Computational Models of Human Intuition: What Does the Evidence Support?

This exploratory position paper from Artificial Intuition Research (AIR) examines what existing research can tell us about computational models of human intuition. It reviews three examples: an artificial agent inspired by the recognition-primed decision model, a semantic network designed to identify connections between concepts, and studies comparing language models with human responses on reasoning tasks. The paper examines what these approaches demonstrate and where their limitations lie. A model may reproduce aspects of a human decision process, identify useful connections, or produce responses similar to those of people. However, these results do not automatically establish that the system uses the same process as a human decision-maker. The paper outlines ways to evaluate such claims more carefully, including response and error patterns, confidence calibration, and evidence about computational processes. It also proposes a focused engineering study to test whether adding a warning or gating component can improve decision-making beyond a simpler uncertainty threshold or a system without that component. The aim is to distinguish demonstrated capabilities from interpretations and open questions. The paper does not claim that any of the approaches reviewed has established a computational model of human intuition. Instead, it identifies practical questions for further investigation and proposes ways to test them. This is an exploratory research output developed through AIR and has not undergone formal journal peer review.