Emotion prediction as computation over a generative theory of mind

Abstract

From sparse descriptions of events, observers can make systematic and nuanced predictions of what emotions the people involved will experience. We propose a formal model of emotion prediction in the context of a public high-stakes social dilemma. This model uses inverse planning to infer a person’s beliefs and preferences, including social preferences for equity and for maintaining a good reputation. The model then combines these inferred mental contents with the event to compute ‘appraisals’: whether the situation conformed to the expectations and fulfilled the preferences. We learn functions mapping computed appraisals to emotion labels, allowing the model to match human observers’ quantitative predictions of 20 emotions, including joy, relief, guilt and envy. Model comparison indicates that inferred monetary preferences are not sufficient to explain observers’ emotion predictions; inferred social preferences are factored into predictions for nearly every emotion. Human observers and the model both use minimal individualizing information to adjust predictions of how different people will respond to the same event. Thus, our framework integrates inverse planning, event appraisals and emotion concepts in a single computational model to reverse-engineer people’s intuitive theory of emotions. This article is part of a discussion meeting issue ‘Cognitive artificial intelligence’.

How could we make a social robot? A virtual bargaining approach

“…Most superficially, but importantly from a practical point of view, computers are increasing able to classify human facial expressions 2 and interpret human actions (e.g. [16]), and the ability to simulate human faces and movements, both in hardware and software, has become increasing impressive. Moreover, remarkable progress in speech recognition and production, as well as more elementary natural language processing in script-like contexts (e.g.…”

“…But both might expect that the signal will actually be used for deception. 16 Thus, computing the meaning of a signal rests on a hypothetical commitment to mutual cooperation through communication (after all, communication is a form of joint action, [4,59]). But it does not require actually believing that the other party will cooperate (although no doubt such cooperation is the typical case).…”

“…). 16 Indeed, where the players have conflicting interests, the equilibrium strategies will be for the sender to choose a signal at random (because any signal that depends on the configuration of bananas and scorpions risks providing useful information to the receiver) and the receiver of the signal should ignore it completely. This is the problem of 'cheap talk' in economics[57].…”

Introduction to ‘Cognitive artificial intelligence’

“…A particularly critical aspect of the challenge of building computational models of other minds—inferring emotional states—is taken up by Houlihan et al . [ 9 ] in their paper ‘Emotion prediction as inference over a generative theory of mind'. They describe a computational model of emotion prediction, the Inferred Appraisals model, that uses inverse planning to infer mental states, which can include individual objectives but also ‘social preferences' such as preference for equity or the desire to maintain a good reputation in the eyes of others.…”

Calculated Feelings: How Children Use Probability to Infer Emotions

“…Intuitive theories specify causal relations involving abstract constructs, allowing people to predict and explain outcomes (e.g., Gopnik & Meltzoff, 1997 ; Gopnik & Wellman, 2012 ). For example, children’s early theory of emotions may specify that having desires fulfilled causes people to be happy (for in-depth discussions of the intuitive theory of emotions see Anzellotti et al, 2021 ; Houlihan et al, 2023 ; Ong et al, 2019 ). If children were limited to only scripts, they would expect any given sequence of events to always end with the same emotional reaction.…”