AI and theory of mind

Theory of mind is considered a key component of human cognition. Our mind can understand the psychological state of others. It is also referred to as understanding and accepting other people’s minds. And now scientists are actively working to endow machines with this skill. To do this, computer scientists are actively collaborating with experts in psychology, psychiatry, and neurobiology. And the results could be very promising.

The introduction of “theory of mind” will help to implement very complex functions in AI systems, allowing deployment of these systems in high-risk and critical, potentially dangerous situations. For example, no one is surprised by drone vehicles these days. Artificial intelligence can identify the actions of other road users fairly accurately. But at the same time, it cannot yet take into account the fact that a person may change their mind very sharply due to personal thoughts and motivation. For example, a child walking peacefully on the pavement may react to a ball, a kitten, or an ice cream van and abruptly change its trajectory to a completely illogical one — rushing across the road. And no predictive system which uses only past data gathered can react to such a situation accurately and with a high level of reliability. It is easier for the human brain to predict the behavior of others, given a particular person’s psycho type and environment (“baby-kitten”, “ice-cream baby”, etc.).

So the need to teach artificial intelligence “theory of mind” is a very relevant task. But it is not an easy task to accomplish. Now there are many datasets available for AI training that help assess observable human behavior. That said, a benchmark has not yet been developed to assess the “theory of mind” abilities of AI. Perhaps data on emotional state, desires, intentions, and other mental state features should be used for training. It is likely that the implementation of the “theory of mind” will require considerable specialist effort.