Re: And is _this_ true?

Steve Goldfarb <slg@xxxxxxxxx> wrote:
If you get the problem to that form, are are theorems about probability
of systems arising randomly. I'm trying to give forms of those to an
audience that didn't study Information and Automata theories.

Are we talking about a system that arose randomly, though? I think that's
part of the problem with your argument. It's not taking into account the
ordering principles produced by environmental forces.

1- That's still random. It just means that there are options more probable
than others.
2- Those forces to not select based on the future, which drives my whole
thesis -- the system is no more probable than unfiltered randomness,
and since the filter reduces the number of possible paths to the result
in question, it actually LOWERS probability.

The part about the probability of producing a single enzyme can stand
on its own. Evolution doesn't touch it. The production of a single
enzyme is an all-or-nothing, totally unrelated to what was there before
and what enzymes it produced. A slighly different DNA sewquence will
likely produce nothing, but if it manages to foster the production of an
enzyme, it would be a different enzyme with totally unrelated chemical
properties. Similarity of DNA doesn't translate to similarity of effect.
One increasing fitness for survival has nothing to do with whether another
produced with similar DNA will increase or decrease fitness. They are
toally uncorrelacted. Each enzyme stands alone, from a statistical point
of view.

That's a hypothesis, not a conclusion..

No, it's experimental data.

Further, you're asking "what's the
probability of producing an enzyme" when producing enzymes was never part
of the goal. Survival was the goal, and it so happened that a particular
sequence of proteins that we call an "enzyme" worked.

Cellular life is based on proteins. Similarly DNA coded proteins are
behaviorally unrelated. Thus, once you get to the bateria level or beyond,
each "letter" in the pattern produced is a protein. Enzymes are just
typical small proteins of known size and encoding.

If we liken the question to that of moneys hitting typewriters and
producing a work of Shakespear's one enzyme corresponds to one letter
in the sequence.

You're arguing that evolution is like giving those monkeys rules which
make Shakesperean plays more likely. But whatever the rules are for which
sequences of letters remain around for future monkeys to tinker with,
the probability of each keystroke remains unchanged.

No, once again your model is wrong. That's not how it works. The letters
are not unrelated - there IS a structure. Evolution is happening at a
structural level, not at an individual letter level.

But we're still at the probability of ONE letter!

It's not simply random variation, either.

Actually, it is. Then there is a filter as to which variations survive.
And that filter has nothing to do with the result -- later emergence of
intellect. Thus, the filter doesn't increase the odds of the result.

....
Evolution's filter is for what survives now. That has nothing to do
with which possibility offers the most opportunities for survival in
the future.

Demonstrably non-true. Organisms which reproduce are more likely to
reproduce than those which do not...

And what does that have to do with producing organisms with different
mechanisms for survival? Did you read the post full through before
replying? You're still talking tactic, not strategy.

This is unlike genetic programming. There, fitness for survival is
measured in distance from solving the final problem. The correlation is
as good as the quality of the fitness function. So, a genetic algorithm
to produce intelligence could in principle produce it -- but requires
design -- Someone knowing the goal and aiming things there.

I had a class on artificial intelligence with John Holland, who was one of
the inventors of the idea of genetic programming. The systems that he'd
built didn't have any knowledge of a final problem, and they
self-organized anyway. It's been a long time since I've dealt with any of
that, but I don't think you're describing it correctly.

I helped someone make money using a genetic algorithm to trade stocks.

FWIW, just check wikipedia. The fitness function tries to measure
distance from desired result. GA is used when you know the result
well, can define the individual steps well, but can't figure out what
sequence of steps gets you to the result.

Look up fitness functions. They are a measure of distance from a goal.
From Wikipedia <http://en.wikipedia.org/wiki/Fitness_function>: "An
ideal fitness function correlates closely with the algorithm's goal,
and yet may be computed quickly. Speed of execution is very important,
as a typical genetic algorithm must be iterated many, many times in
order to produce a usable result for a non-trivial problem."

GA is goal oriented; you're assuming that evolution can produce thought
in a length of time like the age of the universe even if not.

But it all hinges on the lack of correlation. That flippers changing
into legs is more likely than legs existing altogether. And as Herman
pointed out, the odds of having flippers that can evolve into legs is
similar to the odds of having flippers at all -- both require symmetry.
But one still needs to consider the odds of having flippers (the first
90 tosses) and the odds of them evolving into legs. That's just one way
of getting 50 heads and 50 tails among many. How can it be higher than
the odds of getting exactly half of the tosses heads in any pattern?

Once again, your model isn't right. It's not strings of bits getting
randomly flipped over until they form some configuration that you have in
mind. It's structures - physical structures - that interact with each
other in various ways...

No, the fioltering is about structures. The randomness is primarily
about cosmic rays hitting DNA molecules. And the filtering is for
survival today, as per above.

It all boils down to the observation that flippers aren't selected
because of their usability as legs, therefore the possibility of
flippers doesn't increase the probability of legs.

No, this is completely wrong. You're making several mistakes. For one
thing, traits aren't actually selected for, what happens is everything
without that trait is selected against. There's a difference...

Define the difference between saying the fittest survive more frequently
and the less fit survive less frequently.

"flippers" per se weren't actually part of the equation, neither were
"legs." What was being selected against was the inability to have
sufficient mobility of a specific type...

Still, mobility at time A isn't selected for a different kind of mobility
at time B. (And actually it's survival at time A, which happens to
involve one kind of mobility, vs survival enhanced by another kind of
mobility at a later time B.) Therefore, if we're looking at the odds
of that different kind of mobility at time B, that earlier form doesn't
increase the odds. You're ignoring that producing B's reality requires
the odds of the reality at time A times the transition from A to B. Which
has the probability of B as an upper bound.

....
Once again, what do coin tosses have to do with anything? As I said
before, evolution is a process that incorporates probability, but I don't
think it's reasonable to categorize it as a random process.

Everything. Randomness is coin tosses. A nice metaphor for repeated
randomnizatiohn of all types. With or without weighted coins, as long
as you aren't arguing a correlation between one result and the next.

And since we all agree there is no goal to evolution without invoking G-d,
there is no correlation between the effectiveness of mutations without
G-d. It's a bunch of sequential randomness.

Tir'u baTov!
-Micha

--
Micha Berger I slept and dreamt that life was joy.
micha@xxxxxxxxxxx I awoke and found that life was duty.
http://www.aishdas.org I worked and, behold -- duty is joy.
Fax: (270) 514-1507 - Rabindranath Tagore
.

Relevant Pages

  • Re: Generating electricity?
    ... distinct probability distribution. ... Electrons also exhibit "randomness" and probability distribution. ... fair coin coming up exactly 5 heads or 100 fair tosses of a fair coin ... Nobody has claimed that the "prediction" will be right 100% of the time, ...
    (rec.martial-arts)
  • Re: Tedious bridge randomness technicalities
    ... test hyper geometric progressions for randomness ... See small sampling reference above. ... I initially assumed 1/52 probability impact ... Or you can move to larger sample sizes. ...
    (rec.games.bridge)
  • Re: The sense of Continuum
    ... The distance can not get any quantity,and it includes ... Here, the notion of partial randomness, and partial disorder can be used to ... have a %100 probability of existing because they contain no uncertainty. ... due to the topological indeterminacy of the term ~b. ...
    (sci.math)
  • Re: Whats a boy to do?
    ... Expermental probability is about expected outcomes observed over multiple ... so not only does randomness ... Probability is not the right analysis for a singular event. ...
    (rec.outdoors.fishing.fly)