The phenomenal binding problem for neural networks.
Consciousness and cognition February 3, 2026 Chris Percy, Gautam Agarwal
A deliberately simple artificial neural network model can implement functional binding—combining micro-units of information for cognitive tasks—but fails to achieve phenomenal binding, the integration of micro-information into the unified, macro-scale conscious experience typical of human phenomenology. The model's failure highlights a key challenge for theories of consciousness: maintaining a distinction between unconscious and conscious processing while achieving phenomenal binding. Several established theories, including Integrated Information Theory, Orch-OR, and Conscious Electromagnetic Information Theory, map onto possible solution structures based on which parts of the model they elaborate or reject. Each proposed solution requires further development to fully account for phenomenal binding.