Comparing Theories of Consciousness
A comparison needs a common experiment. Specify the stimulus and intervention, the observer, the time of each probe, and the distribution of responses the accounts predict. A model of access can explain which information reaches a decision. A model of reported similarity can explain which experiences a participant groups together. Agreement on one record is evidence about that record. An identification of experience across people requires further assumptions and a test that can distinguish alternatives. Operational states · Reports and interventions
A preregistered comparison
The Cogitate Consortium compared predictions of integrated information theory and global neuronal workspace theory in the same visual task. It found information about conscious contents in several cortical regions. The expected sustained posterior synchronization and the proposed prefrontal ignition at stimulus offset each faced contrary evidence. The study discriminates among specific predictions about location, time and connectivity. A new account can enter that comparison by predicting the same outcomes before seeing them. Study and preregistered predictions
Counterexamples and what they test
- Matching behaviour. A recurrent circuit unrolled across a fixed finite horizon can preserve the same input–output map while changing internal causal structure. Behaviour on this horizon cannot distinguish these circuits. Testing a proposed internal cause requires access to an additional intervention. Unfolding argument
- Substitution. If a change preserves the report while changing a model's consciousness score, the report alone cannot adjudicate the two assessments. Calibrating the score on that same report can instead make agreement true by design. Freeze the calibration and state an independent prediction. Substitution argument · Measurement protocol
- Large integration scores. Aaronson gives simple networks, including expander constructions, with high scores for proposed measures of integration. The calculation tests a measure's response to connectivity. Calling such a network conscious requires a separate criterion for that interpretation. Even in the seven-level model a product of uncorrelated states can increase the basis-dependent coherence score. Aaronson's example and discussion · Product-state calculation
Mechanisms and discriminating observations
Autopoiesis
Which processes maintain the boundary and regenerate the components after a perturbation? A maintained boundary supplies evidence for self-maintenance.
Integrated information theory
IIT specifies cause–effect repertoires and partitions of a causal system. A coherence score in a chosen density-matrix basis does not determine those repertoires. Equal density matrices can be paired with different transition laws.
Free-energy principle
A variational free-energy comparison needs a generative model, observations and an approximate posterior. For fixed state and target, entropy plus relative entropy reduces to cross-entropy. That identity supplies no update that minimizes a specified prediction error.
Global workspace
Test whether information delivered to one subsystem becomes available to other tasks, under a manipulation that changes broadcasting. Global access and a report about experience are separate outcomes.
Higher-order theories
Vary the accuracy of a higher-order representation while matching first-order discrimination. Confidence, type-two sensitivity and reports must have distinct scoring rules.
Predictive processing
Change the prior information while holding the stimulus fixed. Fit rival predictive models before comparing held-out errors. A fixed replacement channel contracts to its chosen target even with no observation input.
Attention schema
Manipulate information about where attention is directed and compare attention performance with judgments about attention.
Quantum cognition
Order effects in answers can be modelled with noncommuting questions. Compare the complete response distributions with a classical model carrying memory of the first question.
Orchestrated objective reduction
The proposal involves a physical reduction mechanism in microtubules. A Hilbert-space model of report frequencies supplies no measurement of that mechanism.
Recurrent processing
Recurrent and time-unfolded feedforward circuits can have identical input–output behaviour on a fixed finite horizon. Internal perturbations can distinguish their wiring.
Neuronal group selection
Track selection and re-entrant signalling among neuronal groups over learning trials.
Adaptive resonance
Vary the match between incoming patterns and learned categories. Measure category resets and retention of previously acquired responses.
Enactivism and 4E cognition
Perturb the sensorimotor loop, including the body or environment, and compare performance when the same sensory input is replayed passively.
Sensorimotor contingencies
Change the mapping from action to sensation and measure transfer after learning the new mapping.
Temporo-spatial theory
Compare spontaneous neural timescales, stimulus timing and the duration of a reported percept.
Dendritic integration
The proposed mechanism relates basal and apical inputs to cortical neurons. A test must distinguish the contributions of these inputs.
Electromagnetic information
A field account requires a manipulation whose effect distinguishes the proposed field coupling from neural coactivation.
Perceptual control
Perturb an environmental variable and test whether action holds a declared perceptual signal near its reference.
Operational architectonics
Specify EEG segmentation and coupling criteria before testing their relation to changes in a report.
Neural correlates of consciousness
Compare matched stimuli with different reported percepts while tracking report demands, attention and performance.
Assembly theory
An assembly index depends on the allowed components and joining operations. It says how an object can be constructed under those rules.
Quantum mind
Different quantum proposals require different physical degrees of freedom, coherence times and readouts. A fitted probability formula cannot identify a carrier.
Dissipative adaptation
Compare driven systems at controlled energy input and measure work, dissipation and changes in configuration.
Russellian monism
The interpretation distinguishes structural relations from claims about intrinsic nature. Relational measurements establish the former.
Multiple drafts
Vary when and how a judgment is elicited. Reports of the same stimulus can depend on the probe and later information.
Affective neuroscience
Distinguish measured approach, avoidance and arousal from the interpretation of a felt emotion, especially across species.
Somatic markers
Record bodily responses and decisions independently, then test whether the former predict subsequent choices beyond prior behaviour.
Interoceptive inference
Perturbation of bodily cues permits a comparison of interoceptive inference with a model driven only by exteroceptive cues.
Subcortical organization
The proposed contribution of subcortical circuitry needs evidence that separates it from cortical contributions and from motor output.
Neuropsychoanalysis
Distinguish affective reports, arousal and physiological regulation when testing their proposed relationship.
Holonomic brain theory
A distributed representation can be tested by the response to local damage. A Fourier description alone does not identify a neural mechanism.
Functional systems
The proposed feedback cycle compares an expected outcome with an attained one. Vary feedback after a matched action.
System-evolutionary theory
Test changes in neuronal recruitment across acquisition and performance of a learned action.
Information synthesis
The proposal relates a current signal to memory and motivational context. Vary these inputs independently.
Consciousness as paradox
A proposed selection process needs an outcome that distinguishes access, inability to report and an active exclusion mechanism.
Projective wave theory
A projective model of perceived space and a physical wave mechanism make separate claims. Calibrate geometry and look for independent wave measurements.
Categories of qualities
Yoneda identifies objects up to isomorphism once the category and its arrows are given. Similarity reports can supply candidate arrows. Demonstrating composition and transfer across observers requires additional tests.
Formal IIT and process theories
To compare process representations, specify the preparations, admissible interventions, partitions and composition laws. A matching symbol or dimension does not establish a functor.
Minimal physicalism and quantum reference frames
Specify the observer’s physical interface and which boundary records the two models predict. Independence under a frame change is a property of those records.
Perceptronium
A proposed phase of matter needs a state space, dynamics and an order parameter. A gate can change value on a smooth trajectory with no dynamical bifurcation.
Quantum-information accounts of experience
Purity and distinguishability have precise meanings for a stated quantum preparation. Identifying them with the privacy or unity of experience needs an independent bridge.
Observer theory
Different observers can identify different states if the allowed tests differ. A common held-out experiment can determine which distinctions transfer.