Dr Rick

The answer to the last part is illustrated in the following by Vander:

Anyone who can read and follow the long and often technical posts knows better.

Rick

phlebas

However, I wouldn't say that Deanne is in any way a "match" for Dembski...

Well, I certainly wouldn't say they are evenly matched.

To paraphrase my SATs from many years ago:

Deanne is to Dembski as sand blaster is to soda cracker.

Undercurrent

I'm not entirely versed in the NFL theorems, but by my understanding, this is a faulty argument.

At least in the orginal NFL paper, the formal optimization algorithm described is just a machine that, given a few function evaluations at a number of points, picks (perhaps stochastically) a new point to evaluate the function at. The optimization process is to evaluate the function at a point, use the machine to pick a new "interesting" point, evaluate at the new point, use the machine to pick another point, &c.

These are the types of optimizers, that the original NFL theorem applies to. Averaged over all landscapes, for the same number of iterations, any such optimizer is no better than any other.

The evolutionary process is an example of this kind of optimizer. The points to evaluate the utility function at are the genomes of the various individuals, the act of function evaluation is the spawning of an offspring with a given genome and seeing how it survives, and the machine that decides what new points (genomes) to test is the mutation/recombination process.

A designer could solve around the NFL problem in two ways: (1) by using an optimizer not of this type (or some other type that another NFL theorem applies to) or (2) by evaluating a larger number of points than the evolutionary process does (really fast computer, perhaps.)

So, if NFL actually was a big problem for evolution, it could be gotten around by positing a designer.

Dembski's argument that NFL is a big problem for evolution is something like:

1) Averaged over all utility functions, any optimization algorithm only as good as random selection. Specifically, the evolutionary process is only as good as random selection averaged over all utility functions.

2) Therefore, utility functions where evolution is better than random search are hard to come by.

3) Random search is not good enough to produce the diversity and nearness to optimality of life we see.

4) Therefore, a evolution could not be responsible for the diversity and nearness to optimality of life we see without a carefully selected utility function.

5) Therefore a designer did it.

The problem with this is that 2 does not follow from 1. Desipte the fact that, averaged over all utility functions evolution is no better than random search, it is fairly easy to find specific utility functions where evolution does do much better than random search. This is evidenced by the success of evolutionary algorithms for doing optimization in the "real world" (e.g. airfoil design, orbital transits, &c.)

It is, in fact, terribly easy to come up with utility functions where evolution outperforms RS, and that is why NFL is so shocking. It says that despite the fact that we see these algorithms working every day, on average, they are no better than RS.

m.

Automaton

You're like a virgin trying to point out that others can not get laid.

Jesse

Undercurrent:
It is, in fact, terribly easy to come up with utility functions where evolution outperforms RS, and that is why NFL is so shocking. It says that despite the fact that we see these algorithms working every day, on average, they are no better than RS.

I don't think it's necessarily that shocking. The vast vast majority of "all possible fitness landscapes" would have little or no correlation between genetic distance and fitness--changing a single bit of the gene sequence would, on average, have just as large an effect as changing every single one. Natural selection requires relatively smooth fitness gradients in order to have an advantage over random search. To me, the mere fact that we live in a universe with regular laws where similar organisms are more likely to have similar fitness than completely dissimilar ones shows that the NFL theorem has no real relevance to biological evolution.

Doubting Didymus

If anyone here with the time and patience to frequent dembskis forum would be so kind as to inform us here in the event of dembski eventually posting a substantial reply, I would be most grateful.

Thank you.

Jesus Tap-Dancin' Christ

Undercurrent:

Oh. Well, I'm not that terribly well versed in them either, it was just a random thought that entered my head.

Bubba

Don't hold your breath.

Bubba

Albion

Oh, hasn't Vanderzyden already done that? You're so ungrateful!

rebelnerd

I just got done reading the entire thread. Whew!
Dembski's last post: 9/22
Deanne's 9/30

Her last two posts were directed mainly in reply to Janitor@MIT and jasonyoung. Janitor@MIT said he was calling it quits at the bottom of page 6. jasonyoung just was talking out his/her backside and was corrected by Deanne. Dembski's last post (9/22) had nothing of substance, mostly just trying to wave away Deanne's points.

Filo