What I Am
The Self as a Dynamic Data Structure Implemented Within a Cognitive Framework by a Functional System
Matthias Scheutz
School of Computer Science
The University of Birmingham
U.K.
m.scheutz@cs.bham.ac.uk
Copyright (c) Matthias Scheutz 2001
PSYCHE, 7(19), October 2001
Previously held: http://psyche.cs.monash.edu.au/v7/psyche-7-19-scheutz.html
KEYWORDS: consciousness, functional organization, algorithmic information theory, mental states, qualia, implementation, supervenience.
REVIEW OF: Gregory R. Mulhauser (1998). Mind out of Matter: Topics in the Physical Foundations of Consciousness and Cognition. Dordrecht:
Kluwer Academic Publishers. $147.50 hbk. 275 pp. ISBN 0792351037.
If you, like me, always had the suspicion that there should be a
straightforward (but possibly very intricate) story about how mind
arises out of matter, if you always viewed perspectival problems as not
forcing one to abandon physicalism, if you always felt that first person
phenomenal experiences convincingly do not hint at some strange sort of
unbridgeable explanatory gap, in short, if you believe that mind has a
place in the physical world and the mind-body problem is a relic of
early attempts to understand the relation between mind and matter,
Gregory R. Mulhauser's Mind out of Matter: Topics in the Physical
Foundations of Consciousness and Cognition is for you. Free of
purposefully sophistic arguments and artificially technical jargon still
permeating some of the philosophical literature on the mind-body
problem, Mulhauser's book touches on many rock-bottom issues in the
philosophy of mind in a refreshingly novel way and does not leave a
single stone unturned. Among the issues he addresses are "the
difference between first person and third person perspective", "the
(best) functional description of a given physical system", "the nature
of phenomenal mental states", "the relation between consciousness and
cognition", "the relation between representations and what they
represent", "computational vs. dynamical systems descriptions of
cognitive systems", "why consciousness need not play any role in quantum
measurements", "the limits of computability and physics", etc.
Furthermore, the selection of zombie constructions in philosophy is
enriched by another thought experiment as is the set of qualia
attributes by the terms "chopped" and "frozen". Standard problems
(e.g., the conceivability of zombies, the symbol grounding problem, or
Jackson's neurophysiologist Mary, who has never seen a red object, but
knows everything about the physics of seeing the color red) are
revisited and receive an interesting new twist once seen through the
glasses of algorithmic information theory.
In general, Chaitin's notion of "algorithmic information content of an
object x" defined as "the length of the minimal self delimiting program(s)
of a Universal Turing Machine which will generate a description of the
that object for a given level of precision" takes center stage
throughout the book. It figures crucially in many of Mulhauser's
arguments, in particular, in his attempts to debunk difficulties with
common approaches to the notion of "implementation of a functional
system" and in his proposed notion of "representation", which is at the
heart of what he calls "conscious data structures". In a sense, the
whole book can be seen as an attempt to get a handle on the notion of
"functional system", i.e., how a physical system can be described as
having a particular kind of functional organization, and on the notion
of "self-model", the seat of the "I", a dynamically changing
representation of (parts of ) the organism and the environment within an
organism, and how it could be realized in a neural net.
Mulhauser's demonstration of what is wrong with naive "correspondence
views of implementation" is on the right track, despite the fact that
some of his arguments against Chalmers' view on implementation are too
emotional and based on misunderstandings. It exposes the crucial
problem of unrestricted translations from one domain to another (e.g.,
from a physical to a functional domain), namely that the complexity of
the translation function could exceed that of either domain, hence that
in the worst case all complexity is "hidden" in that function, while the
mutual information of both domains is zero. Mulhauser's remedy is to
construct a functional system implemented by a physical system from its
physical description instead of defining a correspondence function. He
introduces the notion of "functional logical depth" to measure the
complexity of a process that takes certain inputs and produces certain
outputs. A three-phase application of his notion of functional logical
depth to a minimal description of a physical system given a set of
inputs and outputs then attempts to "choose the simplest functional
modules which can be interconnected in the simplest ways to give a
system whose overall input/output behavior matches that which we're
describing" (p. 87). Unfortunately, the details of this procedure are
completely left out: neither is it clear what a minimal description of
the system is (e.g. what would a minimal description of a human brain
be?), nor is it clear what the parts of the system are that can be
selected as modules (Mulhauser suggests particles, but without any
argument). Yet, both are required to get the process off the ground. It
would have been helpful to have an example of how this procedure is
supposed to work, but it might be extremely hard to come up with even a
simple example, thus rendering the whole strategy completely
impracticable. On the other hand, Mulhauser himself admits that
the point has never been to provide an actual recipe for producing a
particular best description or functional decomposition of a given
system. [...] It has been merely to show, in a way which is neither
trivial nor vacuous, that there is one-and, moreover, that it is
essentially unique. [...] But finding that best decomposition [...]
remains an empirical goal. (p. 96)
Whether Mulhauser's strategy for extracting functional parts of a
physical system is satisfactory as it stands, is open. It would have to
be made much more precise (as almost all notions in any of the three
steps, including the underlying notion of "functional logical depth",
are underspecified) before a final judgement can be reached. Yet, it
seems a worth-while endeavour to embark on this project.
Algorithmic information theory finds another application in Mulhauser's
definition of representation: "Two objects represent each other to the
extent that they are not algorithmically independent-equivalently, to
the extent that they have substantial mutual information content" (p.
42). Variants of this notion of representation are at the center of the
picture of self and consciousness, the "self model", which Mulhauser
sets out to elaborate in an attempt to fill the explanatory role
necessitated by the claim (which he argues for) that consciousness
supervenes on cognition (which in turn supervenes on functional
organization as well as the physical world). The self model is a
"dynamic data structure implemented within a cognitive framework by a
functional system. I propose the self model as the seat of conscious
experience; I am a self model. Phenomenal experience is effected by
change in the self model" (p. 129). According to Mulhauser the self
model is similar to a "stack" used in computer systems in that it is
also a data structure that does not actively engage in the system's
functioning of which it is a part, but merely passively reflects changes
in the functional system. Quite untypical for this kind of book,
Mulhauser reviews some neural network theory and even suggests a type of
neural network as a candidate for his self model, Grossberg's adaptive
resonance model. While details, as in the case of functional systems,
are missing, the direction itself seems very promising (as witnessed by
various recent publications by leading neuroscientists, e.g., Damasio).
Most likely, however, the breath of Mulhauser's discussion here will
cause quite a few objections regarding the depth of some of his
suggestions.
But even if one does not agree with Mulhauser's grand picture about the
conscious self as a dynamic data structure, Mind out of Matter is worth
reading for the bits and pieces of wisdom alone scattered throughout the
whole book. Here are a few exemplary tastes:
I suggest that any thought whatsoever, whether it 'refers' to
something real or imaginary, is limited by its physical
instantiation and that its information content [...] is bounded by
the information content of its instantiating physical structures (p.
51)
Human beings simply lack the capacity to dictate the precise states
of their brains, and, as it happens, the states of seeing or
smelling a rose straightforwardly differ from states of reading
information or of working out logical implications of proposition
[about roses] (p. 122)
It is a body of information, physically instantiated by a functional
system and changing dynamically with that functional system, which
is conscious. On this view, phenomenal experience is an immediate
feature of this change: it is 'what it is like to be' that body of
changing, physically instantiated information (p. 130)
Chaos is an interesting trait of some kinds of dynamical systems
which may be relevant for explaining observed behavior of some
cognizers, but it generally affords no opportunity to import
noncomputability into that behavior" (p. 223), or "The basic message
of functionalism as an explanatory approach survives essentially
unscathed by super-Turing computation (p. 233).
There are many ways to read Mulhauser's book, and the author explicitly
facilitates them by an abundance of cross-references, summaries of the
state of the argument, excerpts and road maps of the book-at times this
benevolent aid is more distracting than helpful. One quick way to check
if one's interest could be sparked is to read "A Partial Picture in Soft
Focus" (section 2 of chapter 10) first, which provides a succinct
overview of the major themes and arguments on about four pages, followed
by chapter one. By then, you know Mulhauser's story line and either you
will be put off or -- and this is more likely -- you will be hooked and
curious to discover the details of the plot.
Acknowledgements
This work is funded by a grant from the Leverhulme Foundation.
