Insight, Illumination, and AI

Much of our education and training is directed toward cultivating skill and expertise in tasks that require sustained procedural thinking. Yet many real-world problems seem to require more than just the ready deployment of procedural thinking. These tasks require insight, which can serve to re-frame problems—or even illumination, the strongest cases of which offer a complete solution without intermediate proof or justification. Although this dimension has been neglected, illumination can also be a property of the solution itself: An inspired result can illuminate a whole region of a problem space, transforming our understanding of how the space itself is structured.

In this working group we will investigate Insight and Illumination (I&I) through the lens of artificial intelligence. We aim to: (1) better understand the mechanisms and conditions of I&I (illuminated by the capacities and limitations of AI systems); (2) ask how AI might amplify I&I in collaboration with human researchers; and (3) examine what the current AI moment reveals — and obscures — about the laborious cultivation of skill that may itself be a precondition for insight, and whose role is thrown newly into question as AI systems automate more of the procedural labor that has traditionally prepared the mind. Most fundamentally, we hope to contribute to the longstanding SFI interest in the phenomenon (and phenomenology) of understanding, with insight and illumination representing minor and major manifestations of this epistemic state.