Evaluating Approximations and Heuristic Measures of Integrated Information

Area of Science:

• Neuroscience and theoretical physics
• Investigating the physical basis of consciousness

Background:

• Integrated Information Theory (IIT) proposes Phi (Φ) as a measure of consciousness.
• Calculating Φ is computationally intensive, limiting its application to small systems.
• Existing approximations for Φ lack rigorous validation against IIT's core principles.

Purpose of the Study:

• To evaluate heuristic measures and computational approximations of Phi (Φ).
• To determine the accuracy of these approximations in estimating Φ values in model systems.
• To assess the potential of these measures for estimating a system's capacity for high Φ.

Main Methods:

• Simulated small networks (3-6 binary nodes) with random connections.
• Constructed state transition probability matrices (TPMs) and generated time-series data.
• Calculated exact Φ, various approximations, and related measures (e.g., state differentiation, signal complexity).

Main Results:

• Φ was closely approximated (r > 0.95) in small binary systems by readily available approximations.
• State-independent maximum Φ correlated strongly with signal complexity (LZ, rs = 0.722), Φ* (rs = 0.816), and D1 (rs = 0.827).
• These measures can efficiently estimate a system's capacity for high Φ or predict low-Φ systems.

Conclusions:

• Readily available approximations can accurately estimate Phi (Φ) in small binary systems.
• Measures of signal complexity, Φ*, and state differentiation show strong correlation with maximum Φ.
• Further validation is needed for larger systems and different dynamics to establish practical alternatives to Φ calculation.