Human–AI Comparative Science, AI Futures, and the Functional Definition of Machine Intuition

This paper develops a comparative framework for human and artificial intelligence around three ques- tions. What can human–AI comparison reveal about the functional organization of cognition? How might a mechanism identified in one system generate a hypothesis about the other? And how should such comparisons inform work on possible and desirable AI futures? We separate predictive AI futures, concerning what AI systems may become under specified technical and institutional conditions, from normative AI futures, concerning what forms of development humans should encourage, constrain, or design for. The paper then develops a mechanism-level account of machine intuition, drawing on the discovery- oriented conception of intuition due to Bowers et al. Rather than treat intuition as rapid recognition of familiar patterns, we define machine intuition functionally as discovery-oriented pre-verification valua- tion: the allocation of scarce verification resources toward candidates whose value is not yet established. We propose a computational realization using surrogate valuation, uncertainty-sensitive candidate se- lection, and an intermediate relational-distance constraint taken from the companion paper’s matching- interval hypothesis. Machine intuition is a functional term here. It carries no claim that AI systems have subjective experience or a human-like intuitive faculty. We also outline a route for transferring mechanisms from AI systems to hypotheses about human functional disorders. An AI abnormality, characterized independently of any hypothesis about humans, is subjected to an intervention native to the AI system; the mechanism of that intervention is then iden- tified; and a cross-system correspondence test decides whether the mechanism has a plausible human counterpart. Behavioral similarity alone does not license the transfer. We also keep failure to support a hypothesis distinct from its positive disconfirmation throughout. Together with the companion pa- per, these three contributions form one research program connecting functional comparison, conditional mechanism transfer, AI futures, and the computational realization of discovery-oriented evaluation.