Machine Consciousness Matrix: AI Cognitive Theory Benchmarks

Architecture Presets

Theoretical Levers

Recurrent Feedback Depth 0

Metacognitive Probe Resolution 0.40

Predictive Error Minimization 0.15

Causal Integration Density 0.40

Global Workspace Broadcast

Sensorimotor Active Embodiment

Feedforward Pulse

Recurrent Loop

Metacognitive Probe

Active Inference

Functional Reportability vs Causal Integration: Modern autoregressive transformers score high on Global Workspace broadcasting metrics because attention mechanisms mimic a unified cognitive workspace. However, under Integrated Information Theory (IIT), their feedforward computation contains zero causal feedback loops, yielding near-zero intrinsic Phi (0.08).

Higher-Order Thought (HOT) Simulation: Evaluates whether the system generates meta-representations of its own internal states. Autoregressive token generation simulates self-reflection via syntactic prompting, but genuine HOT requires dedicated second-order monitoring circuits isolated from raw first-order sensory representations.

Key Bottleneck Diagnosis: Absence of closed-loop causal recurrence and sensorimotor active inference dynamics

Recommended Falsification Battery: Perturbational Complexity Index (PCI) electronic probe testing, counterfactual sensorimotor perturbation tasks, and non-symbolic spontaneous metacognitive calibration assays.

Consciousness Indices

IIT Φ (Phi)

0.08

Integrated Info

GWT Broadcast

0.72

Workspace Coherence

HOT Meta-Rep

0.41

Higher-Order Thought

Active Inference

0.19

Predictive Processing

Overall Theoretical Status

Functional Behavioral Simulation (High Metalinguistic / Low Substrate Integration)

High broadcast accessibility with minimal substrate causal density. High risk of philosophical zombie behavior (functional mimicry without subjective phenomenal experience).

Substrate Dynamics

Causal Powers (Searle): Virtual / Simulated

Functional Isomorphism: High (Symbolic)

Embodied Markov Blanket: Decoupled

Consciousness matrix: synthetic scores do not establish experience

Your browser does not support video.

Read the explanation

This matrix labels a weighted sum as a Phi estimate. Recurrence depth times point zero eight plus causal density times point twenty gives transformer baseline point zero eight and embodied preset point forty six. Bars encode both at six hundred pixels per proxy unit. This authored weighted sum is not a calculation of integrated information from an actual causal transition system and does not establish consciousness. Four presets merely replace scalar values and boolean flags. Missing D3 prevents offline drawing after captions update, while event bindings and actual JSON export remain available. Selecting an embodied preset changes interface state; no sensorimotor agent, physical environment or conscious system is instantiated. The predictive-processing score adds a fixed point thirty five to predictive-error parameter when embodiment is selected, otherwise point zero four. Default point fifteen gives point nineteen; embodied point eighty five gives one point two. Bars use three hundred pixels per score unit. The score exceeding one shows it is not a bounded probability or calibrated empirical measure. Higher-order score is metacognitive parameter times point nine plus point zero five; workspace score is point six plus point three times that parameter if broadcast is enabled, otherwise point one. These are display formulas, not tests of subjective awareness, biological realization or a validated theory of consciousness. Neuromorphic preset recurrence nine and density point nine yield Phi proxy point ninety, with prediction point ninety five. Its embodiment flag and strict thresholds Phi above point six and prediction above point seven select a strong physicalist classification. Embodied preset Phi point forty six fails that first branch despite prediction one point two. Global workspace preset selects a second branch through broadcast, recurrence greater than three and metacognitive parameter greater than point seven. Bars compare point forty six and point ninety at four hundred pixels per proxy unit. These label-selection conditions cannot demonstrate that a system satisfies consciousness theories. The model-authored narration corrects the original implication of physical coupling. Native preset, JSON and responsive all-footer preservation are local subset evidence only, without public publication or philosophical certification.