Anybody know about this?

[Collin]

Okay, so I'm browsing all the evolution forums I frequent, and on a board that hasn't been posted in for three weeks, this guy comes out of nowhere and copy pastes (with permission) the text for a fictional dialogue called "The Minkowski Challenge," by Fady Bahig.

Anyone heard of it? From what I've read so far, it's a rather childish reliance on self-incredulity and the author's knowledge of computers, which Bahig uses to draw attention away from the lack of biological expertise. If anybody has something to say that I've been missing, by all means, help me out.

Amazon's Page: http://www.lulu.com/content/438633

[RBH]

I'm having some trouble connecting a self-published book about some sort of "spiritual journey" (the link is not to Amazon's page; it's a species of vanity publisher) with E/C here.

RBH

[Quetzal]

Someone appears to be spamming internet evolution vs. creationism sites with this garbage. EvC forum recently got a new poster who presented this. Administrator discussion is on-going as to whether the forum will allow it or consider it spam.

RBH: Here's the essence of the "challenge" - the spammer (my opinion) posted a fairly large extract: Minkowski's Challenge (http://www.evcforum.net/cgi-bin/dm.cgi?action=msg&f=25&t=2321&m=1).

[RBH]

Someone appears to be spamming internet evolution vs. creationism sites with this garbage. EvC forum recently got a new poster who presented this. Administrator discussion is on-going as to whether the forum will allow it or consider it spam.

RBH: Here's the essence of the "challenge" - the spammer (my opinion) posted a fairly large extract: Minkowski's Challenge (http://www.evcforum.net/cgi-bin/dm.cgi?action=msg&f=25&t=2321&m=1).Interesting. Thanks for the heads-up. This will bear watching.

RBH

[Collin]

Man... all the weird names going around... Fady Bahig? Albert Godwin? These guys sound like they were raised on cabbage and attended school at their breakfast table.

[SophistiCat]

Looks like an amateur who has no clue about the actual work that has been done in genetic algorithms. Here's Godwin's own the summary of the argument:

1- There is a program that reproduces and mutates, it mutates in a manner very similar to living organisms

2- There is a program that deletes all the offspring carrying a certain sequence of code

3- The program will evolve another one that will reproduce well yet at the same time not contain the targeted sequence of bytes.

4- This is a minor evolution and it can occur (Microevolution) , but you can never force such a program to take a big jump (Macroevolution) by developing encryption, for instance.

Needless to say, this in no way disproves "macro-evolution". The ToE does not say that organisms will develop highly derived features just for the hell of it. An easy, one-time challenege like that presented in the simulation will likely prompt a simple, one-time adaptation. For more complex new features to evolve, selective pressure would have to be more challenging and mutable.

[Collin]

I was surprised, SophistCat. When I used that argument on him earlier this morning, he became shockingly humbled, assuring me that he was no scientist and that he didn't intend to paraphrase what the author was trying to say.

[LambdaCalculator]

1- There is a program that reproduces and mutates, it mutates in a manner very similar to living organisms

It depends on what he means by "very similar". GAs typically produce the offspring themselves rather arbitrarily, while genetic heredity controlls the progeny of biological organisms. That's not what I'd call "very similar", just a convenient approximation for engineering purposes.

2- There is a program that deletes all the offspring carrying a certain sequence of code

No, there isn't. There are multiple means of specifying selection criteria when using GAs, if any involve deleting individual products that contain a certain line of code, I'm not aware of it. That would be a horribly ineffective way to do it.

Actual GAs typically use Bayesian differentiation or Hidden-Markov models to specify selection criteria, and earlier there was always good old artificial selection by an actual user.

3- The program will evolve another one that will reproduce well yet at the same time not contain the targeted sequence of bytes.

Well, you can program a GA to do that, but it's not necessarily true (GAs would be worthless otherwise). See above.

4- This is a minor evolution and it can occur (Microevolution) , but you can never force such a program to take a big jump (Macroevolution) by developing encryption, for instance.

I'm wondering whether this guy even has the foggiest idea what encryption is. I don't see why encryption would ever help in producing a wide array of novel products with GAs, if anything I would think it would make it harder. But perhaps I'm misreading this.

(Edited to correct mistake)

[Quetzal]

Man... all the weird names going around... Fady Bahig? Albert Godwin? These guys sound like they were raised on cabbage and attended school at their breakfast table.

Since the "book" this is supposedly referenced to "proves" "microevolution" and at the same time "proves" there is a barrier preventing "macroevolution" (whatever the heck that is), I've been entertaining the idea that the two names are pseudonyms. These are such blatantly creationist claims, the "God win(s)" is pretty straight forward. Anyone want to try and take a stab at unravelling an anagram with the other (if it is an anagram)?

[Barbarian]

The example posted in that thread is rigged, multiple ways:

(1) it does not allow for large code blocks to first grow to completion and then get connected into the call flow. The maximal size Gen.com can copy is its own size, or at least so I read that ugly assembly. So of course code serving truly novel functions will be truncated before it could get off the ground. Real genome is not so strictly size-limited.

(2) the selector emulates a deterministic Natural Selection: it will surely kill programs with a certain byte sequence. This means that extra fast evolution is required to get to the encryption variant in one jump, or the selection step resets the evolutionary clock. To better emulate NS, the selector program should kill recognized virii with a high (but lesser than 1.0) probability, and perhaps kill non-recognized programs with a low prob. ('s what I mean when I bring up randomness of NS time to time here ...)

If those two rigs are removed to better resemble, evolution of an encryption(*) skill or indeed the apparition of the whole of "Windows Vista for 16 bit" is possible. In fact, what is amazing to me is that it is enough to fix those two aspects and evolution starts working, even if a host of fundamental differences remain between biological evolution and this one.

ETA: in fact it is enough to remove the first rig, although the result would be at par with tornado in a junkyard. That is possible, too. What the guy should really aim for is to build a large replicator, with lots of modules prone to duplication and function hijacking.
________________________
(*) I guess he uses the word "encryption" to mean some scrambling capability to hide particular byte strings from primitive virus finders. The program could evolve into a sequence of blocks of (instruction+nops to pad to a fixed size+absolute jump to the next block), and permutate those blocks randomly in the output, updating the jumps of course. This would indeed not be reliably detectable by a simple byte-block-matching selector.

[Quetzal]

Hi Barbarian,

I've been following the thread over at EvCForum. Excellent response you made.:notworthy: It'll be interesting to see if "Albert" responds.

[SophistiCat]

I'm wondering whether this guy even has the foggiest idea what encryption is. I don't see why encryption would ever help in producing a wide array of novel products with GAs, if anything I would think it would make it harder. But perhaps I'm misreading this.

AFAIK, the book author used the example of the evolution of viri in response to virus-detection software. The simplest virus scanners search for signature byte sequences of known viri. This is what the "selection" part of the simulation did in this case. The simplest response to that is, of course, to mutate the virus code just enough to destroy the original signature. In real life this works only until new virus definitions come out, but in this case I think they never updated virus definitions between iterations - thus the primitive genetic algorithm had no incentive to mutate any further.

Modern virus scanners also employ a number of methods that do not rely on precisely identifying known virus signatures. Heuristic methods try to detect suspicious code in executable files, or suspicious activities during program execution. Code emulators launch a program in a safe virtual environment and examine its instructions for suspicious signs.

Some of these innovations in virus detection can be circumvented or made more difficult by encrypting the main part of the virus. However, since the simulation scanner never progressed past the primitive signature search stage, there was no reason to expect any GA to develop such sophisticated defences.

[LambdaCalculator]

I think I understand this a bit better now.

I think the main problem is that, as Barbarian has pointed out, he has rigged his example to yeild results that he wants and fails to emulate natural selection.

Emulating natural selection is, I would assume, intractable. But an approximated simulation would be far more complex than what he proposes. Off the top of my head I can think of partitioned fitness landscapes and some sort of parametric method for selection as things that may yield results that more accurately resemble the physical world.

Creating simulations like this is no different than concocting ID "theory" in and of itself. The IDers can get whatever results from their "theory" if they set the parameters in their favor (there are no transitional fossils, no new information, evolution can't create "irreducibly" complex systems, etc.). Computer programs like these are no different.

_________________________
EDIT:
(It's also interesting how he has written the entire GA in ASM code. ASM is hardly ever used anymore in any serious software engineering projects (save those with serious performance requirements) precisely because it's hard to read, long and error prone. Wonder why he didn't simply conventional C or a higher-level language like Perl, Python or Ruby to write it. Such would be a much easier and casual read.)

[Barbarian]

Hi Barbarian,

I've been following the thread over at EvCForum. Excellent response you made.:notworthy: It'll be interesting to see if "Albert" responds.
Thanks, Morpho. (where's that blushing smilie when I need it?). He did reply as I expected - he is no expert in assembly, so he cannot evaluate our counterarguments, but he thinks we failed to provide valid ones.

[SophistiCat]

EDIT:
(It's also interesting how he has written the entire GA in ASM code. ASM is hardly ever used anymore in any serious software engineering projects (save those with serious performance requirements) precisely because it's hard to read, long and error prone. Wonder why he didn't simply conventional C or a higher-level language like Perl, Python or Ruby to write it. Such would be a much easier and casual read.)

Because ASM has a one-to-one correspondence with machine code, and mutations operate on machine code.

[RBH]

I haven't been following this discussion closely, but it may be relevant that what evolves in Avida (http://devolab.cse.msu.edu/) are opcode programs.

RBH

[doc_simon]

Hmmm, he appears to be missing a potential function of development. With such a thing, there can be a less obvious mapping from genotype to the phenotype. Hence, even the second program can be rendered ineffective as the genotype would not correlate in predicatable/linear ways with the phenotype.

In effect, the development process (which I believe has been observed in nature;) ) would act as a the necessary decryption algorithm.

I can't be bothered to read the full details, but based on SophistiCat's response, using an indirect geno-phenotype mapping (e.g. development) you can beat this challenge. Do I win a medal?

[doc_simon]

<dupe>

[SophistiCat]

I haven't been following this discussion closely, but it may be relevant that what evolves in Avida (http://devolab.cse.msu.edu/) are opcode programs.

RBH

Which are Assembly light, with a more limited instruction set.

[RBH]

Which are Assembly light, with a more limited instruction set.Still Turing complete, in concert with the virtual machine the opcode runs on.

RBH