Will AI De-Skill Our Social Cognition? The Risks in the Delicate Interplay Between Artificial and Natural Theory of Mind

Theory of Mind (ToM) is a crucial component of human cognition and one of its distinguishing features. The development of Large Language Models capable of successfully performing ToM tests at levels similar or superior to those reached by humans has sparked a fierce debate and great interest in understanding how they do so. At the same time the Artificial Intelligence (AI) industry is actively developing systems able to display human-like skills to better interact with humans. These systems routinely engage with millions of people, providing advice on interpersonal relations and mental health issues and serving as virtual companions in a myriad of roles. This process is driven by the developing ability to simulate interactions that reflect the social cognition processes that drive human behaviour, including ToM. Here we review the evidence and critically evaluate the concept of ToM in artificial and natural intelligence. We discuss the psychological and neural implications for human users interacting with an increasing number of “ToM-equipped” AI systems and potentially a decreasing number of other humans. We argue that, without strong regulations and societal education across the world, there is a real risk that AI will effectively de-skill human social cognition.