Re: (non-flame post) good chess programs intelligent?

From: Chris Cogan (ccogan@telepath.com)
Date: Tue Oct 10 2000 - 14:10:06 EDT

  • Next message: Bertvan@aol.com: "Definition of intelligence"

    Chris
    Interesting and valuable post.
    <snip to get to chase>
    >DNAunion
    >So, after all this, what are others&#8217; views on whether a chess
    >computer possesses some form of intelligence? Have human intelligent
    >agents already created a new form of intelligence that is not biological?

    Chris
    I agree. I think we need to separate intelligence (at least of a sort) from
    consciousness, though consciousness is, in a sense, a kind of intelligence
    (self-awareness, etc.) or at least feeds important data into intelligence
    for it to take advantage of. Computer chess programs *are* intelligence by
    proxy. Interestingly, we can write chess programs that can "play" better
    chess than we can, because we can progressively refine the intelligence
    that goes into the program and that becomes used during play beyond a level
    that we ourselves are normally capable of in a reasonable time (although,
    under time-limit rules, the programs "intelligence" will vary according to
    the speed of the processor and memory and disk, etc.). This suggests that,
    at least in principle, we need to consider intelligence in both
    time-dependent and time-independent senses. We can conceive of minds
    similar to ours that process data much more slowly but that, given
    sufficient time, would generally come to the same or even better results.
    But, in practice, human life imposes restrictions that tend to lead us to
    judge intelligence by the quality of results achieved rather quickly,
    rather than by the same quality of results achieved more slowly. This is
    not always a fair way to evaluate degree of intelligence.

    If life on Earth were initiated by a computer-controlled machine that was
    not itself conscious but which had been well-programmed by very
    knowledgeable biological engineers, then I'd agree that this would be
    intelligent design by proxy, since at least the basic relevant knowledge of
    the designers would have been distilled into the computer program. Such a
    program might not be able to cope with situations that the engineers would
    have been able to deal with readily, but, within its design-scope, it would
    be intelligent.

    Computer chess program intelligence is structured somewhat like conceptual
    intelligence, but it is not the same. In a way, it is "formalized" and
    distilled intelligence. The program uses a minimal or near-minimal amount
    of information about what it's doing, but that information is carefully
    selected to be just the information it needs in order to perform the task
    mindlessly (but intelligently). I don't want to push this kind of
    "intelligence" too far, but I'm willing to grant that it is a kind of real
    intelligence, though some would argue that only the more general
    information processing of current intelligent biological organisms should
    be considered intelligence, because it has so much greater contextual
    ability than do computer programs; it recognizes not only the chess
    position, but (perhaps) the mood of the opponent, the actual shape and
    colors of the pieces, whether they are centered on their squares or not,
    whether the opponent seems listless and uninterested, etc. We do not yet
    have computer programs that can approach this level of functioning (partly
    because of severe hardware limitations; a recent Scientific American
    article estimated that we won't have computers that can truly match the
    sheer hardware processing power of the brain for another fifty years, even
    at an exponential growth rate in processing power per computer. The retina
    of one eye alone does an enormous amount of processing; we would need very
    many of today's current processors to do not only that processing but the
    processing behind the other senses, and so on. This is one main reason why
    nearly all of today's computer programs that do anything "intelligent"
    (drop the quotes?) do so by means of a great deal of abstracting of the
    task requirements from other factors that a normal mind would be also
    processing information about.

    The reason why I reject this evaluation of intelligence with respect to a
    pure "evolutionary algorithm" is that there is no planning or goal setting
    or any such activity in such an algorithm; it is not even as "smart" as an
    exhaustive-search Tic-Tack-Toe program. The evolutionary algorithm can be
    carried out by a program that does not even use a primarily algorithmic
    method to select variations to impose on replications (it could use
    cosmic-ray counts, for example). Further, the function of the algorithm is
    precisely to emulate processes that, in Nature, are unintelligent. Although
    a computer must perform some operations on data to "select" a "genotype"
    for removal from the population, this is only necessary because the entire
    activity is being performed on information, and in a specifically
    information-processing environment. To mimic even basic physics, *we* have
    to write algorithmic code (because the "physics" of the computer's
    information-processing (not the electronics) is completely different from
    real physics).

    Further, selection does not need to be allowed to determine what variations
    are introduced. Thus, selection may only limit the basis for future
    variations by excluding from the current generation those "genotypes" that
    would lead to *different* future genotypes from the ones that will come
    from the surviving genotypes, but it can be excluded from having any direct
    effect. Selection is, strictly, *inherently* non-creative. It *limits* what
    can be created by limiting the population of currently-functional genotypes
    to ones that have certain characteristics, but does not *create* the next
    generation of genotypes or their variations.

    This means that we can consider variation and selection quite separately
    (initially, at least; the consequences of selection in the real world are
    not as simple as indicated here, because of the shifting of resources, the
    production of dead organisms that other organisms can feed off of, and so
    on, so that selection changes the selection factors on the fly).

    The difference between intelligence and unintelligence in this case is that
    intelligence *pre-*selects components or aspects of what is being built,
    out of a number of things that *might* be tried. The evolutionary algorithm
    *post-*selects, *after* unintelligent variations have been made. Further,
    the selections may be as stupid as such physical factors as excessive
    temperature. We certainly would not accuse the weather of intelligently
    selecting out organisms/genotypes that cannot survive in the heat (or cold,
    etc.), and we should not attribute intelligence (in the relevant sense) to
    a data processing environment that, for example, arbitrarily "kills" off
    data strings that mechanically meet some criterion analogous to an
    over-warm environment.



    This archive was generated by hypermail 2b29 : Tue Oct 10 2000 - 14:15:04 EDT