be thinking, and to behave as if you have reasonable knowledge about what it is that the other is doing.
“I believe that everything important in psychology (except perhaps such matters as the building up of a super-ego, that is everything save such matters as involved society and words) can be investigated in essence through the continued experimental and theoretical analysis of the determiners of rat behavior at a choice point in the maze.”
A T-maze is very simple: the animal is placed in a start box; it makes an excursion down an alley, whereupon it arrives at a choice point, and when it gets to the choice point it must go to the left or go to the right in order to receive food. Rats are very good at this task, and will rapidly learn to alternate between the differing arms in the T-maze, knowing that they have been in the previous arm before and that they have consumed the food that was present in that arm. Indeed, it is reasonable to assert that a substantial fraction of experimental behavioural neuroscience is based on these paradigms, where we observe a single animal over time, and we make inferences about what it is that they are likely to ‘know’, given the contingencies that they have been exposed to. We have learned an enormous amount about the brain systems that support these kinds of behaviours. We know, for example, in many studies using tasks such as the T-maze, radial-arm maze, the
plus maze and the water maze, that there are brain systems concerned with representing position in space (for example, refuges, or the location of food); that there are other brain systems that are involved in making choices over time (for example, delayed memory tasks). We know in other experiments the brain systems that are involved when an animal becomes addicted to a substance, or indeed when an animal continues to make what are now non-rewarded choices (so-called “perseverative” behaviour).
Experimental behavioural neuroscience focuses on the analysis of individual animals, summed and segregated into differing experimental groups that differ along some dimension or set of dimensions of interest.
Experiments such as these, while they have been enormously valuable, and will continue to be so, fail to answer some important questions. This can best be illustrated using the example of the T-maze again. Imagine for a moment that instead of placing one animal in the start arm of the T-maze and then freeing it in order to make a choice, that we instead place two animals (a leader and a follower) in the start arm of the T-maze, separated by a glass partition. The leader is released, runs down to the choice point of the T-maze, makes a choice, and gets the food. The question then is: what does the follower do? Will the follower take a predetermined course of action, and let its behaviour be guided by what it has done previously, or will it behave according to what it “believes”, “guesses” or “simulates” the leader having done. In order to do this, the follower must estimate what it believes the leader has done, or at least what it would have done in the leader’s place, that is, it would have eaten the food available to it, and it must infer what it is the leader is likely to have done. Remember, the key thing here is that the follower cannot see what the leader has done. The follower can only estimate what it is likely that the leader would have done, and it can only do that on the basis of what it would have done, had it been similarly placed. So this is metacognition in a simple form: ‘I must behave according to what I believe the other is likely to have done’.
This is a good example of social inference within a social context: in particular, inferences regarding the likely behaviour of other conspecifics. Social interaction offers a useful exploratory interface between understanding spatial cognition, especially allocentric and egocentric cognition, and social cognition involving actions generated by the self, actions generated by others, and understanding how the two of these integrate together. Understanding the direct consequences of interacting conspecifics (where conspecifics investigate each other, engaging grooming, sniffing and other socially motivated behaviours, and how the perception of inferred action might occur, especially in the context where behaviour of one conspecific precedes the behaviour of another and therefore leads to choices that may be different in time (inter-temporal choice) given the behaviour of the leader, and then the follower. This perspective in gives rise to lots of interesting questions: How does knowledge of the actions of others control our own behaviour? How does understanding the computations performed in brain networks during dynamic social interactions allow us to control our own behaviour? What are the neural substrates that underpin the perception of inferred action?**
*Important NB: I anthropomorphise deliberately throughout this piece.
**Writing the above led me to this super paper, by Michael Brown which reviews in great depth some of the same issues, but greatly extends the range to consider the tension between social affiliative and metabolic (hunger) drives.
Social Influences on Rat Spatial Choice by Michael F. Brown
Although there is abundant evidence for social learning and other forms of social influence on behavior, relatively little experimental analysis of the mechanisms involved is available. The present paper reviews a line of research examining social influences on spatial choice in the context of spatial working memory paradigms using pairs of laboratory rats foraging together for food. There is a social affiliation effect – rats are attracted to spatial locations if a familiar conspecific is there. However, there is a countervailing tendency to avoid visits to spatial locations that were previously depleted of food by the other rat. The latter effect is based on working memory for the choices made previously by the other rat. The memories for the previous choices of another rat can affect subsequent choices flexibly, either increasing or decreasing choice tendencies depending on working memory for the contents of spatial locations resulting from the rat’s own visits to the location.
Keywords: social memory, social learning, working memory, spatial memory
Brown, M. F. (2011). Social Influences on Rat Spatial Choice. Comparative Cognition & Behavior Reviews, 6, 5-23. [publication listing]