Re: Ghost in the Machine: What Computer Scientists andNeuroscientists can Learn from Each Other

From: Jim Fehlinger (fehlinger@home.com)
Date: Sat Nov 18 2000 - 07:52:08 MST


Apropos the view of the brain as a Darwininan machine, as
espoused by Gerald M. Edelman and others, I thought I'd
mention a book with a similar theme which I ran across
recently: _Darwin Machines and the Nature of Knowledge_ by
Henry C. Plotkin, 1993, Harvard University Press (paperback
edition, 269 pages; see
http://www.amazon.com/exec/obidos/ASIN/0674192818/qid=974465462/sr=1-2/107-6292930-3771726
The author is described on the book jacket as a professor of
psychobiology at University College in London.

I found that this book complemented my earlier reading of
Edelman (see
http://www.amazon.com/exec/obidos/ASIN/0465013767/o/qid=974464826/sr=2-1/107-6292930-3771726
and
http://www.amazon.com/exec/obidos/ASIN/0465007643/qid%3D974464955/107-6292930-3771726
and see also my earlier remarks about Edelman at
http://www.lucifer.com/exi-lists/extropians.2Q00/date56.html
[search down the page for "A Spring-Powered Theory of
Consciousness"]). Whereas Edelman concentrates mainly on
neuronanatomy and the embryogenesis, development, and
architecture of the brain as the physical substrate of what
he calls "primary consciousness", the details of which he
describes in selectionist terms, Plotkin is concerned more
with theoretical considerations of Darwininan evolution as a
mechanism for matching biological organisms to a world which
manifests fluctuations and stabilities on many different
time scales -- in other words, a mechanism of adaptation.

Plotkin describes the efforts which have been made to
generalize "classical" Darwinism (or neo-Darwinism, as it
became when fused in the 20th century with Mendelian
genetics), whose proper domain is the change over time of
biological species (driven by a two-stage process "with
genetics as the principal engine of variation and selection
acting mainly upon the phenotype" [p. 46]) to a more
abstract, encompassing theory which can be used to account
for variation over time **within** individual organisms as
well as that **between** individuals. As does Edelman,
Plotkin points to the immune system as a solidly
uncontroversial example of intra-organismic change which is
universally acknowledged (the author claims) to occur by a
process analogous to that of classical Darwinian selection
among individuals.

To eliminate the analogical or metaphorical aspects of this
characterization, Plotkin calls upon what he names
"universal Darwinism" (see Chapter 3), a term which Richard
Dawkins originated, and which was intended by him to
indicate that "such processes are not bound to life on Earth
but are the same wherever adaptation and/or speciation are
occurring or have occurred anywhere in the universe"
(p. 60). Plotkin, however, alters the sense of this phrase
to indicate that "it isn't only adaptation and speciation as
conventionally understood that are driven by these
processes, but that here on Earth certain other forms of
transformation of living systems are also caused by these
same processes" (p. 60).

Plotkin describes several different approaches to a
generalized Darwinian theory, one of which he calls the
"g-t-r heuristic" (g-t-r standing for
"generate-test-regenerate"), which the author derives from
the work of American biologist Richard C. Lewontin and
American psychologist Donald T. Campbell, and another of
which is the RIL ("replicator-interactor-lineage")
formulation of universal Darwinism, which Plotkin derives
from the work of Richard Dawkins and American philosopher of
biology David L. Hull. In addition to these scholars,
Plotkin credits a number of others who have produced work
along these lines, including philosopher of science Karl
Popper, sociologist Georg Simmel, physicist Ernst Mach,
ethologist Konrad Lorenz, psychologist B. F. Skinner, and
economist H. A. Simon (see p. 70). I suggest that anybody
who is interested in this subject also perform a Google
search on the phrase "evolutionary epistemology".

The primary thrust of the book, of course, is to explore the
applicability of some form of generalized Darwinian theory
to the characterization learning and intelligence in humans
and other animals with nervous systems as a "secondary
heuristic" of adaptation, working within the lifetime of the
individual and subsidiary to the "primary heuristic" which
has shaped the individual's phenotype. The following
quotes, which I have selected to give the flavor of the
author's arguments, are from the central chapter of the book
-- Chapter 5 "The Evolution of Intelligence":

p. 137

        [The] period of time between receiving a set of
        genetic instructions and their implementation
        through development to the point where those
        selfsame instructions might be returned, via
        reproduction, to the gene pool, has been given a
        variety of names... Konrad Lorenz... called it
        generational deadtime. It is a lag-time that is an
        invariant feature of any system whose construction
        takes time and which is based on a set of
        instructions that cannot be continually updated. In
        the case of sexually reproducing organisms, each
        organism is cast off and cut off from the gene pool
        with a fixed set of instructions that cannot be
        altered. It cannot "dip" back into the gene pool to
        augment these instructions if it finds they are not
        good enough. In this sense, genetically, it is on
        its own.

p. 144

        The uncertain futures problem concerns an organism
        going through life, equipped only with instructions
        given at conception (and hence perhaps only correct
        at that time) on how to survive, and having to
        interact with a world that may be different from
        that in which its life began. There are, of course,
        two possible outcomes -- the problem is solved or it
        is not... Extinction is always caused by an
        inability to cope with the uncertain futures
        problem, and given the estimates that well in excess
        of 98 per cent of all species that have ever existed
        are now extinct, it is clear that, in the end, all
        life succumbs to its uncertain future... None the
        less, [the] struggle to delay the inevitable
        succumbing can be relatively successful. There are,
        after all, species that have existed for millions of
        years.

p. 145

        [One way] for an organism to reduce the amount of
        significant change it has to deal with... is to
        reduce the period of time between conception and
        reproductive competence... This means that the
        ratio of life-span length to numbers of offspring is
        low... This is a characteristic 'life-style' of
        animals known to ecologists as r-strategists...
        These r-strategists usually live less than one
        year...; they develop rapidly; they are usually of
        small body size; and they normally reproduce over
        just a single period. This is the life-history
        strategy common to most invertebrate animals and
        clearly one that is... relatively successful...

p. 146

        [A] second very general way of dealing with
        change... [is to] change the phenotypes so they can
        change with and match the changing features of the
        world. This, in turn, comprises two general
        strategies. The one results in changes between
        phenotypes, and relies on the chance or radical
        component of the primary heuristic [the mechanism of
        classical Darwinian evolution, involving genetic
        variation and selection among competing
        phenotypes]... [Large] numbers of offspring are
        produced,... each... different from the others...
        r-strategists... often do combine a short life-span
        with a quite prominent radical component of the
        primary heuristic...

p. 147

        Another... strategy is to match change with
        change... by giving rise to change **within**
        phenotypes... I will call this the tracking option.

p. 149

        How can such changes be tracked? The only way to do
        it is to evolve tracking devices whose own states
        can be altered and held in that altered condition
        for the same period of time as the features of the
        world which they are tracking... Such tracking
        devices would be set in place by the usual
        evolutionary processes of the primary heuristic and
        hence would operate within certain limits... These
        additional knowledge-gaining devices comprise a...
        secondary... heuristic.

        ...[T]here are two such classes of semi-autonomous
        knowledge-gaining devices or secondary heuristics
        that can track change in this way. The immune
        system is one; the intelligence mechanisms of the
        brain are the other.

        ...Think of the world as comprising sets of
        features, some unchanging and others changing at
        different rates and with different degrees of
        regularity... [I]f the frequency of change is less
        than that set by generational deadtime for
        extracting genetic instructions from the gene pool
        and then returning them to it, then the conservative
        component of the primary heuristic will be able to
        'see' these changes and will furnish adaptatons to
        match them. But if the frequency of change is
        faster than the frequency set by generational
        deadtime, then though the primary heuristic will be
        able to see the long-term stabilities upon which
        these changes are superimposed and will be able to
        detect the margin within which these more rapid
        changes occur, in order to track the precise values
        of these changes the primary heuristic will have to
        evolve devices that operate at a much higher
        frequency -- at a frequency high enough to be able
        to track these values. If the high-frequency
        changes are unstable,... the tracking
        device... need [only] command an immediate
        compensatory response... However, if these
        changes... [are] short-term stabilities, then these
        tracking devices must comprise a secondary heuristic
        that is able to change and maintain new states that
        match those features of the world that are being
        tracked. Such brain mechanisms... are what we
        know... as rationality or intelligence.

p. 151

        In general, the kinds of animals that would evolve
        intelligent adaptations are relatively long-lived
        and produce relatively few offspring in their
        lifetime; that is, the ratio between the length of
        life and number of offspring is high. Such animals
        usually develop quite slowly and have a relatively
        large body size. They are what ecologists refer to
        as k-strategists...

p. 166

        What might [the] process be by which the secondary
        heuristic of intelligence gains knowledge? There
        are only two candidates... selection and
        instruction, corresponding to Darwinian and
        Lamarckian theories of evolution respectively. The
        principal difference between them is that
        selectionism involves an over-proliferation of
        entities, the generation of which is unconnected
        with the organism's need at the time (which is why
        the process is sometimes referred to as blind); a
        small number of them are conserved after they have
        been tested against the organism's requirements...
        Instructionalism, on the other hand, involves the
        production of an entity, a product of intelligence,
        after the need for it has arisen (which is why the
        process is considered to be directed); this process
        requires a malleable substrate that is moulded into
        an adaptive state by the environmental feature that
        forms the external end of the relationship. Put
        crudely, in the case of instruction the environment
        rules; in the case of selection, internal or
        organismic states lead.

p. 169

        [R]ight now we do not know how the brain
        [works]... in... sufficient detail... to be able to
        say... whether it works in a selectional or an
        instructional manner. Traditionally, the technical
        neuroscience literature has been based primarily
        upon instructionalist theory. However, in the last
        twenty years or so, elegant and powerful theorizing
        by the likes of Jean-Pierre Changeux in France and
        Gerald Edelman in the United States has shown that
        selectionist models of brain development and
        function are both plausible... and interesting...

        What, in terms of... brain states, might correspond
        to the replicators, interactors, and lineages of the
        secondary heuristic?... A replicator is a pattern
        of activity in a nerve sheet... and corresponds to a
        memory... [T]he pattern of activity that
        constitutes [a] memory... is not always present --
        it is usually a potential pattern rather than one
        that is actually present. This is just another way
        of saying what we all know to be the case, that we
        do not live with all of our memories activated all
        of the time. But they can be re-constituted,
        re-membered, re-plicated... Memories **are**
        replicators. They are the secondary brain heuristic
        analogues of genes... in the sense that memories
        are produced again and again as copies each time
        that the brain state in question is reconstituted --
        that is, each time we conjure up some specific
        memory.

p. 171

        [W]hy should the brain be seen as a Darwinian kind
        of machine rather than as a Lamarckian machine?...
        Forced to take sides,... there are two... reasons
        for choosing the selectionist camp. One is the
        problem of creativity...
        Intelligence... involves... the production of novel
        solutions to the problems posed by change --
        solutions that are not directly given in the
        experienced world... Such creativity cannot occur
        if change is slavishly tracked by instructionalist
        devices. So what we see here is that while
        selection can mimic instruction, the reverse is
        never true... Instructional intelligence comprises
        only what has been actually experienced... Indeed,
        according to D. T. Campbell, the father of modern
        evolutionary epistemology, selectional processes are
        required for the acquisition of any truly new
        knowledge about the world: 'In going beyond what is
        already known, one cannot but go blindly. If one
        goes wisely, this indicates already achieved wisdom
        of some general sort.' Instruction is never blind.
        Selection always has an element... of blindness in
        it. At the heart of all creative intelligence is a
        selectional process, no matter how many
        instructional processes are built on top of it.

        The [other] reason for choosing selection over
        instruction is one of parsimony and simplicity. If
        the primary heuristic works by selectional
        processes, which it most certainly does,... and if
        that other embodiment of the secondary heuristic
        that deals with our uncertain chemical futures,
        namely the immune system, works by selectional
        processes, which is now universally agreed, then why
        should one be so perverse as to back a different
        horse when it comes to intelligence?

        A nested hierarchy of selectional processes is a
        simple and elegant conception of the nature of
        knowledge. There will have to be good empirical
        reasons for abandoning it.

The article "Ghost in the Machine: What Neuroscientists and
Computer Scientists Can Learn from Each Other" at
http://www.doubletwist.com/news/columns/article.jhtml;$sessionid$WLUGKNIAAA5EBWBCHIVSFEQ?section=weekly01&name=weekly0130
(which Max More provided the link to in his earlier
message), suggests that interpreting the mechanisms of
learning and intelligence in the brain in Darwinian terms is
a "metaphor", and quotes to that effect Dr. Terrence
Sejnowski, director of the Computational Neurobiology
Laboratory at the Salk Institute: "But the Darwin metaphor
for the brain is limited, he adds. While Sejnowski doesn't
deny the scientific support for synaptic plasticity and
neurogenesis, he thinks the parallel to Darwinism is
inexact. 'If you want to be strict--if you don't just want
it to be a literary metaphor--there needs to be a process of
duplication. You need to take something that is successful,
make many copies, and mutate it. There's nothing that
corresponds to that in the brain.'"

Plotkin's book and other writings on evolutionary
epistemology are, in effect, an attempt to address
Sejnowski's dismissal as a mere metaphor the application of
the adjective "Darwinian" to the operation of the brain.
The attempt to create a theory of universal Darwinism, as
Plotkin defines the term, is precisely the attempt to put a
Darwinian characterization of the brain on a firm
theoretical foundation. One gets the impression, however,
that not all philosophers and scientists are convinced of
the respectability of this enterprise. Nevertheless, the
following remarks in Plotkin's text apply to Sejnowski's
objections: "[T]he way that evolutionary processes actually
do occur in the immune system, in the brain or at the level
of the species is still largely unknown. But there is no
reason not to expect considerable differences of detail.
**The actual mechanisms in each case**, of course -- and one
cannot repeat this point often enough -- **are entirely
different**" (Chapter 3 "Universal Darwinism", p. 100).
"[A] brain state can be reconstituted when such a state was
not actually present all the time in the past. This
characteristic of neural replication, of making a copy, is
different from the replication of genes. In the latter
case, a specific chemical structure is constantly present
and, at certain points in the cycle of the interactors
carrying those genes, copies, literal copies, are made of
the 'original'. Clearly neural replication is different.
Here copies of brain states present some time in the past
are reconstituted, such structures being only potentially
present, not actually so, in the intervening time. How such
actual brain states are re-established from potential brain
states is, for the present, a process whose details are
entirely unknown by brain scientists. We simply do not know
how it works" (Chapter 6 "Aspects of Human Knowledge",
p. 219).

In other chapters, Plotkin acknowledges Dawkins' invention
of the term "meme", and describes attempts to apply
universal Darwinism to cultural evolution. Plotkin admits
that such attempts are even less precise and more
speculative than the attempt to characterize learning and
intelligence as the subsidiary, secondary heuristic in a
hierarchy of Darwinian evolutionary processes whose primary
heuristic is the alteration of the phenotype through
successive generations. Nevertheless, Plotkin asserts that
it may someday be theoretically justifiable to characterize
cultural evolution as a "tertiary heuristic" operating
subsidiarily to the other two (see p. 225). Plotkin also
mentions Karl Popper's characterization of the evolution of
scientific knowledge in Darwinian terms, and remarks "As
Popper himself stresses, his view is not meant merely to be
understood metaphorically. He means it literally: 'All this
may be expressed by saying that the growth of our knowledge
is the result of a process closely resembling what Darwin
called "natural selection"; that is, **the natural selection
of hypotheses**... From the amoeba to Einstein, the growth
of knowledge is always the same: we try to solve our
problems, and to obtain, by a process of elimination,
something approaching adequacy to our tentative solutions.'"
(pp. 69-70).

In the final chapter of the book, the author applies the
evolutionary view of the human capacity for knowledge
developed in earlier chapters to illuminate the arguments of
philosophers Immanuel Kant and David Hume. Kant wrote that
some knowledge, such as the understanding of mathematical
relationships and of causality, derives both from
observation of the world and from built-in intuitions or
propensities to take advantage of the relevant observations
-- the so-called a priori synthetic. Plotkin interprets
this by bringing to bear the conclusion drawn earlier in the
book that the human mind is not a tabula rasa, but is
prepared by the "primary heuristic" of evolution in ways
that give the "secondary heuristic" a head start, so to
speak. People seem to be born into the world knowing, in
some sense, what it is they need to learn. Several lines of
evidence for this priming are presented from experimental
psychology; linguists, of course, have acknowledged since
Chomsky that humans are predisposed from birth to learn
speech with astonishing ease and rapidity. While Kant
interpreted this a priori knowledge as a barrier between the
human intellect and the real world, an evolutionary
perspective interprets this a priori knowledge as the result
of human minds having been adapted by evolution to learn
about the world: the secondary heuristic of intelligence,
which makes philosophical contemplation and discourse
possible, owes its existence in the first place to aeons of
adaptive sculpting by the operation of the primary
heuristic.

Scottish philosopher David Hume shocked his contemporaries
by denying the possibility of certainty; again, in an
evolutionary interpretation, this is simply a statement of
the "uncertain futures" problem, a fact about the world
which makes intellect adaptive in the first place. In the
following passage (from Chapter 7 "The Philosophical
Problems in Perspective", p. 243), Plotkin alludes to the
arguments of Hume, who criticized inductive inference
because the "principal of the uniformity of nature" on which
it is based is not a logical truth. This passage also
contains a capsule summary of Plotkin's view of the
relationship between inter-organismic and intra-organismic
evolutionary processes (what are called throughout the book
the primary and secondary heuristics):

        In the face of this Humean uncertainty -- Humean
        because modern science can now interpret what the
        philosopher was pointing to as a factual condition
        of the universe and not merely a logical point --
        nature has come up with an elegant and effective
        response. Unable to rely upon just one level of
        evolution and one unit of selection as the means of
        gaining knowledge about just one range of
        frequencies of change (that range being limited at
        one end by the change becoming undetectably slow to
        the point that survival is not threatened by a
        failure to respond to it, and at the other by the
        generational deadtime of each species), subsidiary
        evolutionary processes have evolved, each with its
        own units of selection, and each able to gain
        knowledge about changes that are occurring at
        ever-higher frequencies. So in the real world the
        Humean uncertainty is converted into a pragmatic
        issue of dividing the world into band widths of
        frequencies of change and fluctuation, and emplying
        knowledge-gaining mechanisms that are able to match
        the rates of perturbation of the world with organic
        structures able to alter their own states at
        equivalent rates.

        Three and a half billion years of life on Earth tell
        us that though individuals, individual species, and
        whole larger groupings of living forms may come and
        go precisely because Humean uncertainty can be
        literally fatal, the biotic system as a whole
        endures, being rather adept at solving the problem.
        Life really is quite good at the knowledge game.

Since Plotkin mentions Gerald M. Edelman and Jean-Pierre
Changeux in the same breath, I ordered a book by the
latter: _L'Homme Neuronal_ (Fayard, 1983). I have never
before attempted to read a serious book in French
(never having gotten much beyond _Le Petit Prince_, and that
was a long time ago), so I may have bitten off more than I
can chew. Nevertheless, I still get a thrill from being
able to use the Web for international shopping -- I ordered
the book from a Paris on-line bookstore,
http://www.chapitre.com

Cheers!

Jim F.



This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:32:01 MST