AI offers cognitive science naturalistic models for generalizable theories

A new paper on arXiv (2502.20349) argues that progress in AI provides cognitive science with opportunities to adopt naturalistic stimuli, tasks, and behaviors in experiments, alongside computational models that handle them. The authors review research across neuroscience, cognitive science, and AI showing that naturalistic paradigms can elicit distinct behaviors and engage different processes than traditional lab tasks. They contend that learning from naturalistic data yields qualitatively different outcomes, potentially resolving aspects of natural intelligence and ensuring theories generalize beyond controlled settings.