… with Mung. In a recent comment Mung asserted that
If Darwinists had to put up their hard earned money they would soon go broke and Darwinism would be long dead. I have a standing $10,000 challenge here at TSZ that no one has ever taken me up on.
Now, I don’t have \$10,000 to bet on anything, but it is worth exploring what bet Mung was making. Perhaps a bet of a lower amount could be negotiated, so it is worth trying to figure out what the issue was.
Mung’s original challenge will be found here. It was in a thread in which I had proposed a bet of \$100 that a Weasel program would do much better than random sampling. When people there started talking about whether enough money could be found to take Mung up on the bet, they assumed that it was a simple raising of the stake for my bet. But Mung said here:
You want to wager over something that was never in dispute?
Why not offer a meaningful wager?
So apparently Mung was offering a bet on something else.
I think I have a little insight on what was the “meaningful wager”, or at least on what issue. It would lead us to a rather extraordinary bet. Let me explain below the fold …
Mung accepted that Weasel programs reach their goal far faster than random sampling. However Mung also said (here) that
Weasel programs perform better than blind search because they are guided. I didn’t think the performance was in dispute, nor why the performance was better.
and elsewhere characterized Weasels as succeeding because of “intelligence” as opposed to ignorance.
So let’s imagine what might happen if we took Mung up on the \$10,000 bet. We would bet that the Weasel would succeed, because of cumulative selection. Mung would bet that (because of “intelligence” or being “guided”) the Weasel would succeed. The stake would be held by a house of some sort, which would not take a commission.
The Weasel would be run. It would succeed. So the house would declare that we had all won. The stake would be given to the bettors, in proportion to their bets. But alas, no one actually bet against the Weasel. So the winnings would be zero. Everyone, Mung and the rest of us, would get their stake back, and that’s all.
To bet against Mung, we have to come up with some event that distinguishes cumulative selection from “intelligence” (or being “guided”). That seems to be the issue on which Mung was offering a \$10,000 bet, and declaring (here) us all to be “pretender[s]” because we would not put up or shut up.
So there it is. We’re all betting on the same side, and no one will win or lose a penny. Unless Mung can come up with some test that distinguished “intelligence” or being “guided” from cumulative selection.
Now I am possibly misunderstanding what the bet actually would be. I hope that Mung will straighten us out on that, so that we can understand what test is proposed, and place our bets.
Finally!
Mung,
This is the equivocation on the word ‘random’ that I pointed to in my first post. You seem to think it applies only to the neutral case – where there is no fitness differential (or rather, it’s zero). There’s also the possibility of confusion with ‘aimless’.
So, you could write a program that behaved entirely ‘at random’. Great. What would that show?
Oh please. The deny everything approach. How quaint.
It would show that cumulative selection is not the cause of the success of the Weasel program.
Mung,
Well, was it?
Mung,
By taking selection out of the program entirely – or preventing it from being cumulative, by using an ever-shifting target – you would prove that cumulative selection is not the cause of ‘success’ in a program that has them.
Which part of “I don’t know what you mean by cumulative selection” do you not understand?
Mung,
Which part of “cumulative selection” do you not understand? Selection is a bias favouring some genotypes over others in survival/reproduction. ‘Cumulative’ adds in iteration of this process over successive generations.
How about this: The “target” phrase is initially generated at random, and then it can subsequently randomly change at a random number of positions, every random number of generations.
This would then reflect the “random” changes of the environment that selection has to track, instead of some pre-specified static target (initially randomly generated or not). Like the environments in nature where continents move, rivers slowly carve new paths, changes in solar output slowly change the climactic conditions and so on and so forth.
And we can just stop the program at some arbitrarily picked number of generations and then compare the results to a random string generator.
My bet is this infused-with-randomness WEASEL will still beat a blind random seach without selection. By many orders of magnitude.
That part where the algo loops producing descendants with some random mutations and selecting the fittest ones maybe?
We now know the other components can be random, so if the weasel works because it’s not purely random , the only candidate left is selection: selection is the reason why the weasel works.
It wasn’t that hard, was it?
Of course. Now you tell me. Do you really think that no part of Weasel was meant to be like real life or that Weasel was intended to be nothing like real life?
Which parts of the program need to not be random for cumulative selection to succeed? I think you’ve come closer than anyone else so far in understanding the problem.
So we write this program that does everything at random that a Weasel program would need to succeed except we leave out the ‘selection’ part and we leave out the ‘cumulative’ part.
Now we add the ‘selection’ part but continue to leave out the ‘cumulative’ part. What would we need to add to the program in order to accomplish that, and would the program still succeed?
Next we add the ‘cumulative’ part. What would we need to add to the program in order to accomplish that, and would the program then succeed?
Mung,
It was intended to show the difference between a process of replication with selection, and one without. In that respect only, it is like real life. The ‘halting problem’ you alluded to is certainly not an issue. I don’t know why you brought it up.
I’d rather drop that word entirely from the discussion to be honest. It is a minefield of equivocation.
If you are of the opinion (which I used to share, and Dawkins still does) that selection is ‘non-random’, then you can’t put selection in a random Weasel and still have it be random – cumulatively or otherwise.
But better, I think, to say what we actually mean. Random targets, random mutations, the random element of selection (ie drift), and the random wanderings of a shifting target are different things connected by a single, overworked word.
Let’s call that Iterative Selection, because that’s all you’ve actually described.
So it’s the cumulative part I don’t understand. Actually, I do understand it. All too well. Getting people to admit to it is another problem entirely. In the Weasel program what is it that accumulates? It’s the number of letters matching the target phrase. Can we at least agree on that?
Generate the target phrase at random and what accumulates is still the number of letters matching the target phrase.
ETA: So you’re another proponent of cumulative random genetic drift?
Haha.
Well, just because a designer decides when to stop a program, and just because the designer decides that what now exists is the target, that doesn’t mean it was designed.
In some people’s mind.
Mung,
Through the process of iterative selection, absolute fitness rises. It does so cumulatively.
So cumulative selection can only work if we know the target beforehand?
Isn’t that what Mung has been telling you all along?
Mung,
I don’t understand why they can’t have cumulative selection first, and then chose a target phrase at random second.
phoodoo,
No. I already gave an example where the computer picks the target, and doesn’t tell us. You do need some element of consistency, but the Weasel will easily track a shifting target. It rather depends how fast it shifts – rather like mutation rate. NS won’t work in a capricious environment.
Nope.
phoodoo,
In the example I gave – ‘random’ target picked at runtime – that’s exactly what happens, if I understand you correctly. The program’s already written when the target is established – including the selection routine.
Allan Miller,
So it relies on the computers ability to keep a secret?
phoodoo,
Not critically, no.
But how do you know the computer won’t change its mind, and make a new phrase? Or perhaps it never made the random phrase at all to begin with, because it didn’t feel it was necessary?
Or maybe it made a mistake and picked the wrong one.
No, we don’t have to know what the target is. At all.
If the computer is honest you mean?
That’s like saying that you don’t understand how you can’t have gradual adaptation first, and then have an environment later.
In the WEASEL, the target phrase is analogous so the environment. That’s what selection is adapting the organism to. In the WEASEL, the organism is the string that the iterative process of mutation and selection is working on.
So what is happening is that selection is gradually adapting the organism to the environment over generations.
In real life, environments change slowly. Islands move, sea levels change, climates change and so on. So if you want to make the WEASEL more realistic, make the target phrase change slowly too. You can even make it change randomly, so the individual letters are mutated at randoml.
If you did that, it would still work, and it would work much better than randomly guessing without selection.
This is simply false.
Computers have minds?
No, I don’t mean anything about the honesty of the computer. You asked if we had to know the target beforehand, and we don’t, the program can run without the programmers ever knowing what phrase the computer picks as an “environment”, and it would still manage to adapt the “organism” to the “environment” by doing cumulative selection.
No, it is simply a fact.
Because computers can’t do that. It is not an ability they have.
Apparently we can’t even agree on that. Wow. It’s the fitnesses that accumulate? That doesn’t even make sense.
Mung: In the Weasel program what is it that accumulates? It’s the number of letters matching the target phrase. Can we at least agree on that?
Sure, we can agree on that.
Double post.
And let the Weaseling begin!
phoodoo,
One suspects that you are somewhat unfamiliar with actual computers …
… and have misunderstood the idea of picking absolutely any target string from the space of possibilities.
Bizarrely, Mung (computer literate) and phoodoo (?) are both desperately casting about for Ways in Which a Weasel Wouldn’t Work. That there are such ways is not in dispute, and much effort could be spared in agreeing that there are, indeed, such circumstances. No-one is attempting a wager on those.
As Mung just told you, this is completely false. If the target phrase were analogous to the environment, that would mean that in evolution, organisms are naturally selected that most closely resemble the environment.
So the fish that are wettest get selected, and lions that are sort of sandy and dry with some coarse brush get selected.
Why this would help them survive better is anyone’s guess.
Mung,
To someone who struggles with even the basic concept of fitness, perhaps. In the implementation, we measure the Hamming Distance of a string from the target. Strings closer to the target have higher fitness than those further away. Since the distances are steadily decreasing as the Weasel runs, the fitnesses must be going up. The starting population has lower fitnesses than any member of a later population. As one could easily see by ‘freezing’ such a prior population, a la Lenski, and comparing survivorship with more recent individuals in a separate contest.
Allan Miller,
Well, if a computer can’t change its mind, or not do what it is told to do, how can you make the claim that it is the computer and not the designer who is choosing the number?
A computer just does what it is told right?
Poe’s law needs to be renamed after phoodoo. smfh
I’ll try this again.
How about a WEASEL that works like this:
The “target” phrase (which is a string of symbols), which here is supposed to be analogous to a changing environment, is initially generated at random and then it will subsequently slowly change, at a few random positions, every few generations.
This would then reflect the “random” changes of the environment that selection has to track, instead of some pre-specified static target. Like the environments in nature where continents move, rivers slowly carve new paths, changes in solar output slowly change the climactic conditions and so on and so forth.
And then there’s the organism, which in the program is also just a string of symbols, is randomly generated to begin with. And it goes through the same iterative process of mutation and selection, by comparing it to the “environment” string, as in the original weasel.
And we can just stop the program at some arbitrarily picked number of generations and then compare the results to how randomly guessing the string, without selection, works.
Would this be an acceptably realistic WEASEL test for our resident antiweaselists?
Mung,
Fitnesses don’t change cumulatively as a result of random genetic drift. So no.
Alan, please try to think.
If you freeze a population, and then measure its fitness, it is exactly the same as the next population you freeze and measure. Because the only measurement you have for fitness is what exists.
There is no population of bacteria in Lenskis experiments that are more or less fit than any other population.
Its pretty ironic that you are accusing others of not understanding this whacky concept you call fitness.
Rumraket,
Can we stop evolution at any point in time, and measure which organisms are most closely matched to the environment? Is a dog or a fish more closely matched to the environment?
phoodoo,
If there are N possible variants, and the computer has a chance 1/N of constructing one of those variants, I don’t see what ‘choice’ the designer is making, in any conventional sense of the word.
Ask Mung about randomisation techniques. No use asking me; I’m in league with Satan and will tell you anything to gain a soul.
phoodoo,
So Lenski’s observation that fitness (as measured by competitive superiority among more recent frozen samples) has been on the increase is … what? Not an actual observation?
The Rumraket theory of evolution: Organisms evolve by accumulating little bits of their environment.
Unfortunately they don’t care about realism. All they know is that this is about evolution, it started with Dawkins and they must resist the idea at all cost
How, indeed, would a string “METHINKS IT IS LIKE A WEASEL” fare in a population of “JHVJ HVDJHV JHD VHVJ *&*%%^”? Does it not have higher fitness? Why not?
I get what you’re saying, and that is where the analogy breaks down, as it is obviously not the case in real life that organisms adapt by becoming the environment they are adapting to.
My point is merely that, in order for the program to reflect the actual evolutionary process, something has to constitute the environment, as in that thing to which the evolving entity, is adapting.
So the degree of match between the environment string, and the organism-string, would be analogous to the degree of adaptation of the organism to it’s environment in reality.
The string of symbols are merely analogies to properties of the environment, and properties of the organism. And when they match, this is meant to reflect that the properties of the organism are well matched (as in well adapted) to the properties of the environment.
What is the problem?
We can measure relative reproductive success in different environments, for different species, yes.