Consciousness
Why and how brain activity gives rise to subjective experience is still unexplained.
Open in the interactive tree →We can correlate brain states with reports of experience, but nobody knows why any physical process feels like something (the 'hard problem'). Rival theories, such as global workspace, integrated information, higher-order and predictive-processing views, disagree about which systems are conscious, including animals, patients and AI.
As of October 2026
The largest test so far, an adversarial collaboration between supporters of integrated information theory (IIT) and global neuronal workspace theory (256 participants, funded with over 6 million US dollars), was published in Nature on 30 April 2025; it tested pre-registered predictions of both theories and substantially challenged key tenets of both without settling the question. In September 2023, 124 researchers signed a letter calling IIT pseudoscience; in a later survey of consciousness researchers only 8 percent fully agreed with that label. No accepted test exists to tell whether a coma patient, an animal or an AI system is conscious.
What is missing
- A theory accepted across the field that explains why experience exists, not just its brain correlates
- Objective measures of consciousness that work without verbal report (coma patients, infants, animals)
- Experiments able to distinguish rival theories, since their predictions often overlap
- Recording and perturbing whole-brain activity at cellular resolution in conscious subjects
- Agreement on what counts as evidence of machine consciousness
Becomes possible once solved
- Reliable tests for awareness in coma and under anesthesia
- Evidence-based ethical rules for animals and AI
- Understanding of disorders of consciousness and perception
Open steps
- Consciousness without speech High AI leverageObjective measures of awareness for coma patients, infants and animals that work without verbal report.
- Experiments that separate theories Medium AI leverageDesign experiments in which rival theories predict clearly different outcomes, since many of their predictions overlap today.
- Whole-brain cellular recording Low AI leverageRecord and perturb whole-brain activity at cellular resolution in conscious subjects.
- Criteria for machine consciousness Low AI leverageAgree what would count as evidence that an AI system is conscious, given that the theories differ on what matters.
- A theory of why experience exists Low AI leverageA theory accepted across the field that explains why experience exists, not just its brain correlates.
Where AI could help
Low AI leverage. The gap is a missing theory and experiments that separate rival theories; AI offers data tools but cannot settle what experience is.
- Detecting covert awareness in EEG or fMRI of patients who cannot respond
- Analyzing large adversarial-collaboration datasets to test what each theory predicts
- Building computational models of rival theories to derive sharper, testable predictions
Shown so far
- In August 2024 the New England Journal of Medicine reported that customized AI analysis of EEG in 241 unresponsive brain-injured patients found about 25% (60) who covertly followed commands. source
Prerequisites
- Cognitive revolution1956
- Split-brain studies1961Split-brain patients first showed how experience can divide inside one skull
- Functional MRI1992
- Brain wiring maps2024
Unlocks
- Consciousness test2040s?
- Whole-brain emulationopen