As Tom English and others have discussed previously, there was a book published last year called Introduction to Evolutionary Informatics, the authors of which are Marks, Dembski, and Ewert.
The main point of the book is stated as:
Indeed, all current models of evolution require information from an external designer to work.
(The “external designer” they are talking about is the modeler who created the model.)
Another way they state their position:
We show repeatedly that the proposed models all require inclusion of significant knowledge about the problem being solved.
Somehow, they think it needs to be shown that modelers put information and knowledge into their models. This displays a fundamental misunderstanding of models and modeling.
It is a simple fact that a model of any kind, in its entirety, comes from a modeler. Any information in the model, however one defines information, is put in the model by the modeler. All structures and behaviors of any model are results of modeling decisions made by the modeler. Models are the modelers’ conceptions of reality. It is expected that modelers will add the best information they think they have in order to make their models realistic. Why wouldn’t they? For people who actually build and use models, like engineers and scientists, the main issue is realism.
To see a good presentation on the fundamentals of modeling, I recommend the videos and handbooks available free online from the Society for Industrial and Applied Mathematics (SIAM.) “[Link]”.
For a good discussion on what it really means for a model to “work,” I recommend a paper called “Concepts of Model Verification and Validation”, which was put out by the Los Alamos Laboratories.
You charge they don’t know squat about modeling because they say the modelers are including, in this case, their own world view o problems that need to be fized THEN you don’t show why this is wrong. yopu just say YES ITS OKAY.
I have been reading recently einsteins book on his theories recently and he shows how the direction came from the concept of fRAME OF REFERENCE in physics.
The iID guys here are saying evolutionism has gone wrong because of a frame of reference from thinking beings as opposed to random chance.
Evolutionists have evolution solving problems which in fact are only problems for thinking beings and so unlikely mutations/selection could solve problems.
they got a good point that error is lurking in basic evolutionary thought.
In their Introduction to Evolutionary Informatics, Marks, Dembski, and Ewert go to much effort to argue that evolutionary algorithms succeed because of the information built into them about the solution to the problem. For some, like Richard Dawkins’s “Weasel” algorithm, the target is coded right into the program.
But for numbers of others, they succeed only in showing that some parameters of the algorithm have been “tuned” so that they succeed better. This is an extremely weak argument. For example, for Dave Thomas’s Steiner Tree algorithm, the program is presented with a set of points in a plane and tries to find the shortest tree connecting them. The problem is defined by the set of points. To argue that it had the solution built into it, front-loaded, we would have to have something coded into the program that was specific to the particular set of points that poses the problem.
And yet, however much the program’s parameters have been tuned, the tuning is long over before the problem is even posed. And the tuning does not change from one instance of the problem to another, which means from one set of points that need to be connected, to another set.
So whatever has been done, the algorithm cannot have the solution front-loaded into it. DME do not make clear the distinction between tuning the general algorithm, and providing information specific to the problem that is posed, aside from the set of points itself. If they had made the distinction clear, it would have been obvious that for the Steiner-Tree solver, they have not proven that the requried information is front-loaded.
Joe Felsenstein,
The authors first present themselves as scrutinizers of mathematical models of real-world biological processes. But instead of talking about the validity of math models they change the subject to the performance of math algorithms.
An algorithm, by itself, is just math. The variables and the relationships between them need not have any correspondence with any real world situation.
It’s fine to critique their discussion of math algorithms. But I think that the greater concern is that they are dodging the main issue of math modeling, which is realism.
I have to say, Robert Marks (especially) appears impervious to constructive criticism. I’ve emailed him more than once regarding Tom English’s critiques here at TSZ. At least Dembski used to admit to revising his work on seeing valid criticism (though without acknowledgement).
In cases like the Steiner Tree genetic algorithm, its developers were not trying to model biological phenomena closely, but instead were trying to show that evolution-like processes could do well at solving problems — that they often end up with solutions far better than random ones.
Marks, Dembski and Ewert’s response was to try to show that, although they did well, this was only because the relevant information is front-loaded into the algorithm. They present themselves as having shown that, but the did not show that at all. Whenever they make arguments about fine-tuning of parameters, they are talking about parameters whose values are the same for all problems that the algorithm tries to solve.
So I am not sure that their work should be criticized for not addressing the biological realism of the algorithm. Instead it should be criticized for not showing what it claims to have shown. The Steiner Tree algorithm does not have front-loaded information specific to any individual case presented to it. And they claim that it does, and that they have shown that.
Thank you for your OP. There is a lot of confusion about this here at TSZ, or perhaps it’s just pretending to not see what is so obvious.
The OP seems to be saying they are right about that and that there’s nothing remarkable about it. Yet much ink has been spilled here at TSZ denying it.
Why do you suppose that is Joe?
@Mung: The OP does not distinguish between (1) information that helps the evolutionary algorithm solve problems in general, and (2) information that is “inclusion of significant knowledge about the problem being solved”. In other words, information that specifically helps the algorithm solve the particular case of the problem.
In cases like the Steiner Tree algorithm, the latter type of information is not put into the program. Marks, Dembski, and Ewert do not make the distinction between the two types of information, and neither did the OP.
MDE are using the presence of information of type (1) to imply that there is information of type (2). I’m denying type (2) information is there, and will continue to do so until you show me that it is there.
Joe and Mung,
I thank you for your extensive comments. But when Mung likes my post and Joe is dubious I know that I am not communicating successfully.
The OP is about mathematical models of real-world processes such as biological evolution. It is not about algorithms that solve math problems. I hope to expand on this discussion in the next day or so.
This is not so.
you are saying evolution is a real world process. YET thats the contention here.
Math can show real world processes .
The opponents are saying math doesw not work for evolutionary processes.
its like your presuming evolution is true, for your readers too, and then you dismiss the ID folk.
Say it ain’t so!?
i think math being applied to biology only works after biology processes has been proven. it isn’t a aid to proving bioloogy processes.
this is a error of analysis and i see it here where these modelings are being done.
Robert Byers,
You did catch some ambiguous wording in my comment. When I used the term “evolution” I meant only change over time. I said:
I should have said:
An evolutionary model would include such things as random variation and selective reproduction in the population. And it would be legitimate for people to challenge whether such a model was realistic in how much significance the modeler gives to such factors.
The point of my OP is that you can’t challenge a model by simply pointing out that some aspect of the model was put in there by the modeler. This is what DME try to do in their book. If they want to challenge the validity of a model they should be talking about realism. The best I can tell, a source of their confusion is a conflation of (1) mathematical models of real-world processes and (2) evolutionary algorithms for solving math problems.
In the OP, Freelurker says:
That quote from the OP correctly characterizes Marks, Dembski, and Ewert’s argument. And I have been arguing that what MDE have done actually does not show this, for a number of the evolutionary algorithms they use as examples.
The issue MDE raise is not whether those algorithms model evolution well enough. They are discussing all models that incorporate evolutionary processes, including those that are used to solve engineering problems.
I think that the difference between Freelurker and I in this discussion is that Freelurker wants to concentrate on those examples in DME’s book that actually are trying to model aspects of biological evolution. For those, Freelurker might argue that they are inadequate models of evolution, and that more adequate ones would not contain built-in information about a “problem being solved”.
The models that I am discussing are the ones in DME’s book that try to find good solutions to engineering problems such as finding Steiner Trees for sets of points on a plane. The effectiveness of the models at finding solutions is argued to show that real-world evolutionary processes can be effective at finding genotypes of higher fitness. DME set out to counter this by arguing that all such models “require information from an external designer” in order to work. And they have not shown this for cases like Steiner Tree.
What is it that is obvious?
All models are wrong, but some models are useful.
This is a good video:
https://www.youtube.com/watch?v=sxxkcd29J2I
If what Marks, Dembski, and Ewert had said was “Here are all these models of evoluton — but they’re not realistic enough models, so they don’t prove that natural selection works” then that would be a relevant answer.
But what they said was: “We show repeatedly that the proposed models all require inclusion of significant knowledge about the problem being solved.” And they included both models of biological evolution and “evolutionary algorithms” that solve engineering problems. For these they tried to show that they would not work to improve adaptation, or to solve engineering problems, unless their designer included in them information about the problem being solved.
They declared that they had shown this. Thing is, for many of their models, they hadn’t actually shown it.
Well they say they did. You say they didn’t.
So it requires a careful study of examples and then whether THEY show the models are inclusive of solitions already etc or YOUR SIDE shows they are independent.
they have brought a scientific criticism to this subject and so it has risen toi a higher level for investigation.
To show they are right or show they are wrong REALLY DOES require more sentences and discussion.
Not saying I can do it but until them THE judge is out.
not me, but I think a thread could be done on the issue worthy of the subject.
Its a great assertion by ID thinkers. its important and is possibly another reason for why evolutionism fails and so why its on the way out.
these guys are taking the subject on ON a careful intellectual level .
Prove them wrong or right could be important to the credibility of critics or of evolutionism.
They are using imagination and then hypothesis and so bringing scientific methodology to a contention within science.
ID thinkers once again are the innovators of modern thought on the merits of evolutionism.
You’re describing conclusions. But stating conclusions is not the whole story. There are also the arguments each side presents. I have shown that they are wrong. So far no one has refuted me on that. Care to try? (Hint: just stating more conclusions does not count).
I will have to read more carefully. i’ll try. Its not my thing. If i fail to comprehend it good enough I will just say so.
To bad one of them couldn’t do it as i understand they are aware of this blog. If its a good point then it can be made or lost here as well as anywhere.
I will try this. Oddly freelurker helps a bit.
Freelurher stresses the real world issue of showing evolutionary processes proved by these models.
JF separates tuning the general algorithm AND providing info specicif to the proble,.
THEY are saying its a failure because Specific INFO is being provided. So its not a random thing and so not from evolutionary process.
This steiner tree can be true but thats not their analogy.
They are asserting the evolutionary models are not like the Steiner tree (without saying that actually).
Their insight is just that information is tilted towards a problem that must be solved.
Your steiner tree exists before the problem is introduced. Fine!
They are saying its not that way in this evolutionary models.
They are trying to say everything is frontloaded, to use that word, .
Your steiner tree is too innocent. the evolution stuff is aiming at solutions aggressively.
P.S. A issue is also that if a mutation can do anything to make a bodyplan a wee bit different but functioning tHEN a line of reasoning alone always could have mutations evolve anything.
there is a problem in evolutionism when it can say simply mutation plus time equals new population equals creation of anything in biology. ID folks need to be wary that getting into these evo models that it can be a tar baby.
Joe Felsenstein,
First let me point out to people who have not read DME’s book that those authors do not offer any models of their own. They present themselves as scrutinizers of the work done by actual modelers. They make certain arguments and then conclude that the modelers are putting information into the models, as if this has some significance. People on this blog have posted before pointing out problems in their arguments. My point is that regardless of the adequacy of their arguments, their conclusion is just a discovery, on their part, of the nature of modeling.
Hopefully I have clarified above that I’m not talking about the adequacy of specific models. And as I tried to say in the OP, a model is nothing but information about a real-world situation (it’s a representation.) Talking specifically about modeling processes involved in the history of life, it’s not just the modeler’s selection of values for parameters that puts in information; it’s also the selection of the specific processes put in the model such as variation in populations, selective reproduction, poofing, and telekinesis. The real issue is whether the information makes the model more valid, meaning more realistic.
It was very appropriate that above you put the words “problem being solved” in quotation marks. This gets at another point I’ve tried to emphasize in my comments. (This topic probably deserves an OP of its own, but I don’t know when I will get time for it.) When we are using math to justify claims about real-world processes then we are doing mathematical modeling, and we should be using the appropriate terms and concepts. We, including DME, create lots of room for confusion and mischief when we conflate math models and math algorithms. We should talk either about the validity of math models or the performance of algorithms and not switch back and forth. In the above example, “problem being solved” makes sense in the context of people using optimization/search algorithms but we are anthropomorphizing if we use that term in the context of modeling natural processes.
And the burden of realism is different between process models and algorithms. Someone using a genetic algorithm to design a circuit board is not concerned about whether the algorithm realistically represents the way circuit boards breed. Someone using a simulated annealing algorithm to design a circuit board is not concerned about whether the algorithm realistically represents the way a circuit board solidifies from a liquid.
Freelurker,
Marks, Dembski, and Ewert are trying to argue that the information that is in the genome that enables the organism to be well-adapted does not originate in the process of natural selection.
For biological evolution, they Active Information, a measure of how smooth the fitness surface is, and argue that this is where the information comes from. To them, the fitness surfaces that allow natural selection to succeed are rare and special, and indicate the intervention of a Designer who “front loads” the process with information. (Tom English and I have argued at Panda’s Thumb that smooth fitness surfaces are inherent in physics and do not require special intervention by a Designer).
For “evolutionary algorithms” or “genetic algorithms” that solve engineering problems, they attempt an analogous argument, showing in some cases (like Dawkins’s “Weasel” algorithm) that the information about the target is provided in advance by the programmer. They try to make the analogous argument for others, such as Dave Thomas’s Steiner Tree algorithm. For those, I am arguing that they have failed, that the adjustments of parameters that Thomas makes in the design of the algorithm do not involve information about the particular problem being solved, because Thomas’s adjustments are made long before any problem is presented to the Steiner Tree algorithm.
Are you agreeing with MDE that evolutionary algorithms used to solve engineering problems contain information about the desired solution? For Steiner Tree, I think not.
To my surprise, I almost understand this.
Marks, Dembski, and Ewert, in their book, do not wave aside the Steiner Tree algorithm as using too unrealistic a model of evolution. They do assert that the information enabling it to succeed at finding a good Steiner Tree is frontloaded into the algorithm. They are trying to argue that both realistic models of evolution and “evolutionary algorithms” solving engineering problems only succeed because of frontloading. (I am arguing that they are wrong about that in the case of Dave Thomas’s Steiner Tree algorithm).
I wonder what would happen if the Geary test were used as the oracle for a weasel like program.
If it’s just a statistic for autocorrelation, wouldn’t it just end up with
“MMMMMMMMMMMMMMMMMMMMMMMMMMMM” ?
I was thinking you would select the most random like.
If I understand correctly, then you will end up with one of many possible random-looking sequences (heh, it’s almost as if you were looking for sequences that score least on the “looks designed to me” scale).
In the special case where string length = alphabet size, then you will arrive at one of the N! permutations of the alphabet. Since all ‘optimal’ scores are two steps away from other optimal scores (any position swap), you may get a frozen accident, depending on the mutation and generation size parameters. If OTOH string length =/= alphabet size, then every optimal score has optimal neighbours, and the result will meander through this optimal space…
Caution: all this is based on my assumption about how Geary would score extremely-lowly-auto-correlated sequences. I am not familiar with the test.
Joe Felsenstein,
JF
If they didn’t show how the Steiner tree was frontloaded then why could I!?
Did they think they did just by saying so?
i don’t know what their sentences said.
Anyways thats one example. they think the others are frontloaded.
if a Steiner tree really works and is a example of evolution , then it must just be a issue of how mutations can do anything. I think there is a greater flaw in all this.
Thats a aside.
Someone needs to read and quote why they think Steiner tree etc makes their case.
Joe Felsenstein,
You were saying that DEM were making unjustified and unrealistic assumptions about nature, and you were correct.
The OP and all of my subsequent comments have been about the validity of models, especially models of evolutionary processes. (Or evolutionary models of processes – however one wants to put it.) I’m going to now follow my own admonition, the one I made earlier, against changing the subject to the performance of evolutionary algorithms. People may not catch the change of subject and this may promote further conflation of math models and math algorithms.
As it happens, there is a great example of such conflation in DME’s book, specifically in reference to Dave Thomas’ application of evolutionary algorithms to the Steiner tree problem.
The authors say:
And they refer to Dave Thomas’ writings on the Steiner tree problem as an example of this.
But what Dave Thomas actually does is: (1) take a geometry problem, (2) model it as a math problem, (3) solve the math problem (approximately, using an evolutionary algorithm), and (4) map the math solution back into geometry. Contrary to what DME say, at no point does Dave Thomas model any processes, evolutionary or otherwise. (Dave Thomas himself correctly describes his own work.)
I agree but do point out that the whole reason that Dave Thomas made his evolutionary algorithm was to show that processes analogous to evolution were effective in finding a good solution to an engineering problem. He could show people who claimed that natural selection would not work that for that problem and evolutionary algorithm did fine. It was, as you note, not a model of biological evolution in any more detailed sense.
By the way, here is a recently updated discussion by Dave Thomas of the Steiner Tree debates, including a refutation of Winston Ewert’s arguments about his Steiner Tree evolutionary algorithms. It contains many links to these controversies.
Well its a little mathy and picturey.
the thing wrong in all this is not about what selection can do to solve any biological problem but is really saying a useful mutation can do anything to start a path to solving any biology problem.
ID thinkers still might/can even this on BUT its all still the same darwin idea that small changes can do/did anything in biology.
if you do have a faith in mutations then you could do anything.
however i wrote a thread on TSZ here once that by this reasoning you COULD DO ANYTHING biologically however impossible.
its all lines of reasoning that leads to proving nothing.
its saying the impossible can be done and so they say evolutionism as a creative force can never be seen as impossible.
yet its impossible being justified by faith in mutations.
THATS the great intellectual flaw in all this.
Not boundaries to what mutations could do within present evolutionary claims.
In fact i don’t easily see why mutationism /selection couldn’t be claimed to fix anything when it can fbe reasoned to do the impossible.
Anyways as usual it has nothing to do with scientific biological investigation.
its just math and infinity.