Generative AI as a transformational logic for cognitive neuroscience

Abstract Cognitive neuroscience faces a paradox: neural data are abundant, yet conceptual synthesis has stalled because dominant contrast-based approaches show where activity differs but not how cognitive operations relate or transform. Here, we propose a generative-transformational logic grounded in AI and neural geometry, treating cognition as lawful mappings among neural states. Generative models can learn latent transformations linking states across tasks, contexts, and individuals. Because transformation success is testable, this framework enables counterfactual simulation and connects data-driven modeling with theory-driven inference. It moves cognitive neuroscience from mapping correlates toward algorithmic explanations of how the brain generates and reorganizes cognition over time.