Comments on Aur’s “From Neuroelectrodynamics to Thinking Machines”

Abstract

In [1] Dorian Aur heatedly criticizes what he calls the ‘‘current neurophysiological doctrine,’’ which relies on the measurement of neural events on a millisecond time scale, that is, spikes or action potential. Aur’s intention is no other than to terminate one of the most fundamental ideas in neuroscience since the pioneering work of Edgard Adrian in the 20’s, the functional relevance of these nerve impulses as carriers of information. In Aur’s view, the spike timing and other related forms of neural coding expressed in terms of temporal observables are no more than epiphenomena. The principles of neural computation must be found in the spatial distribution of electrical processes that occur during the action potential. Thus, ‘‘current mainstream provides a weak understanding of computations performed in the brain,’’ because it ignores the ‘‘hidden information’’ embedded in the complex microscopic interactions inside the cell, during the millisecond time frame of a spike. Thus, ‘‘intrinsic computational processes’’ are decided at a much slower time scale and smaller space scale than is commonly assumed in neurophysiology. Neuroelectrodynamics (NED), a new theoretical framework that borrows from Hamiltonian mechanics, Thermodynamics, Quantum Physics and non-Turing computation, is surmised as the ‘‘change in paradigm required’’ to understand ‘‘brain language.’’ This review highlights the methodological pitfalls and conceptual errors introduced in the model suggested by Aur. First, it is shown that the mathematical equations proposed are not adequate for the studied system, that is, the brain, and second, a discussion on the aftermath of the dismissal of spike trains as carriers of relevant information, as stated by Aur, is sketched. First, with regard to the methodological aspects, Aur makes a claim for ‘‘adequate techniques’’ in order to understand ‘‘the neuron’s language.’’ For Aur, the diversity found in actual recording of action potential propagation in nerve cells needs to be explained in terms of the spatial distribution of electrical charges inside the neuron. The spike directivity vector is presented as the tool put on place to reveal the hidden information laying in the intracellular interactions inside the cell. Thus, while mainstream neurophysiology assumes that it is the timing of the spike that matters, Aur announces a new approach to understand neural computation, up to now unperceived by neurophysiologists, in which meaningful patterns are built upon spike directivity vectors that quantify transient charge density taking place during action potential. The methodological implications of Aur’s approach are of a devastating complexity, owing to the stratospheric dimensionality of the neuron models needed to capture the dynamics of ions, molecules and proteins inside every single cell. It is hard to imagine how one may come to grips with the dynamics of such a gargantuan system. There are millions of proteins inside each neuron! Surprisingly enough, Aur’s bet is Hamiltonian mechanics, which is mainly geometry in phase space [2]. Although Aur’s modeling choice is entirely legitimate, the actual model proposed does not apply neither aims at the physical reality for which claims to be conceived, the brain. It goes without saying that a model is always a simplified description of some features of a system, for example point neuron models are simplifications unable to simulate J. Gomez-Ramirez (&) Autonomous Systems Laboratory, Universidad Politecnica de Madrid, Jose Gutierrez Abascal, 2, 28006 Madrid, Spain e-mail: jd.gomez@upm.es