​Social Intelligence in Humans versus Machines: A New Framework for Understanding Imitation, Intent, and Theory of Mind

The human mind is vastly complicated. Even with major scientific advances, our simplest intelligent actions remain difficult to model computationally.

Among these is the fascinating question of how we come to understand that other beings think and feel like us—what is known as the “theory of mind.” Rajesh Rao’s project pairs recent advances in computer science and neural network theories with the results of studies on human qualities, exploring social intelligence in an innovative framework.

We have the remarkable ability to learn from social interactions with fellow humans. In other words, we have social intelligence: the ability to learn by watching, inferring intent, and imitating others. This ability leads to the development of a “theory of mind” in humans. Current approaches to machine intelligence lack such an ability. This project will develop a new interdisciplinary framework for social intelligence based on computational models and empirical investigations. Specific aims of the project include:

1. Empirical and Computational Framework for Causal Learning, Intent Inference and Imitation: Based on empirical data from developmental science, the team will develop a computational framework for causal learning, intent inference, and imitation based on Bayesian inference in probabilistic models.
2. Neural Implementation in Sum–Product Networks: We will implement the probabilistic models above using sum–product networks, a new type of neural networks that emulate operations performed by dendrites of biological neurons while implementing efficient Bayesian inference. Results will be compared to neurophysiological and brain imaging data.
3. Theory of Mind and Social Intelligence in Humans vs. Machines: Building on (1) and (2), the team will investigate a probabilistic model of how theory of mind develops in humans. They will explore the implications of embodied human intelligence vs. disembodied machine intelligence in the development of theory of mind and social intelligence.

The outputs of the project will include presentations, research papers and popular science articles elucidating the differences between human and machine intelligence in social environments. This research will open the door to studying the relationship of intelligence to human empathy, beliefs, ethics, and other qualities in an empirically-informed computational framework.