Can AI develop a unified theory of consciousness solely from neural data without human input ?

Some researchers claim that deep learning models, trained on vast amounts of brain data, could one day reverse-engineer consciousness itself. If successful, this would redefine what it means to be human and grant AI unprecedented authority over ethical questions. Opponents argue consciousness cannot be reduced to data patterns or algorithmic processes. The implications for personhood, moral consideration, and existential risk are profound.

Current AI approaches to consciousness rely on either proxy tasks (e.g., predicting neural activity or behavior) or architectures framed by existing theories (global workspace, predictive coding), but none has produced an empirically grounded, unified theory of consciousness from neural data alone without human-specified theory. The key obstacles are the absence of an agreed neural correlate of consciousness and the lack of gold-standard datasets that unambiguously label conscious vs. non-conscious neural patterns. Some groups use large-scale neural recordings plus machine-learning classifiers to infer states of awareness, yet these remain correlational and theory-laden rather than generative theories. As of 2024, no AI has synthesized a standalone, falsifiable theory of consciousness purely from data without importing human theoretical commitments.

— Enriched May 10, 2026 · Source: best-effort summary, no public reference