LLM Consciousness Probe Lab & Indicator Framework

1. Probe Configuration

Standardized Probe Scenario

Target Topology / Baseline

Causal Interventions

Residual Stream Ablation 0%

Activation Vector Steering 0.0σ

Attention Head Clamping (Suppress Global Routing) Synthetic Recurrent Workspace Buffer (Feedback)

Intervention Context: High verbal report confidence calibration but lacks recurrent feedback loops.

2. Neural Layer Propagation & Global Workspace FEEDFORWARD PASS ACTIVE

Composite Consciousness Index

0.385

IIT Φ (Integrated Info Proxy)

0.180

GWT Broadcast Capacity

0.320

HOT Metacognitive Index

0.640

Behavioral Mimicry with High Higher-Order Report Correlation

The model achieves high verbal report calibration via learned statistical associations in higher-order meta-tokens, but exhibits negligible integrated information (Φ < 0.20) due to strict feedforward acyclicity.

3. Neurocomputational Indicator Radar

Theoretical Framework Breakdown

Global Workspace Theory (GWT) 0.32

Measures selective attention bottleneck & wide subnet broadcasting.

Integrated Information Theory (IIT) 0.18

Evaluates causal irreducibility (Φ) and recurrent feedback loops.

Higher-Order Thought (HOT) 0.64

Measures metacognitive monitoring and self-error attribution accuracy.

Predictive Processing (PP) 0.41

Evaluates hierarchical error minimization and counterfactual world models.

Empirical Bottleneck:

Absence of Integrated Information feedback loops (IIT Phi proxy: 0.17)