Entropy (Basel, Switzerland)
March 25, 2025
Chris Percy, Andrés Gómez-Emilsson
2 citations
Neuroscientific theories of consciousness must address how separate micro-units of information combine into a single, unified conscious experience—the phenomenal binding problem. This paper examines how Integrated Information Theory (IIT) v4.0 offers a solution by proposing that particular entities called 'complexes' define existence. While this works in a static framework, it creates difficulties when applied to dynamic systems. The authors identify a dilemma for IIT: non-local entity transitions versus contiguous selves, termed the 'dynamic entity evolution problem.' Three potential ways IIT could dissolve this dilemma are described. The paper contributes to IIT's shift from static to dynamic analysis.
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.
arXiv Preprint Archive
January 22, 2026
Derek Shiller, Laura Duffy, Arvo Muñoz Morán et al.
The evidence against large language models (LLMs) from 2024 being conscious is not decisive, though it is stronger than the evidence against consciousness in simpler AI systems. The Digital Consciousness Model (DCM) provides a systematic, probabilistic framework for assessing consciousness in AI, incorporating multiple leading theories rather than a single one. It allows comparison across different AIs and biological organisms and tracks how evidence evolves as AI develops. The DCM's initial results show that while current LLMs likely lack consciousness, the case against them is far from settled.
Journal of Consciousness Studies
February 1, 2025
Chris Percy
Dissatisfaction with traditional methods like logical deduction and experimental falsification for evaluating theories of consciousness has led to exploration of alternatives, including a method termed 'listed requirements.' A structured literature search and critical review identified five candidate lists, which are a promising but insufficient start. The longest list contains 11 items, but across the five lists 19 unique items appear, and taxonomic analysis surfaces at least 30 potential candidates. Four limitations of the method are discussed, arguing it is best used as one tool within a broader assessment strategy. The conclusion outlines a workplan for a sufficiently complete working taxonomy.