Available online at www.sciencedirect.com
Physics of Life Reviews 9 (2012) 456–457 www.elsevier.com/locate/plrev
Comment
Self-organisation, inference and cognition Comment on “Consciousness, crosstalk, and the mereological fallacy: An evolutionary perspective” by Rodrick Wallace Karl Friston The Wellcome Trust Centre for Neuroimaging, University College London, Queen Square, London WC1N 3BG, United Kingdom Received 23 September 2012; accepted 25 September 2012 Available online 26 September 2012 Communicated by L. Perlovsky
Keywords: Self-organisation; Information theory; Ergodic; Entropy; Free energy
The treatment of cognition and [co] evolutionary dynamics offered by Wallace [1] is deep and thought provoking. It provides a rich banquet of theorems, perhaps too rich and long – to enjoy it fully, one would have to be very hungry and have a broad palette. However, Wallace is not just showcasing his encyclopaedic knowledge of theoretical biology and statistical physics; there is genuine purpose here. So what is his basic message? My understanding is as follows: the mereological fallacy of attributing conscious attributes to the brain can be dissolved in a simple and compelling way by contextualising cognition in an embodied or situated sense; in other words, by considering the fundamental constraints on self-organising systems that endure in a changing world. In this view, cognitive processes – conscious or unconscious – necessarily emerge at multiple scales and are not constrained to the machinations of the brain. For example, Wallace reviews wound healing, tumour control, immune function, genetics and culture as instances of cognition at different timescales. So what permits this cognitive or informational attribution? The story starts with the very existence of adaptive (biological) systems and their implicit ergodicity: namely, their long term occupancy of any state converges (almost surely) to the probability of being in a particular state. Put simply, the fact that biological systems persist in the face of environmental fluctuations means that they revisit particular states and – through the ergodic theorem – induce a probability distribution over their states. The existence of this probability distribution enables Wallace to harness limit theorems in communication and information theory to dissolve the gap between the physics of systems and their information theoretic representation. In short, if systems have an enduring form they are ergodic in some (weak) sense and this endows them with a dual representation in terms of (thermo) dynamics and information or cognitive processing. Wallace then goes on to examine “the necessary conditions of the asymptotic limit theorems of communication theory to” beat “the mathematical thicket back one layer”. This thicket represents the complicated phenomenology of stochastic differential equations describing the (co-evolutionary) dynamics of adaptive systems. We are then treated to a brief review of several perspectives; DOI of original article: http://dx.doi.org/10.1016/j.plrev.2012.08.002. E-mail address:
[email protected]. 1571-0645/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.plrev.2012.09.005
K. Friston / Physics of Life Reviews 9 (2012) 456–457
457
calling upon things like index theorems, Landau’s theory of punctuated phase transitions, the rate distortion theorem, Kuhn–Tucker optimisation and much more. I imagine I was asked to write this review because of the close relationship between the treatment provided by Wallace and the (variational) free energy principle [2]. I close by highlighting their remarkable consilience: crucially, the motivation for free energy minimisation is rooted in the ergodic theorem and says that homeostasis necessarily requires the entropy of a systems states to be minimal, which requires the minimisation of self-information (or variational free energy) at each instant in time. This minimisation provides the same informational perspective on exchanges with the environment described above. In minimising free energy, every system becomes an inference machine or cognitive engine in the sense of Helmholtz [3,4]. The connection between inference and free energy was established by Richard Feynman who “describes in great detail how information and free energy have an inherent duality” [1]. This inferential perspective resonates nicely with “the central point of Atlan and Cohen [5], in the context of the immune system, in that cognitive function involves comparison of a perceived signal with an internal learned or inherited picture of the world”. This is precisely the form of Bayesian inference that emerges from the free energy principle, which – when extended to cover action – leads to the “broad spectrum of functions that can only be embodied by the full construct of individual in context” [1]. Acknowledgements The Wellcome Trust funded this work. References [1] [2] [3] [4] [5]
Wallace R. Consciousness, crosstalk, and the mereological fallacy: An evolutionary perspective. Phys Life Rev 2012;9(4):426–53 [this issue]. Friston K. The free-energy principle: a unified brain theory? Nat Rev Neurosci Feb 2010;11(2):127–38. Helmholtz H. Concerning the perceptions in general. In: Treatise on physiological optics. New York: Dover; 1962 [1866]. Dayan P, Hinton GE, Neal RM. The Helmholtz machine. Neural Comput 1995;7:889–904. Atlan H, Cohen I. Immune information, self-organization, and meaning. Int Immunol 1998;10:711–7.