From: Ben Goertzel (ben@goertzel.org)
Date: Sun Aug 25 2002 - 15:25:34 MDT
Gordon wrote:
> On Sunday, August 25, 2002, at 04:08 PM, Cliff Stabbert wrote:
>
> > Our bodies and minds represent millions years of *successful*
> > evolution. We are, after all, the ones who made it.
>
> Evolution only has to be successful enough. It does not have to find
> the right answer.
No finite system can find "the right answer" to all problems. Algorithmic
information theory has proved this quite clearly.
But your comment does indirectly pertain to some interesting issues...
I agree that evolution has not provided optimal solutions to all problems!!
The confluence of our sexual and excretory organs is an oft-cited
demonstration of this ;)
I also feel that evolution is far from an optimal problem-solving heuristic.
In fact, recent computer science work shows that even in cases of "black box
optimization" where the GA (a computer simulation of evolution) has
typically excelled, one does much more efficient problem-solving using
probabilistic methods such as Goldberg and Pelikan's "Bayesian Optimization
Algorithm". (Search for "BOA Goldberg Pelikan" for the technical
references). But these algorithms require a centralized controller whereas
evolution is a distributed process... The BOA algorithm is like having a
probabilist analyze the genomes of all organisms on the planet, and reason
about which combinations of genes tend to provide fitness, and create new
genomes for new organisms accordingly. This is more efficient than
evolution with its messy crossover & mutation operators, but it's not an
option Nature had [unless you want to posit a new kind of Bayesian
scientific creationism ;) ].
In short, there is some meat to the claim that the evolutionary process is
an inefficient approximation to Bayesian optimization...
But the story's not done yet.... Genetic drift played a big role in the
origin of complex, intelligent life -- selection wasn't the whole story.
There's reason to believe that evolution harmonizes better with the
self-organizing processes that cause some genetic drift, than BOA type
processes do.... Thus some mixture of evolutionary and explicitly
probabilistic approaches may be optimal for creative evolution...
This was a digression from our thread on rationality, though ;)
-- Ben
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:40 MDT