Re: And is _this_ true?

Steve Goldfarb <slg@xxxxxxxxx> wrote:
So the question is why Micha insists that you can't apply the genetic
algorithm to genetics. He hasn't made that clear, and his failure to make
that distinction has caused this entire discussion to be mostly noise.
Yes, you've alluded to that point but it's such a radical POV that I think
you needed to be much more explicit that that's what you were saying.

I've been saying that evolution isn't a GA WRT producing thought,
because its fitness function has nothing to do with thinking. Or any
other eventual goal; not even future survival. (A point you yourself
made repeatedly.) It's all tactics, what keeps you alive NOW.

In a Genetic Algorithm written as software, the fitness function
is artificial, and measures closeness to the eventual, future,
goal. Evolution has no eventual goal, no strategy for future generations.

Thus, until /after/ thought evolves as a survival mechanism, there is
nothing to make thought more likely than any other survival technique.

Even if we come up with some proto-thought, there is no reason why
something closer to thought is more a likely strategy than something
further away. And so, not only do you have to deal with the odds of
proto-thought exactly in the same fashion as the original question,
you then haven't changed the odds of the rest of the trek. This was my
metaphor of the state of the world after 90 out of 100 coin tosses.


Another way of saying that thought is no more likely than any other
survival technique is saying that invoking evolution to explain the
history of how it arose says nothing about the odds of thought being
the actual winning answer. You might as well simply analyze the odds in
terms of the straight bit set -- you'll get the same answer.

You are using a system that has no statistical correlation to the
result we're analyzing, and trying to convince me that it explained the
likelihood of that result. It is comparable to picking balls by size
and explaining the details as though that changes the odds of the color
of the ball picked. You might as well just count how many of the large
balls are red and ignore the selection process.

And those who engage in Automata Theory and Artificial Life research
acknowledge grapple with this problem. There is no explanation for the
rise of new automata. (Automaton: (roughly) a set of interacting parts
that work together in accomplishing a single function.) By our models
of statistics, even with evolution, they shouldn't be arising as often
as they do.


You'll notice a similarity between my description of automata and Behe's
irreducible complexity. The difference is that he is claiming that
evolution can not produce them. And thus disproving his thesis is easy.
I am arguing that evolution is producing them far more frequently than
the law of large numbers would lead me to expect. (Law of large numbers:
(roughly) if you flip a coin many many times, you should get closer and
closer to 1/2 of your tosses being heads. Or a die landing on 6 around
1/6th of the time.)

I happen to believe this is why popularizations of evolution tend to
fall into anthropomorphications like, "Beatles evolved the ability to
fly in order to..." Evolution has no "in order to", but there is little
way to explain it *plausibly* without invoking the notion. So, the lazy
writer drifts to those terms anyway.


And don't try to say that you were only talking about the unlikeliness of
sentience, because you later argued that the development of even one
enzyme via evolution is a statistical impossiblity.

I clarified, once I clarified it to myself, that this is essentially
a distinct bit of argument. If the first bit of this post explains why
looking at the process of how we combine our die rolls is irrrelevent,
this part explains why I feel it's fair to treat the existence of this
enzyme or that as the unit, the die roll, under consideration.

The randomization is on a DNA mutation level. The filtering of natural
selection is on the protein/enzyme level or above. Similar segments of
DNAs do not produce more simular enzymes. Very different strings that
produce similar proteins will survive just as well.

So, you can take the smallest string of DNA that could specify any one
protein and treat it as the "size" of your die. Each randomization is
of that configuration of nucleotides or longer.

(The mathematical similarity of the two arguments is that both are showing
a lack of statistical correlation between the filtering and the result
it is being invoked to explain.)

Now I ask: Rolling every die possible at the same time, how long does it
take before I would "expect" a die roll produces one meaningful result? I
looked at that question, and the answer doesn't fit in the age of the
universe.

(Using "expectation" here in the statistical sense, but the numbers of
"tosses" involved are so large, the odds of defying expectation by any
margin are very very tiny.)

Another way of saying it is that the number of mutations that actually
aid survival are so few that by the time there is an organism with a
single useful new protein to actually go through the filtering process
of surviving better than its ancestors you already exhausted the age of
the universe.

Without getting to evolution's successive iterations.

:-)BBii!
-Micha

--
Micha Berger You will never "find" time for anything.
micha@xxxxxxxxxxx If you want time, you must make it.
http://www.aishdas.org - Charles Buxton
Fax: (270) 514-1507
.

Relevant Pages

  • Re: Lennys Counter Argument
    ... There is extensive evidence on evolution way, ... Bullshit, Sean. ... Where's the statistics in Einstein's seminal paper on relativity? ...
    (talk.origins)
  • Re: Revised Tautology FAQ - Thread-5
    ... He has foisted evolution ... point has nothing to do with whether natural selection is a tautology. ... misconstrued as the fittest surviving? ...
    (talk.origins)
  • Re: Revised Tautology FAQ - Thread-5
    ... Who is to say if a stronger species did not die out to allow survival ... He has foisted evolution ... point has nothing to do with whether natural selection is a tautology. ... misconstrued as the fittest surviving? ...
    (talk.origins)
  • Re: Ben G on reinforcement-learning and the wirehead problem
    ... Evolution has the goal of survival, ... intelligent machines becuase they are good at surviving. ... rewards to motivate it to survive, such as the body damage sensors. ...
    (comp.ai.philosophy)
  • Re: Why didnt at least one dinosaur species become human-level smart?
    ... I'd suggest that you're actually mistaken to think there's a huge gap. ... But I take that stance when I argue about what I think the brains are doing ... In the history of the evolution of life, ... parents - everything we would "learn" about surviving before we became ...
    (comp.ai.philosophy)
Loading