Subjects: Evolutionary computation. Information technology–Mathematics.1
Yes, Tom English was right to warn us not to buy the book until the authors establish that their mathematical analysis of search applies to models of evolution.
But some of us have bought (or borrowed) the book nevertheless. As Denyse O’Leary said: It is surprisingly easy to read. I suppose she is right, as long as you do not try to follow their conclusions, but accept it as Gospel truth.
In the thread Who thinks Introduction to Evolutionary Informatics should be on your summer reading list? at Uncommon Descent, there is a list of endorsements – and I have to wonder if everyone who endorsed the book actually read it. “Rigorous and humorous”? Really?
Dembski, Marks, and Ewert will never explain how their work applies to models of evolution. But why not create at list of things which are problematic (or at least strange) with the book itself? Here is a start (partly copied from UD):
- It is not a textbook, it is a tract: The authors expect their readers to know important verses of the Bible by heart (“Secondly we believe a la Romans 1:20 and like verses that the implications of this work in the apologetics of perception of meaning are profound”), but that they have not heard of the most common technical terms (“JPG: pronounced JAY-peg”). The maths is used not to enlighten, but to impress: it is not just preaching to the choir, it’s preaching to the choir in Latin.
- The nature of this book allows the authors to skip over all the problems of their ideas and omit difficult definitions: while they talk about “searches” for dozens and dozens of pages, they never define what a “search” is.
One of the most problematic sentences is on page 173: “We note, however, the choice of an [search] algorithm along with its parameters and initialization imposes a probability distribution over the search space”.
Does it really? They authors have tried to show this in a couple of ways in various papers, and each of their approaches seemed to be ridden with further problems. So, they just side-step this crucial bit of their theory. - The conclusion for the section on proportion betting seems to be wrong (Section 4.1.2.2.12 “†Loaded Die and Proportional Betting”.) The authors claim:
The performance of proportional betting is akin to that of a search algorithm. For proportional betting, you want to extract the maximum amount of money from the game in a single bet. In search, you wish to extract the maximum amount of information in a single query. The mathematics is identical.
But if there are two fields of equal size, and I lost my keys in the first one with probability 2/3, in the second one with probability 1/3, it makes sense to search the whole of the first field, and only afterwards the second one. On average, it takes longer to switch between the fields with probabilities 2/3 and 1/3, respectively (even if switching does not take any time) – that’s because the doubling rate parameter does not apply to this problem.
DiEb:
An apt analogy.
There was a recent “ID the future” podcast with Marks HERE. I listened for around 10 minutes. It was absurd, but that’s cheaper than buying the book.
Then it is what I would call “a snow job”.
So?
So?
Mung,
So, have you read it?
Are you asking that because you don’t think it matters?
First thought: “Yikes! I accidentally hit the ‘post’ button.”
LOL – but I’m waiting for all the “fancy stuff” you have got on this book! Perhaps you should use † and ‡ to warn DEM that your posts include mathematics? And don’t forget to address yourself as “humble”, following the excellent example of DEM….
I have read some of it, but not all of it. Do you have an answer to either of my questions?
First, I don’t think it was designed to be a textbook. And your OP reads quite like a tract itself, along with your own style of preaching to the choir here at TSZ.
What difference does it make if it’s not a textbook, did you buy it thinking it is a textbook and now you want your money back?
I’d just add that in addition to the technical issues with what is a “search”, there is a simpler problem with MDE’s refutation of the effectiveness of natural selection (and that is what they are basically aiming at). They argue that on average “evolutionary search” cannot do better than drawing a single randomly-chosen genotype at random.
The problem is, that the only way they can argue this is to include among all possible “evolutionary searches” a huge fraction of crazy and dysfunctional “searches”. For example, they have to include ones that tend on average to go downhill on the fitness surface, as well as ones that ignore the fitness surface.
As soon as we restrict ourselves to processes that have genotypes that have fitnesses, and that have the fitnesses actually affect whether or not the organism survives and reproduces, then the set of “evolutionary searches” is much better behaved, much better at finding improved fitnesses. And their theorems go away, and specified information can be incorporated into the genotype without any necessity that it be provided from outside by a Design Intervention.
Tom and I showed this in a post on 29 March 2015 at Panda’s Thumb
1. Luckily, I only invested 1.50€ – thanks to the Fernleisystem and the Tierärztliche Universität Hannover – but I have no idea why they bought the book.
2. Tract/Textbook – I was expecting something of a little bit more substance: it would be nice to have a treatise which includes all of the current knowledge on DEM’s theories.
3. Sorry, English isn’t my first language. “So?” is a question which is very hard to translate!
Now you really don’t have to go about trying to convince the rubes in the pews here at TSZ that this is false, they already know this is false. Or even if it’s not false, they know it doesn’t apply to programs such as WEASEL, ev, or Avida (or any genetic algorithm, for that matter). And it follows rather naturally that DEM are obviously mistaken.
So who are you trying to convince?
Yes, indeed, you did – and I enjoyed the read! But imagine a flat-earther writing a book on geography: you would be right to laugh him out of court because his “theories” are not applicable to reality.
My aim is lower: I just want to show how his geometry does not work within his own framework.
Mathematicians. As should DEM.
but then you’re just sneaking in information that can only come from an unknown intelligence!. Dont cha know it’s either gawd or total law-less chaos? No free lunch baby!
The following book is specifically advertised as a textbook.
Natural Computing Algorithms.
Can someone direct me to where they define what they mean by search? Perhaps they avoid using the term, but somehow I don’t think so.
Mathematicians should read their peer-reviewed work. This book is specifically written for a lay audience.
Sheesh. Seriously?
It’s neither. It’s a scam.
Far more technical books use the term without explicitly defining what they mean by it. So I consider it a time-wasting (ho-freaking-hum) nitpick that does nothing to actually advance the conversation. It’s preaching to the choir.
Does DiEb have a definition of search which conflicts with the way DEM use the term?
Have your read that book?
What technical issues are you talking about? How do you define search?
I’d define a search and a search algorithm along the lines of Wolpert and Macready in their paper on “No Free Lunch Theorems For Optimization”: in most of the search problems you have a space and function on this page with a range of (at least partially) ordered values. The target is the element where the function reaches its optimum, so optimization and search are two sides of the same medal.
For an unassisted search, the function is just the characteristic function of the target, other problems provide more complex functions: for the WEASEL, you have the Hamming-distance. You can think of the Traveling-Salesman-Problem (TSP) as a search for the shortest way through all cities.
In the first two examples, you know the optimum of the function beforehand, so you can identify your target during the search.
For the TSP, the optimum is not known upfront. But at least theoretically, you can enumerate all possible ways and identify the optimum, ergo find the target.
I have not read the entire book. I have read some of it. Enough to issue my challenge. Enough to know they refer to searches and search engines and that (as yet) I’ve not found where they define what they mean by the term.
So you see why, to me, it appears like you’re nitpicking?
Well, golly. Do you suppose that might be because most technical books do not provide mathematical analysis of search?
You’ve just changed the question. Where have Marks et al. provided the formal definition of search required for the math they claim to have done?
You like the idea that Marks et al. have established a body of mathematical analysis, and know that you had better not get into the details.
The technical issues that Tom and Dieb are raising.
It doesn’t matter how I define “search”. MDE define a set of things that they call “evolutionary searches”. They argue that they have proven (in the DEM papers earlier) that these on average perform no better than chosing a random genotype. Tom and I have shown that this is only because they included so many utterly crazy processes in their set of “evolutionary searches”.
Do you disagree with us about that?
Handbook of Natural Computing
Again, when they write about search in these books, do they define what they mean by the term? It’s a 4 vol set. I’d really appreciate it if someone could direct me to where they define what is meant by search.
I guess they just assume people know. Or maybe they had technical difficulties.
Now you know why I bought the tract version.
Of course I disagree. You say it doesn’t matter how you define search. Yet you complain about how DEM use the term. It’s like you don’t like their use of the term but don’t have anything else to offer in it’s place. The irony.
I posted a definition of the term.
Is they way the term is used in their book incompatible with that definition?
This stuff shouldn’t be in court. it should be in science circles or the general public up on the issues.
Evolutionism clinging to layers/judges to make their case is a sign of a problem in making a convincing science case.
there is a greater problem yet.
on both side in doing these math things IT shows anything can be done with math relative to origin question.
Yet in reality math offers nothing of intellectual substance or accomplishment to biology subjects.
Its al squeezing numbers that only work AFTER presumptions on biology origins have been made.
However in a contention these presumptions are not agreed to.
So math contributes nothing to help one or the other side in the contention.
A bigger problem withy the whole concept of stats on biology origins like evolutionism.
Really!? Here’s how the book ends:
Really? They’ve made no such claim in their technical papers. Now they glop a bunch of stuff together in a general-readership book, suppressing technical details because they have changed definitions repeatedly over the years, and declare a grand defeat of “undirected Darwinism.”
This is not just technically wrong, but also ethically wrong.
Of no use at all in mathematical analysis.
The question is how the people who claim to have done the math have defined the term. And the answer is that they have not stuck to one definition in their published papers, and are concealing that fact now with nebulosity in their book.
I see the term active information used in three different senses in Chapter 5, only two of which I recognize from the technical papers.
In the section 4.1.2.2.1 that DiEb points to, there’s yet another meaning attached to active information. (IIRC, what MDE call active information in that section was active entropy in their treatment of the utterly botched Horizontal No Free Lunch Theorem.)
The upshot is that Marks, Dembski, and Ewert are not even functioning at high enough a level as mathematicians to avoid needless contradictions of themselves. And yet they present themselves to the “choir” that DiEb refers to as high priests.
It’s not just technical error. It truly is a scam. Somebody (Ide Trotter?) has put up $250 thousand for the Center for Evolutionary Informatics.
There are a few reasons why I simply find it hard to care.
1. The book which is the object of the OP is not a book of mathematical analysis.
2. There appears to be no mathematical analysis supporting genetic algorithms as models of evolution. Yet the textbooks seem to have no issue referring to search, search algorithms, and search engines and popular audiences seem to have no problem thinking they are evidence for evolution.
3. I smell a double standard.
4. The OP is about their book and thus it is about the programs discussed in their book, such as ev and Avida. Are we really not talking about “search” when it comes to these computational icons of evolution?
5. When is someone going to stand up and admit that ev and Avida are not models of evolution. And if someone claims that they are not models because they are actually evolution in action, then the analysis applies to evolutionary search. To paraphrase Joe, it doesn’t matter really what you call it.
6. I have yet to see someone present an alternative.
Actually Tom, I am utterly ambivalent about their mathematical papers because I am not qualified to asses them. I neither like nor dislike “the idea that Marks et al. have established a body of mathematical analysis” and I don’t rely on it.
No, you didn’t wait. And I’m glad for that (hell very well may freeze over first). Thanks for reusing my header.
The question of whether the math applies is anterior to the details of the math (The “weak” case of conservation of information is just Markov’s inequality with
applied to both sides. [George Montanez obfuscates a similar application of Markov’s inequality — “search” constant, target varying — in his dissertation.] In the “strong” case, the probability mass assigned to the target by a Dirichlet-distributed probability mass function is Beta-distributed. Like, wow, I just generalized their result. Designer works in mysterious ways.)
It doesn’t bother Joe F. in the least that “natural selection” can be both cause and effect. I should be devastated if “active information” has more than one sense?
Fine, Mung. But the book is devoid of biology. The point, in the end, is that mathematics suffices to establish that “Darwin was right or wrong,” and that Marks et al. have in fact established the latter mathematically.
I’m not making any sort of pronouncement on the general issue of whether design in the cosmos is empirically detectable. But I am saying that the “evolutionary informatics” project that Marks and Dembski undertook about a decade ago has turned out to be an utter flop. Mind you, when Marks joined up with Dembski, I supposed that we’d be seeing something of substance. (And I was affiliated with the Evolutionary Informatics Lab for a time, in protest of Baylor’s refusal to host its website with the standard AAUP disclaimer attached.) I am amazed, time and again, to see such smart guys (Dembski and Marks, not Ewert) make such dumb errors. I think it’s purely a matter of starting with conclusions that you “know” have to be true, and then scrambling to produce the argument. That’s backwards, of course. (And tu quoque does not change the fact.)
Speaking of tu quoque.
As I indicated in my review of the book, all other questions pale in comparison to this one. The answer is no. In all of the “conservation of information” theorems MDE have published over the years, the target (solution set of the problem) is determined independently of the search — however the search is defined. The modelers are making no such claim. Their models are of a type specifying an evolutionary process along with an event that tends to occur in the process. You, yourself, noted that the match of the instruction set and the event of interest in the Avida EQU experiments. The evolutionary process and the event of interest are jointly specified. Each depends on the other. If you want to argue that there’s something wrong with that, then do so. But don’t tell me that the “conservation of information” math applies. It does not.
Balderdash. When the “target” and the evolutionary process are jointly specified, each depending on the other, the “conservation of information” math does not apply.
I wouldn’t word it that way now. But the choice of the word can is correct. And I’ve wondered if you’re reading it as will always. The point of the Avida model of evolution of complexity (EQU) is that it depends on the circumstances. There is no claim that the circumstances always exist — just that they sometimes exist.
Genetic algorithms are not models of evolution. One of the things about the book that offend me is the assiduous conflation of evolutionary search and evolutionary model.
The book is a report on mathematical analysis, and on applications of that analysis.
Mung quoted a definition of search that said, in its main part, this:
And there we can see the problem with MDE’s book and the DEM papers that it reports on. Because the definition of “evolutionary search” MDE use is not confined to processes that look for improved performance. They include, equally, processes that look for worse performance. Plus processes that choose outcomes without regard to performance.
And given that they include all these with equal likelihood, is it surprising that they end up with a theorem that says that the average outcome of all these processes is that improved performance is no more likely than if we just drew a genotype at random?
If we take Mung’s preferred definition and take all processes that “lead to a desired performance”, or “improve the relative performance of individuals”, we would get a different result. Namely, that on average the outcome of these models of evolution would be that MDE’s “active information” would in fact come to be in the genome.
MDE have found that active information does not come to be in the genome, but they do it by including tons of processes that in no way can be thought of as modelling evolution.
So thanks to Mung for the definition. As vague as it is, it rules out a shitload of the processes that MDE (and DEM I, II, and III) include. And that inclusion is why MDE get no-better-than-random performance of “evolutionary search”. Which they proudly announce as refuting “Darwinism”.
Mung:
Were you less of a Mung, you might grasp that for a word or phrase to have more than one meaning is not a problem. Equivocation is.
Mung,
Nonetheless you know their mathematical conclusions are correct regarding avida, right? Right?
Joe Felsenstein,
It bears mention that Ewert twice “responded” unresponsively to our post. In his first shot, he ignored the GUC Bug model entirely, and said that we agreed with him more than we realized. Then he “responded” by
1. changing the definition of search,
2. changing the definition of active information,
3. changing our model, and
4. synthesizing a claim we did not make.
Swell guy.
So MDE have gotten a response from a world-class theoretical biologist (not entirely spoiled by my contribution). But they want nothing to do with it in their book. They’d rather tell their readers that Dave Thomas’s GA for Steiner trees is a model of evolution. And perish the thought that they’ve failed to understand the difference between algorithmic models of evolution and evolutionary algorithms used in search for solutions to given problem. They’re entirely too brilliant to have mistaken something they don’t understand for something they do understand.
The comments to this post at Panda’s Thumb seem to be lost. Do you know whether there is any way to retrieve them?