Peter Cariani’s First Paper Proposal

A cybernetic model of brain function

An integrative model of brain function is proposed whose functional organization involves adaptive cybernetic, goal-seeking, switching, and steering mechanisms embedded in percept-action-environment loops. The brain is conceived as a network of reciprocally-connected, re-entrant loops (McCulloch, Lorente) within which circulate neuronal signals that build-up, decay, and/or actively regenerate.

The basic signals themselves are temporal patterns of spikes (temporal codes), held in the spike correlation mass-statistics of neuronal ensembles. The simplest temporal codes are interspike interval codes that subserve auditory pitch perception and somatosensory flutter-vibration. Complex temporal codes afford vectorial representations, multiplexing of multiple signals in spike trains (Lettvin) and broadcast strategies of neural coordination. We have proposed neural timing nets as time-domain alternatives to conventional rate-based connectionist architectures.

A functional mental state consists of the set of neural signals circulating in the system at any given point. Neuronal loops function as pattern-amplifiers. Under most circumstances the loop-gains are weakly-attenuating, whereas when circuits are disinhibited, loop-gains switch to weakly-amplifying. Compatible sets of neuronal signals can be mutually reinforcing and stabilizing, such that there is an ongoing autopoietic self-production of neuronal signals (pace Maturana) that can maintain steady informational states (von Foerster’s neural dynamical system eigenstates , Pask’s mutually stabilizing interactions of concepts). There are thus temporal interference patterns and pattern-resonances between mutually compatible neural signals (Lashley) that can be modeled in terms of mutually interacting, attracting temporal pattern-vectors that yield convergent meaning.

Cybernetic and informational functions can be mapped onto brain structures: sensory and motor processing to subcortical ascending and descending pathways and unimodal cortex; pattern analysis, recognition, transformation, and generation to cerebral cortex; switching of motor programs and attentional amplificative facilitation of sensory signals via disinhibition to basal ganglia, real time steering functions to cerebellum; mnemonic consolidation re-broadcast to hippocampus, goal-mechanisms to reward-based anticipatory prediction and remodelling mechanisms of midline dopamine centers.

Music may produce its manifold effects on us because it weakly impresses temporal patterns onto neural circuits that normally utilize those patterns for basic functions, thereby inducing particular kinds of mental states. Music also combines novelty and predictable structure to effectively drive reward (pleasure) circuits.

The model bears neurophenomenological interpretations. Those sets of signals that are actively regenerated in global loops that subserve working memory (Baars’ “global workspace”) determine the experiential contents of conscious awareness, whereas the ability itself to coherently regenerate neural signals is held to be the organizational requisite for waking consciousness (not unlike Pask’s concept of consciousness as organizational closure, albeit of neural signals). Circumstances that abolish awareness entirely (anesthesia, coma, seizure) disrupt this global regeneration process. Alternate states of consciousness (sleep, hypnosis, meditation, trance) involve different, more localized regenerative patterns.

Those circular-causal processes generated from within are experienced as self-generated thoughts, whereas those contingent sensory inputs that switch neural activity patterns from without are experienced as external sensations. Those autonomous sensorimotor routines that come to effectively control some aspect of the outer world become unconscious automatisms. Fields of internal vs. external causation thus may provide functional and phenomenal boundaries for the self.

TRADITIONS: NEUROBIOLOGY, EPISTEMOLOGY

Cybernetic traditions:

  • 8) Neurobiology; consciousness studies
  • 3) Experimental epistemology; constructivism; philosophy of science

2 thoughts on “Peter Cariani’s First Paper Proposal

  1. clarissal

    As I was reading your abstract, I was thinking of both Francisco Varela’s work and the more reason biophysical approaches to thinking about the principles of neural networks in birdsong through the work of a group of interdisciplinary scientists at Duke http://www.dibs.duke.edu/research/research-themes/circuits-and-behavior. I can’t claim any expertise in the field of neurobiology, but I am particularly interested in the application of non-linear equilibrium theories in a cybernetical reading of neuronal mapping, and your paper seems to be performing an aspect of it. However, one of the problems in neurophenomenology has been the difficulties in bridging between a still vague understanding of consciousness and neural mechanisms that had frustrated me in my earlier years as a graduate student trying to explore the concrete potential of this research. I really want to see how you will be bridging that connection, and in a sense, it parallels what I am trying to do in my paper, though not in a way completely similar, between the phenomenological representation of theory and experiment via the Monte Carlo method, and the more subjective invocation quantum interpretations that both Von Neumann and Foerster had tried to do. Even then, I still feel that cybernetics, as it stands provides, a map for potential connections but not exactly the directive for how one can practically go forward.

    Reply
  2. Peter Cariani

    I worked (am still working) on the neural code for pitch — what are the neural requisites and differences in activity that evoke the experience of one pitch vs. another? The problem involves figuring out which aspects of neural activity co-vary with the pitches that we hear. We found that the neural correlates of pitch, at the earliest stages of neural representation (in the auditory nerve) involve temporal patterns of spikes (interspike intervals — the time durations between spikes). From distributions of observed interspike intervals we can predict with high accuracy and reliability what pitch will be heard.

    The neural coding problem naturally leads into the nature of the neural representations and their relationship to the contents of our conscious awareness. This is the problem of the neural coding of pitch vis-a-vis conscious awareness, what I call the neurophenomenological problem. We need to develop bridge laws that predict what we experience in terms of distinguishable qualities from the neural activity patterns. This is related to, but not exactly equivalent to, the neural coding problem. They can be different because there are some perceptual functions, like blindsight, where there is some discrimination that is possible by the system of which the subject is nevertheless unaware (the neural mechanisms and represetnations need to be there, but there are additional organizational requirements for that representation to enter conscious awareness).

    The neural coding problem and the neurophenomenological problem are pretty straightforward once one begins thinking about correspondences between neural activity on one hand and mental functions and experiences on the other. Several years ago, I was tasked by a foundation to design a research funding program to tackle the problem of the neural basis of consciousness, so I have had a great deal of time to think about these issues.

    It is possible to be very clear about consciousness, despite all the philosophical and conceptual confusion that is out there. We need to understand what are the essential differences in neural activity patterns that distinguish the conscious waking state from the unconscious state under general anesthesia (the neural requisites of conscious awareness, NCCs) and also the essential differences that distinguish different contents of awareness (green square vs. orange circle, note A from F#, NCCCs neural correlates of contents of consciousness).

    I think that the main problem in understanding the brain that confronts us is identifying the signals of the system — how information is coded. Before DNA we knew that there existed mechanisms of inheritance somehow bound up with chromosomes, but not how these worked. After DNA we now know the basic informational organizational framework of all organisms. When we crack the neural code, the clouds will similarly part, and we shall have a broad understanding of how it all works. Until we crack the code, we are just groping in the dark.

    Unfortunately, most of neuroscience does not deal with information per se (it’s been taken over by the molecularists), and the problem of the neural code is barely on the radar screen (many establishment neuroscientists believe the dogma thatf rate coding and connectionist networks provide an adequate explanation for how the brain works. Our funding systems only support consensus science (one thumbs down in a panel vetos a proposal), so anyone who has serious doubts about the current dogmas is going to have a hard time getting his or her research funded.

    Re: Maturana and Varela, I have often thought that they might conceive of signal processing in the brain in terms of mutually-stabilizing self-production networks — it would seem to be a natural kind of theory for them , yes? But I have never been able to find anything in their writings that clearly spells out how they think about these things. I tried looking in Varela and Thompson’s book The Embodied Mind, but the theorizing about brain function and consciousness rapidly devolved into Buddhist jargon which I could not follow. Although they are critical of symbolic cognitivism, they accepted conventional wisdom re: neural codes and connectionist architectures.

    Maturana is much more of a neuroscientist than Varela, and knew McCulloch, Pitts, and Lettvin while at MIT, so he may well be thinking (as they did) about temporal codes and interactive dynamics of neural signals in loops, but I have never had an opportunity to ask him about it.

    Much of birdsong recognition involves rhythmic pattern, which is also robustly temporally coded at all levels.

    Although I think that the brain may be largely modelable in terms of homogeneous excitable media that propagates pulse-coded patterns through axonal delay lines and temporally-shaped membrane recovery dynamics (Raymond-Lettvin theory of anesthesia), it is absolutely necessary to focus on informational functions (psychological functions) if we are to understand how the brain works as an informational system. Biophysical dynamics alone will not get us there — the dynamics need to be related to specific functions.

    Cybernetics is very important in neuroscience because it provides a functional framework for how the system is organized. Nervous systems exist to coordinate behaviors that realize embedded goals related to homeostasis, survival, and reproduction. Nothing makes any sense in biology or psychology outside the context of a purposive, functional framework. Without an integrative framework we have the situation that prevails in molecular biology — lots of little structural details, but little deep understanding of the whole organization of the system.

    I will try to give my high-level account of this cybernetic framework in terms of brain function in my talk on Tuesday morning.
    Sorry about the length of this.

    Reply

Leave a Reply

Your email address will not be published.