Understanding Social Cognition with Delta-Rule Active Inference Learning

Trait, attitude and preference learning encompasses the encoding of stable characteristics from observed behaviors, which are then used to make predictions and influence how one interacts with an individual in various contexts. Being able to understand the risk-taking tendencies of others is a complex example of such social inference job. In this research, we used a Bayesian framework to explore how humans can gauge another person's risk-taking tendency. We used a sequential scenario where an observer watched the other person's choices between a high-risk gamble and a guaranteed smaller reward. We proposed an approximate Bayesian observer to assess an agent's risk attitude. This learner utilizes a probabilistic generative model to model the decision making process of others and then employs a variational Bayesian method to invert the generative model. Our research adds to the accumulating evidence that inverting generative models are an essential computing tool for understanding social behavior. The learner updates the posterior estimation of the other's risk attitude on a trial-by-trial basis, with the discrepancy between the model predictions and the choices observed followed the widely accepted prediction error framework namely delta rule. By combining the algorithmic advantages of delta rule with the computational advantage of Bayesian framework, we fashioned a more effective and comprehensible learner. We showed that the accumulated uncertainty the observer builds up while predicting the agent's choices is the main factor that determines inferential uncertainty, i.e., the uncertainty the learner has in estimation of the agent's hidden attitude, which is the impetus for learning and discovery. The model begins with a high learning rate and gradually reduces it as more trials take place. This reflects the way humans learn from examples sequentially, exploring and then making use of the information as they progress. As more data is accumulated from someone who follows a consistent decision-making process, the observer's faith in his own judgments should grow, and he should be less affected by each new observation.