What it is like to be a bit: an integrated information decomposition account of emergent mental phenomena

Andrea I. Luppi, Pedro A. M. Mediano, Fernando E. Rosas, David J. Harrison, Robin Carhart‐Harris, Daniel Bor, Emmanuel A. Stamatakis

TL;DR

This paper proposes a new way to understand consciousness by breaking it down into its fundamental information-theoretic components, moving beyond simply quantifying it. This "Integrated Information Decomposition" (ΦID) framework allows us to formally define consciousness as potentially emergent or non-emergent, and to identify distinct "modes of consciousness" linked to specific information processing patterns.

What Problem Does This Paper Address?

Neuroscience and philosophy have long grappled with the relationship between consciousness and its physical basis, particularly the concept of "emergence" – how complex properties arise from simpler components. While Integrated Information Theory (IIT) offers a way to quantify consciousness by measuring "integrated information" (Φ), it doesn't fully explain how this integration arises or how it relates to emergence. This paper aims to bridge this gap by decomposing integrated information into its constituent parts, allowing for a more nuanced understanding of consciousness and its potential emergent properties. It also addresses theoretical limitations in existing IIT measures, such as the possibility of negative values.

Methods

The core of this paper is the application of Partial Information Decomposition (PID) and its multivariate extension, Integrated Information Decomposition (ΦID). PID breaks down information flow between systems into distinct "atoms" like synergy, redundancy, and unique information. ΦID extends this to multi-target systems, allowing for a detailed analysis of how information is processed.

The authors adapt IIT's measure of integrated information (Φ) from its 2.0 version, which is compatible with Shannon's information theory. They then use ΦID to decompose this Φ into its constituent atoms. This decomposition leads to a revised measure of integrated information, Φ<sub>R</sub>, which focuses on synergy and transfer atoms, ensuring it's always non-negative. They also introduce the Φ<sub>R</sub>-ing ratio to quantify the efficiency of information processing for consciousness.

Furthermore, the paper leverages ΦID to formally define causal emergence, distinguishing between "downward causation" and "causal decoupling," and linking these to specific ΦID atoms.

Datasets

Not discussed in the paper. The paper focuses on theoretical contributions and outlines empirical avenues, but does not present new experimental data or datasets.

Key Insights and Ideas

  • Consciousness as a Composition: Consciousness isn't just a single quantity (integrated information) but a complex composition of different information-theoretic "atoms."
  • Emergence is Not All-or-Nothing: Whether consciousness is emergent depends on its specific information-theoretic composition, not on consciousness itself being inherently emergent. Some aspects of consciousness might be emergent, others not.
  • Modes of Consciousness: Different combinations of information-theoretic atoms can lead to qualitatively different "modes of consciousness," providing a framework for mapping subjective experience to neural dynamics.
  • Selfhood as Causal Decoupling: The paper proposes that the sense of self is linked to "causal decoupling," a specific type of emergence where the macro-level system has causal influence over its own dynamics, independent of its micro-level parts.
  • Addressing IIT's Limitations: The ΦID framework provides a way to derive a non-negative measure of integrated information (Φ<sub>R</sub>) and a measure of processing efficiency (Φ<sub>R</sub>-ing ratio), addressing theoretical shortcomings of earlier IIT measures.

Key Results

  • Formal Definition of Emergence: The paper provides a formal, information-theoretic definition of causal emergence and distinguishes between downward causation and causal decoupling.
  • Consciousness and Emergence Relationship: It establishes that consciousness can be emergent, non-emergent, or a combination, depending on its ΦID composition.
  • New Measures: Introduction of Φ<sub>R</sub> (a non-negative measure of integrated information) and the Φ<sub>R</sub>-ing ratio (efficiency of information processing for consciousness).
  • Hypothesized Link to Selfhood: The paper posits that selfhood is associated with causal decoupling (persistent synergies), and that this can be dissociated from consciousness itself (e.g., "self-less consciousness").
  • Empirical Predictions: The framework generates testable predictions about how different information-theoretic atoms might correspond to specific phenomenological experiences, particularly in altered states of consciousness.

Limitations and Future Work

  • Theoretical Framework: The paper is primarily theoretical, outlining a framework and making predictions. Empirical validation is the crucial next step.
  • Mapping Phenomenology: While the paper proposes a framework for mapping phenomenology to information-theoretic modes, the precise correspondence between specific atoms and subjective experiences is still to be determined.
  • Complexity of Real Systems: The paper often uses simplified bivariate systems for illustration. Extending the ΦID framework to more complex, multivariate systems (like the human brain) is an ongoing area of research.
  • Baseline State Definition: Defining a "baseline" state of consciousness for comparison is acknowledged as a non-trivial challenge.
  • Future Work: The authors suggest exploring the relationship between ΦID and other theories like Global Neuronal Workspace Theory (GNWT) and predictive coding/Free Energy Principle. They also emphasize the need for empirical studies to validate the proposed links between information-theoretic atoms and phenomenology, particularly in altered states of consciousness and across different species.

Summary updated

Abstract

A central question in neuroscience concerns the relationship between consciousness and its physical substrate. Here, we argue that a richer characterization of consciousness can be obtained by viewing it as constituted of distinct information-theoretic elements. In other words, we propose a shift from quantification of consciousness-viewed as integrated information-to its decomposition. Through this approach, termed Integrated Information Decomposition (ΦID), we lay out a formal argument that whether the consciousness of a given system is an emergent phenomenon depends on its information-theoretic composition-providing a principled answer to the long-standing dispute on the relationship between consciousness and emergence. Furthermore, we show that two organisms may attain the same amount of integrated information, yet differ in their information-theoretic composition. Building on ΦID's revised understanding of integrated information, termed Φ<sub>R</sub>, we also introduce the notion of Φ<sub>R</sub>-ing ratio to quantify how efficiently an entity uses information for conscious processing. A combination of Φ<sub>R</sub> and Φ<sub>R</sub>-ing ratio may provide an important way to compare the neural basis of different aspects of consciousness. Decomposition of consciousness enables us to identify qualitatively different 'modes of consciousness', establishing a common space for mapping the phenomenology of different conscious states. We outline both theoretical and empirical avenues to carry out such mapping between phenomenology and information-theoretic modes, starting from a central feature of everyday consciousness: selfhood. Overall, ΦID yields rich new ways to explore the relationship between information, consciousness, and its emergence from neural dynamics.
BibTeX
@inproceedings{luppi2021what,
  author = {Andrea I. Luppi and Pedro A. M. Mediano and Fernando E. Rosas and David J. Harrison and Robin Carhart‐Harris and Daniel Bor and Emmanuel A. Stamatakis},
  title = {{What it is like to be a bit: an integrated information decomposition account of emergent mental phenomena}},
  booktitle = {Neuroscience},
  year = {2021},
  doi = {10.1093/nc/niab027},
}
Figure