The Question
Imagine you receive this message from a friend: “It finally happened, I got the confirmation letter in the mail this morning, let’s go out tonight to debrief.” As soon as you read this, your brain starts building a complex model: You might immediately know that your friend is talking about a job offer they have been waiting for, for over a year, and that they will be both excited and nervous, because the job means moving countries and their partner doesn’t want that; you might immediately think about whether you can move your 6pm yoga class tonight to go and meet your friend, or whether it might make sense for you to pick them up from home on your way back from work, or whether they are working in their office today.
When we talk with our friends, listen to the news or make plans with colleagues, our brains make sense of these complex situations by using background knowledge to create rich and powerful inferences about the mental states of other people. We are constantly reasoning about the things they know, their likely goals, whether to trust another person, whether to change our minds and when to make new plans together or seek new information.
This kind of social reasoning is central to human life and vital to the buildup of knowledge and cohesion in social discourse. Without reasoning about other minds, our ability to learn and share information would be critically limited, yet these core cognitive processes go beyond what contemporary models in modern cognitive science and neuroscience are capable of capturing. Even the most powerful modern artificial intelligence systems lack many aspects of social reasoning, and we currently lack models and methods that would help us to address this fundamental limitation.
The Approach
The project How Reasoning About Other Minds Supports Human Culture: Large-Scale Experimental Simulations of Cultural Evolution With Human Participants aims to fill this gap by conducting large-scale behavioral experiments in which people engage in social interactions with strangers to solve problems, make plans and discover new concepts together. These experiments will leverage new methods in experimental psychology to create situations that expose the rich structure of human social reasoning, and leverage computational methods to extract structure from large-scale behavioral data. Using modern experimentation, the project’s researchers will be able to construct networks of interacting participants who learn from each other over time, and study the evolution of participant behavior in micro-societies.
The experiments will generate large-scale datasets that can be used to test the predictions of formal computational models of social reasoning, and to improve machine learning systems that currently lack many forms of social reasoning. These insights will not only help us understand one of humanity’s key cognitive skills, but will also help to inform the design of digital systems that better support constructive discourse and learning in modern digital discourse arenas, such as social media platforms and conversational AI systems.
The How Reasoning About Other Minds Supports Human Culture project is being led by Bill Thompson at the University of California, Berkeley (USA).
