A K-Means-Based Strategy for Estimating the MIP in Integrated Information Theory

Abstract

This article proposes a strategy to address the computational challenge of finding the Minimum Information Partition (MIP), a key concept in Integrated Information Theory (IIT) for quantifying consciousness. Due to its combinatorial complexity, solving the MIP problem is intractable for large systems. To tackle this, we adapt the k-means clustering algorithm, transforming the problem into a clustering task. This approach offers a scalable and efficient alternative to exhaustive searches, making it feasible for complex systems. Our method advances the practical application of IIT, enabling its use in larger biological and artificial systems. Additionally, it provides a theoretical foundation for estimating integrated information and exploring approximate solutions when exact computations are unfeasible. The proposed strategy balances computational efficiency and accuracy, demonstrating its potential for handling high-dimensional data while maintaining reasonable performance. This work represents a step toward making IIT more applicable in real-world scenarios, fostering further research in consciousness studies and intelligent system design.