Can Integrated Information Predict Consciousness Using Transition Matrices of Brain States? An Exploratory Study

Integrated information theory (IIT) is a leading formal theory of consciousness, and its Python toolbox PyPhi enables system analysis. Computational and empirical constraints make it difficult to test whether higher integrated information values correspond to higher consciousness levels. Using data from a prior study that applied machine learning techniques to derive transition probability matrices among brain states at different consciousness levels, the authors found that integrated information values were not associated with conditions or brain states characterized by greater consciousness. Limitations and future directions are discussed.

Study at a glance

Abstract

Integrated information theory (IIT) is one of the most advanced formal theories of consciousness, featuring a Python toolbox (PyPhi) which allows us to analyse a system according to the corresponding theoretical framework. Computational and empirical limitations make it hard to test the hypothesis that higher values of integrated information are associated with a higher level of consciousness. We leverage the availability of data collected by a previous study (Huang et al., 2020) which is amenable to an IIT 3.0 analysis employing the PyPhi toolbox. The Huang et al. study employed a mix of supervised and unsupervised machine learning techniques (k-means, SVM) to obtain and validate transition probability matrices among brain states for different levels of consciousness. We observed that the integrated information values are not associated with the conditions and the brain states characterized by greater consciousness level. Limitations and future opportunities of our approach are discussed.