First, read through this extremely informative article and the abstracts to these three articles:
here
here
here
then continue ...
According to Dr. Marks work with evolutionary algorithms and computing and intelligent systems, evolving functional information is always guided by previous functional information toward a solution within a search space to solve a previously known problem. [endogenous information] = [active information] + [exogenous information] or as “j” at Uncommon Descent explained, “[the information content of the entire search space] equals [the specific information about target location and search-space structure incorporated into a search algorithm that guides a search to a solution] plus [the information content of the remaining space that must be searched].”
Intelligence is capable of the guiding foresight which is necessary and a sufficient level of intelligence possesses the ability to set up this type of system. This is based on observational evidence. So, if functional information only comes from intelligence and previous information, then where does the information necessary for abiogenesis (the original production of replicating information processing systems) and evolutionary change come from?
Evolution seems to be, for the most part, guided by the laws which allow biochemistry and natural selection which both result from the laws of physics. The laws of physics are at the foundation of our universe which is now seen to be an information processing system. If the universe processes information, and if biochemistry and natural selection is a result of the laws of physics, then the information for evolution by natural selection [and other necessary mechanisms] is at the foundation of our universe (as an information processing system) and represented in the finely tuned relationship between the laws of physics and life’s existence and subsequent evolution. IOW, the universe is fine tuned, that is intelligently programmed, for life and evolution.
My point is that abiogenesis and evolution are not accidental; they are necessarily programmed into our universe, arriving from the fine tuned information at the foundation of our universe, yet do not arrive strictly from laws and chance (stochastic processes) alone, since information processors and functional information are not definable in terms of theoretical law. This is similar to the ending pattern of a shot in a pool game. The pattern of balls is created by the laws of physics once the first pool ball it set in motion, however the ending pattern itself is not describable by natural law. It is a random pattern/event. But, a “trick shooter” can fine tune both the initial set up and the starting shot in order to create a desired pattern in the form of a trick shot. Just like the ending pattern of the shot in the pool game, information and information processing systems are not describable by law. Again, the ending pattern is a random pattern/event. However, information processing systems, which are necessary for evolutionary algorithms to search a given space, and the functional information to search that space, and the information to guide the search do not arise by random accident within any random set of laws. Along with the argument from CSI and my other argument (scroll down to first comment), the best explanation is that these factors of the appearance of life (an information processing system) and evolution (the production of CSI) are programmed into our universe by intelligence.
That can be falsified by creating a program which generates random laws. If these random laws will cause any information processing system which generates functional information to randomly self-organize, just as any pattern of pool balls will self organize randomly after any and every shot, then the above hypothesis is falsified. Dr. Robert Marks is already examining these types of claims, and along with Dr. Dembski is refuting the claims that evolution creates CSI for free by critically examining the evolutionary algorithms which purportedly show how to get functional information for free. Dr. Marks is using the concept of CSI and Conservation of Information and experimenting in his field of expertise – “computational intelligence” and “evolutionary computing” – to discover how previously existing information guides the evolution of further information.
Subscribe to:
Post Comments (Atom)
6 comments:
Dembski & Marks: "Conservation of information theorems indicate that any search algorithm performs on average as well as random search without replacement unless it takes advantage of problem-specific information about the search target or the search-space structure."
There is no mathematically valid "Conservation of Information Theorem". But, in any case, we're in luck. It turns out that biological evolution is working within a very non-random environment, including directional forces such as gravity and sunlight.
CJYman: "That can be falsified by creating a program which generates random laws."
All that might prove is that evolution works very poorly in a universe composed of random laws. It doesn't demonstrate that those laws are necessary, or a product of selective observation, or that there is or is not an underlying symmetry that explains the origin of those laws.
In any case, you need to grapple with the fact that evolution works very well on Earth.
Zachriel:
“There is no mathematically valid "Conservation of Information Theorem".
Ummmm ... yep ... it’s similar to NFL Theorem and has been validated through experimentation and is represented by the mathematical measurement of active information against endogenous and exogenous information. That is one of the things that Dr. Marks is working on. You know, I’m really beginning to wonder if you know anything about information theory. Do you have time to start with some basics ... er ... never mind, every time I bring up the basics, you ignore me and continue to spout your invalidated obfuscations and assertions as if information theory didn’t exist.
From: “Conservation of Information in Search: Measuring the Cost of Success”
“Schaffer’s Law of Conservation of Generalization Performance [27] states a learner ... that achieves at least mildly better-than-chance performance [without specific information about the problem in question]... is like a perpetual motion machine...
The No Free Lunch Theorem (NFLT) [33] likewise establishes the need for specific information about the search target to improve the chances of a successful search.
.[U]nless you can make prior assumptions about the ... [problems] you are working on, then no search strategy, no matter how sophisticated, can be expected to perform better than any other. [15]. Search can be improved only by incorporating problem-specific knowledge into the behavior of the [optimization or search] algorithm.
[33].
English’s Law of Conservation of Information [11] notes the futility of attempting to design a generally superior optimizer without problem-specific information about the search.”
Zachriel:
“It turns out that biological evolution is working within a very non-random environment, including directional forces such as gravity and sunlight.”
So, evolution (as the generation of algorithmically complex and specified information) is directed by the non-random aspects of our environment such as by gravitational and electromagnetic fields? Care to explain what you mean by that and provide some evidence?
Furthermore, are the laws which craft our environment actually non-random? Are the cosmological constants non-randomly chosen from a set of possible laws? If not, then although law itself does not behave randomly, the set of our laws would be completely random and thus the environment itself would be random since it would be fashioned by a completely arbitrary and random set of parameters.
Zachriel:
“All that might prove is that evolution works very poorly in a universe composed of random laws. It doesn't demonstrate that those laws are necessary, or a product of selective observation, or that there is or is not an underlying symmetry that explains the origin of those laws.
Nope. NFL Theorem and Laws of Conservation of Information predict that a random set of laws do not provide anything greater than random highly statistically probable results and will produce neither information processing systems nor algorithmically complex and specified information. So, how do we test this? So far, problem specific active information has been shown to be necessary in order to evolve a system of complex specified information. In fact, in the infamous EV evolution program, the creators stated that the program created 131 bits of information from scratch. However, upon closer analysis, after accounting for the front loaded guiding active information, the only search space left is equal to 9 bits of information: probability of success = 1 in 512. But, to make matters a bit worse, the program is so badly designed that it introduces negative information – it performs worse on average than random search of that 9 bit search space.
What that would show is that there is guiding, problem specific, active information at the foundation of our natural laws, and that no matter how many times the “tape is rewound,” evolution will necessarily converge upon specific points and ultimately what some European scientists call an Omega Point – the end goal of evolution, as front-loaded by active information.
Here
(bottom pg 9. - top pg. 12)
and
Here
(last para. before conclusion)
are thought provocative and informative hypothesis which are relevant to this part of our discussion.
Zachriel:
“In any case, you need to grapple with the fact that evolution works very well on Earth.”
I think it is you who needs to grapple with the fact that evolution works very well on earth. It is up to science to discover how that is so. Back to ID Theory and information theory (NFL Theorem and Conservation of Information) ...
CJYman: "Ummmm ... yep ... it’s similar to NFL Theorem and has been validated through experimentation and is represented by the mathematical measurement of active information against endogenous and exogenous information."
Mathematical theorems are not validated through experimentation. In any case, please provide a cite to the theorem in a reputable peer journal of mathematics. Let me add that David Wolpert, the co-creator of the No Free Lunch Theory, says that Dembski's arguments are "fatally informal and imprecise" and "written in jello".
CJYman: "Conservation of Information in Search: Measuring the Cost of Success"
Let me know if it gets published in a reputable journal of mathematics.
CJYman: "In fact, in the infamous EV evolution program..."
You cite yet another unpublished and significantly flawed paper.
Zachriel:
"Let me add that David Wolpert, the co-creator of the No Free Lunch Theory, says that Dembski's arguments are "fatally informal and imprecise" and "written in jello"."
Well then, I’m sure you’ll be able to explain to me the arguments which provide evidence that Dembski and Marks work is written in jello. In fact here is a paper for you to peruse, in which Dembski and Marks refer to biological evolution and how NFL Theorem still applies: “Active Information in Evolutionary Search.”
Here is the Abstract:
“In his critique of intelligent design, HÄaggstrÄom [12] claims that the No Free Lunch Theorem (NFLT) [10, 24, 30], properly understood, poses no obstacle to biological evolution. He therefore rejects claims to the contrary [8]. To prove his point, HÄaggstrÄom cites several examples of evo- lutionary optimization. Yet his examples prove the opposite. As he admits, "In almost all concrete optimization problems, we have some prior information...." Far from showing that the NFLT places no restrictions on evolution, HÄaggstrÄom's examples show that the success of evolutionary searches depends on prior information concerning target location and search space structure. Consistent with the NFLT, this prior information, now increasingly referred to as "active information," is always external to the search and thus never a free lunch.”
As to being “informal and imprecise” ... is that before or after this paper: “Conservation of Information in Search: Measuring the Cost of Success.”
Zachriel:
"Let me know if it gets published in a reputable journal of mathematics."
Sure. I’m sure it’ll get there eventually, after much kicking and screaming.
Zachriel:
"You cite yet another unpublished and significantly flawed paper."
Upon reading the rebuttal, it seems that the only significant flaw was in the statement that the EA in question performed worse than random search and this misunderstanding seems to have been the result of a bug in the software. Other than that, there is no rebuttal of the equations for measurement and necessity of problem specific active information within evolutionary algorithms. Am I missing something or are you missing something?
Dembski and Marks work provides a quantification of problem-specific information and they have discussed how active information is necessary for EAs to work (which has not been refuted) and all this conforms to the NFL Theorem and Theorems of Generalization Performance and the measurement of active information provides the foundation for Conservation of Information.
As to being “informal and imprecise” ... is that before or after this paper: “Conservation of Information in Search: Measuring the Cost of Success.”
Wolpert wasn't referring to the concepts in that paper; he was talking about Dembski's book NFL which was written before Dembski ever heard of Marks' concept of active information.
The paper you mention doesn't even state a conservation of information theorem, much less prove it, although it does refer to Schaffer's and English's conservation theorems. Would you like to take a stab at stating the conservation theorem that D&M are referring to in the title?
Sure. I’m sure it’ll get there eventually, after much kicking and screaming.
I'm quite certain that it will not. Care to bet on this?
Upon reading the rebuttal, it seems that the only significant flaw was in the statement that the EA in question performed worse than random search and this misunderstanding seems to have been the result of a bug in the software.
You are correct. That was not intended as a full rebuttal -- it simply pointed out one of the reasons that their numerical results are wrong.
Other than that, there is no rebuttal of the equations for measurement and necessity of problem specific active information within evolutionary algorithms.
I don't know what you mean by equations for measurement. If you mean the definition of active information, then of course definitions can't be refuted. And the "necessity of problem specific active information within evolutionary algorithms" is a tautological claim, since any search algorithm that performs better than random has active information simply by definition of active information. As such, the claims that are correct are also devoid of substance. All they tell us is that ev runs better than random, which we already knew. The only substantial claim was that his evolutionary algorithm performed worse than chance, which is wrong.
they have discussed how active information is necessary for EAs to work (which has not been refuted)
Of course it hasn't been refuted, since it's tautological.
CJYman:
"As to being “informal and imprecise” ... is that before or after this paper: “Conservation of Information in Search: Measuring the Cost of Success.”
secondclass:
"Wolpert wasn't referring to the concepts in that paper; he was talking about Dembski's book NFL which was written before Dembski ever heard of Marks' concept of active information.
Ah, yes, that’s because it was written as a popular science book and Dembski has said before that the book was never meant to be the technical explanation. His technical papers, which I’m sure can be found on his site are his technical papers. I know ... redundant ... his technical papers are his technical papers. But, apparently I have to explain this extremely simple redundancy since it was assumed that his popular science book *was* his technical paper while ignoring his *actual* technical papers.
secondclass:
"The paper you mention doesn't even state a conservation of information theorem, much less prove it, although it does refer to Schaffer's and English's conservation theorems. Would you like to take a stab at stating the conservation theorem that D&M are referring to in the title?"
Well, I’ve read through Dembski’s formulation of a Conservation of Information Theorem and as far as I can tell, it shows how the information necessary to find an improbable target requires at least that amount of information to find that original information necessary to find that target. I know, extremely non-technical and in my own words. Basically, if it takes problem specific information to find CSI, then it takes at least that same amount of problem specific information to find the problem specific information that was used to find the CSI. IOW, if random search will not find CSI, then a random set of laws (that is, not guided by previous problem specific information) will not generate an algorithm to find the necessary evolutionary algorithm which can find the CSI at better than chance performance.
I’ve explained it a little better at the end of this post.
CJYman:
"Sure. I’m sure it’ll get there eventually, after much kicking and screaming."
secondclass:
"I'm quite certain that it will not. Care to bet on this?"
Nah, let’s just wait it out and rationally discuss why it would or why it wouldn’t eventually get published. I’m all about reasonable argument here and, to myself, gambling just doesn’t come across as rational in the sense of having a point. But, that’s just me.
CJYman:
"Upon reading the rebuttal, it seems that the only significant flaw was in the statement that the EA in question performed worse than random search and this misunderstanding seems to have been the result of a bug in the software."
secondclass:
"You are correct. That was not intended as a full rebuttal -- it simply pointed out one of the reasons that their numerical results are wrong."
“Were wrong” (as based on a program fault) ... it has been revised. And, where’s the rest of the rebuttal? Do you have anything substantial?
CJYman:
"Other than that, there is no rebuttal of the equations for measurement and necessity of problem specific active information within evolutionary algorithms."
secondclass:
"I don't know what you mean by equations for measurement. If you mean the definition of active information, then of course definitions can't be refuted. And the "necessity of problem specific active information within evolutionary algorithms" is a tautological claim, since any search algorithm that performs better than random has active information simply by definition of active information. As such, the claims that are correct are also devoid of substance. All they tell us is that ev runs better than random, which we already knew. The only substantial claim was that his evolutionary algorithm performed worse than chance, which is wrong."
“*All* they tell us is that EV runs better than random?!?!?!” It also tells us that the reason it runs better than random is because of knowledge of the target being incorporated into EV.
Thus, it also tells us that EV did *not* create any new information (as a measure of probability of search). These are quite significant and important claims, since they seem to run contrary to the claims that the creators of some EAs are trying to sell to both the scientific establishment and the public.
I can see how you can say that “the necessity of active info in EA is a tautological claim,” however that does not make it any less true as definitions can be tautological yet true and useful. To understand the point it is better phrased as such ... “The difference between an inefficient and an efficient search algorithm is active information. Active information is knowledge of the search problem incorporated into the search algorithm thus causing it to perform better than chance performance." Now, what causes active information? If active information is based on target/problem information then whatever causes active information (and thus evolutionary algorithms as per the tautological claim) must be able to incorporate knowledge of a future problem to be searched into a search algorithm.” Of course, you could also make the claim that active information can arise through random processes (ie: a random set of laws), but then you’d have to test that claim and/or at least begin by providing some mathematical groundwork.
So, you *are* missing something quite significant. The equations I was referring to were those probabilistic equations which were used to show that it is the target specific programmed structure which guides the search algorithm and that increases the probability of finding the target. IOW, evolutionary algorithms are programmed to be efficient in search for a specific problem in which characteristics of that problem must be known and incorporated into the search algorithm before the search begins.
These probabilistic measurements are discussed on the evolutionary informatics website in Conservation of Information: Measuring the Cost of Success, under points “A. Repeated Queries,” “B. Frequency of Occurrence,” “C. Partitioned Search”, and “D. Random Mutation.”
As such, the measurement of active information is a measurement of target specific information (knowledge of target characteristic) which is added to a search algorithm which causes that search algorithm to become an evolutionary algorithm and thus perform efficiently.
CJYman:
"they have discussed how active information is necessary for EAs to work (which has not been refuted)"
secondclass:
"Of course it hasn't been refuted, since it's tautological."
So, then you agree that EAs always have knowledge of problem characteristics programmed into it by a system which has knowledge of characteristics of the future problem to be searched, which is the only way to achieve better than chance performance? Well, that’s great, since it is really the only reasonable explanation for the success of EAs that I’ve seen and it’s also founded upon NFL Theorems and COI Theorems. Well, unless you’ve got some better testable hypothesis ... welcome to the Naturalistic ID ranks.
Post a Comment