MIT has launched a new interdisciplinary course that invites students to grapple with deep philosophical questions about artificial intelligence and rationality, rather than offering definitive answers about them. The class — titled AI and Rationality — is co-taught by computer science professor Leslie Kaelbling and philosophy professor Brian Hedden, and is designed to help future researchers and professionals think critically about what it means for an artificial system to be rational, autonomous, and capable of decision-making.
The course explores the intersection of computer science and philosophy, fields that have long overlapped in formalizing concepts like belief, learning from experience, and rational action. In discussing these ideas, students consider questions such as how rational agency might be ascribed to machines, how different disciplines define rationality, and whether traditional logical and mathematical frameworks adequately capture the nature of rational thought — in humans or in AI systems.
Rather than teaching a fixed body of doctrine, instructors emphasize developing foundational tools for critical thinking. They want students to reflect on their assumptions about AI as they progress into careers in research, industry or government, preparing them to confront challenges and uncertainties that cannot be predicted today. This includes reexamining notions like logical consistency, the limits of mathematical models, and the subjective views humans bring to questions of reasoning.
Students from diverse academic backgrounds are engaging with the material, discovering that definitions of rationality can vary widely between disciplines. Some, for example, find that human behavior often defies strict logical or mathematical standards, prompting them to question whether human irrationality or machine behavior challenges conventional ideas about what it means to think or decide rationally. The collaborative nature of the class highlights the value of blending philosophical inquiry with technical understanding in the evolving field of AI.
