Conservation of information sparked scientific interest once a recurring pattern was noticed in the evolutionary computing literature. In grappling with the creation of information through evolutionary algorithms, this literature consistently revealed that the information outputted by such algorithms always needed first to be programmed into them.
That’s simply not true. Creationist and ID literature made that claim, but it was soundly refuted - here, for example, where Sal Cordova insisted that a genetic algorithm (GA) was only regurgitating what was programmed into it, but failed to find the solution that the GA found despite having the code. Neither Cordova, Ewert or Dembski himself have ever explained how a programmer can smuggle into a GA something that they don’t know, or why, if GAs have information smuggled into them, they don’t always produce the same answer.
The paper is full of the usual rubbish about Weasel programs, evolving words, Axe’s work, etc etc etc. At 58 pages long, life is too short to waste reading this.
There is similar stuff in Marks, Ewert, and Dembski’s book “Introduction to Evolutionary Informatics”. They examine search algorithms looking for (for example) solutions to the Steiner Tree Problem. They find a place in the code where there is evidence that the code has been tweaked to make it do better. However, these tweaks are not specific to the particular problem the code is trying to solve. If it were to have the answer coded in, then that would be different for every set of points that the algorithm was confronted with. But the tweaking is not different in each case. MD&E give the impression that they had found specific information, that guides the algorithm in solving the particular case. They found nothing of the sort in the case of the STP, and also in some other cases. So Dembski’s statement that “the information outputted by such algorithms always needed to first be programmed into them” is, as Roy says “simply not true”.
For some 2019 discussion making the same point, see a thread at The Skeptical Zone which has comments by some of us explaining the falsity of the front-loaded-information assertion.
IIRC, in one of their papers they mistook the STP for the minimum spanning tree problem, in another they waffled on about Tierra outputs without ever having run it, and on one of their Weasel stomping days they miscounted the number of matching letters they had.
It is best to ignore those mistakes. Can we “steel man” (the opposite of “straw man”) their argument and make a valid version? Answer: no, not even close.
PS I wrote TSP in my previous comment the second time I referred to the problem. Of course it should have been STP.
A disconnect I notice frequently is that scientists typically think about evolution as an explanation for life’s diversity since the last common ancestor, while many apologists critique it as if it is the atheist’s alternative to any sort of divine action. Further, apologists tend to think of the alternative to intervention as uniform randomness.
In this case, perhaps what Dembski et al. are trying to get at is that any input from the programmer/scientist which biases the outcome in a particular direction–in contrast to uniform randomness–represents an input of information. It may not be the exact details of the eventual solution, but it is a thumb on the scale at odds with their notion of evolution. To my mind, it is more of a fine-tuning argument than an anti-evolution argument. But I think for most apologists, that is not an important distinction; anything which suggests input is needed beyond what would occur naturally is a point against atheism and thus evolution.
Of course, I could be wrong in this instance. I’ve not yet had the opportunity to read either work in their entirety. But I’ve written my own evolutionary simulations and encountered the criticism of “smuggling in” the answer. So I’ve given some thought to try to figure out where that sentiment is coming from, because at face value it has little connection to the code I’ve written.
On the other hand, using only Unix Words, an exhaustive computer search shows that running through all possible evolutionary paths starting from the word A, evolution cannot reach the words HYMN, ENVY, TOFU, OBEY, and IDOL.
A-O-OF-OFF-TOFF-TOFU
A-AY-BY-BEY-OBEY
A-I-ID-IDO-IDOL
A-AN-EN-HEN-HYEN-HYMEN-HYMN
Dembski needs a better vocabulary and a better dictionary since he apparently doesn’t know the word ‘toff’.
The information, as I’m sure you are aware, comes from the environment that determines the selection. In the case of a computer simulation, the fitness function represents the environment. A technical criticism of this might be the “smoothness” of the fitness function, but a Genetic Algorithm can still do pretty well even with the worst case scenario of a completely random fitness function.
The smoothness of the fitness surfaces is strongly related to physical laws that imply that forces decline rapidly with distance, so a chemical reaction in my earwax has little effect on the synthesis of proteins for my toenail. And there is separation in time as well. So a gene affecting one part of the body interacts very little with one affecting another.
I’ve made this criticism before (at PT, 10 years ago). Dembski in his latest 58-page paper, spends a bit of time refuting it by claiming that I argued that all fitness surfaces are smooth. Then he shows off his command of physics by saying that there are some very non-smooth processes. A gross straw-man argument. There are still ample reasons to believe that many fitness surfaces are much smoother than a “white noise” surface. And in those cases we do not need any supernatural intervention to get the fitness surface to be smooth.
Sure, I’m aware that the process of selection over variations results in genomes storing information connected to the environment. And I imagine that Dembski et al understand that as well. Which is why I feel like there must be more to the criticism, especially when they point at features of the code beyond the fitness function. Hence my hypothesis that they mean something broader when talking about information.
Of course, such overloading of terms doesn’t really help with clarity. And maybe they consider that a feature rather than a bug. Or maybe they think they are being clear, but just coming from a very different frame of reference.
Well, they have now changed their emphasis from Complex Specified Information, which is related to Szostak, Hazen, Griffin, and Carothers’ “Functional Information”, to Algorithmically Specified Complexity (ASC). They also continue to promote the conservation of information, which we have been critiquing here. Neither of these concepts has been developed in a way that shows that high values of information cannot be achieved by repeated rounds of natural selection.
Harder-core creationists have their own notions of information, which seem to me different entirely.. It is all very confusing. Different notions, and no open disagreements between their proponents.