Characterizing the neuroarchitecture of human Theory of Mind, particularly as applied to interaction with AI teammates
Abstract
As the capabilities of AI agents rapidly advance, it becomes increasingly important to find ways that enable such agents to effectively team with human participants. Such teaming is relatively new, so it is important to determine how to construct situations in which human and AI agents work together effectively.Much of the interaction depends on how human participants perceive the qualities of the AI agent: do they trust the AI agent#s technical competence? Do they sense something like irrationality or emotionality in theAI agent? Whatever the human participant thinks of their AI teammate (or, for that matter, a human teammate), it is presumed to be mentally represented in the parts of their mind and brain referred to as their Theory of (another entity#s) Mind, or ToM, network. This project aims to determine the contents of human participants# ToM as they interact with AI agents or human participants.The research will use two recent, powerful methods to determine the content of humanparticipants# minds, particularly their ToM. One method is to use fMRI, functional magnetic resonance imaging, to measure their brain activity as they interact with a teammate. The imagingoccurs at a rate of 1 Hz and a resolution of 3 mm3, providing adequate data to assess mental content. This approach has repeatedly previously shown that the ToM system is activated when a human participant engages with another person.A second method is to apply machine learning to the brain activation data to decode particular mental states. For example, this method has previously been used to decode thoughts of particular physical objects, scientific concepts, and emotional states. Here, the goal will be to decode various ToM representations of an AI or human teammate.These methods will be applied in several projects in which various types of tasks are performed by a human participant interacting with an AI agent or with another human participant, and various properties of the teammate may be relevant. The overarching goal of the project is to provide information about human-AI agent interaction that can inform the design and organization of human-AI teams.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 24, 2023
- Source ID
- N000142312682
Entities
People
- Marcel Just
Organizations
- Carnegie Mellon University
- Office of Naval Research
- United States Navy