So. William Dembski, the supposed “Isaac Newton of Information Theory” has a new paper out with co-author Robert Marks. Since I’ve written about Dembski’s bad IT numerous times in the past, I’ve been getting email from readers wanting me to comment on this latest piece of intellectual excreta.
I can sum up my initial reaction to the paper in three words: “same old rubbish”. There’s really nothing new here – this is just another rehash of the same bankrupt arguments that Dembski has been peddling for years. But after thinking about it for a while, I realized that Dembski has actually accomplished something with this paper: in his attempt to argue that evolution can’t possibly outperform random-walks without cheating, he’s actually explained exactly how evolution works. He attempts to characterize that as cheating, but it doesn’t work.
Let me start with a quick review. Dembski has, for years, been pushing an argument based on some work called the No Free Lunch (NFL) theorems. The NFL theorems prove that average over all possible search landscapes, no search algorithm can outperform a random walk. The NFL theorems are true and correct – they’re valid math, and they’re even useful in the right setting. In fact, if you really think about it, they’re actually quite obvious. Dembski has been trying to apply the NFL theorems to evolution: his basic argument is that evolution (as a search) can’t possibly produce anything without being guided by a supernatural designer – because if there wasn’t some sort of cheating going on in the evolutionary search, according to NFL, evolution shouldn’t work any better than random walk – meaning that it’s as likely for humans to evolve as it is for them to spring fully formed out of the ether.
Thes doesn’t work for a very simple reason: evolution doesn’t have to work in all possible landscapes. Dembski always sidesteps that issue.
Let me pull out a metaphor to demonstrate the problem. You can view the generation of a notation for a real number as a search process. Suppose you’re given π. You first see that it’s close to 3. So the first guess is 3. Then you search further, and get closer – 3.14. That’s not quite right. So you look some more, and get 3.141593. You’ll get closer and closer to a notation that precisely represents π. Of course, for π, you’ll never get to an optimum value in decimal notation; but your search will get progressively closer and closer.
Unfortunately, most real numbers are undescribable. There is no notation that accurately represents them. The numbers that we can represent in any notation are a miniscule subset of the set of all real numbers. In fact, you can prove this using NFL.
If you took Dembski’s argument, and applied it to numbers, you’d be arguing that because most numbers can’t be represented by any notation, that means that you can’t write rational numbers without supernatural intervention. Of course, that’s rubbish. The way that we create notations for rational numbers works, because natural numbers are a small, structured subset of the set of all real numbers. We don’t expect them to work for all possible numbers, and any proof about notations that relies on reasoning about how well a notation works on all real numbers is absolutely irrelevant to a discussion of notations for the rationals.
The same thing is true of evolution. Ignoring all of the (very serious) problems with modeling evolution as a search process, evolution doesn’t need to work in all possible search spaces; it needs to work in one particular set of search spaces that by their nature have a lot of structure. Evolutionary processes exploit that structure.
Now, after that downright Oracian introduction, let’s get to Dembski’s latest paper.
It gets off to a very bad start. It starts with a section titled “The Creation of Information”. This is, frankly, a muddled mess. Unfortunately, I think that that is deliberate. You see, Dembski uses very peculiar definitions of information; or, to be more precise, he doesn’t use any consistent definition. He pretends to use Shannon information theory, but he tends to vacillate between the Shannon formulation; his own mangled probabilistic formulation; and Kolmogorov-Chatin. In this section, he purports to talk about creating information – but what he actually does is try to muddy things up by mixing up a tiny bit of Shannon theory with a variety of non-mathematical philosophers talking about the meaning of information. That’s not a particular useful endeavor: what Chesterton meant when he talked about acts of will as self-limitation is not the same thing as what Shannon meant when he said that information is the elimination of possibilities. The whole section is just at attempt to confuse things by equating the mathematical definition of information in Shannon theory with philosophical definitions of information. The problem is that philosophers define information in terms of meaning; mathematical information theory doesn’t give a rats ass about meaning: a spinning neutron start generates more information (without any intrinsic meaning) in a second than Dembski’s meaningful actions will in his entire lifetime.
That leads directly into section two: “Biology’s Information Problem” – a thoroughly redundant section of the paper. Whether his argument makes sense or not, this is just a rehash of what he said in the previous section, and what he’ll repeat in later sections. “Life can’t create information without intelligence, because information doesn’t exist without intelligence, because according to my conflation of mathematical information and philosophical meaning, it makes no sense to talk about information absent intelligence”. Of course, when it comes down to it, that’s Dembski’s entire argument: by definition information requires intelligence; therefore if living things contain information, it must have been created by intelligence. (If I were Michael Egnor, I’d just wave my hands and say that it’s just a tautology, therefore it’s meaningless, and be done with it.)
Section three is Dembski once again rehashing hiss argument against Dawkins’ “weasel” example. The selection function in “weasel” knows the target, and therefore it’s cheating by “smuggling” information into the search, and evolution couldn’t do it without cheating. God but I’m tiring of his repeating that same idiotic argument. It’s a tiny, silly, throwaway example that demonstrates one minor feature of how evolutionary processes work; it was never intended to be a complete example of how evolution produces life. Based on his refutation of Dawkins, Dembski concludes “Evolution, despite Dawkins’s denials, is therefore a targeted search after all.” Sorry Bill, but that’s bullshit, and you know it.
Section four: “Computational vs. Biological Evolution” gets even worse. It’s Dembski’s attempt to pull his “Universal Probability Bound” into the picture. The UPB is one of Dembski’s dumbest ideas: it’s an argument that there’s some probability threshold where anything less probably is absolutely impossible. (Of course, he doesn’t want to admit that it’s his own stupid argument – so he attributes it to Seth Lloyd. And of course, what Lloyd actually said is quite different from what Dembski tries to imply that he meant.) What’s sad is just how badly Dembski builds an argument around this.
Anyway, he wants to build up the argument that you can’t possibly have an evolutionary “search” produce what we see of life in the entire lifespan of the universe. He starts by talking about an IBM supercomputer than runs at just over a 1 petaflop, and how large the search space that it could explore is: if it took one floating point instruction per sample from the search space, then in the lifespan of the universe, it could have searched 1034 samples. That’s supposed to impress you – and it’s used in the following paragraphs as part of one of the typical idiotic probability arguments about life – that’s the fastest computer ever, and it could only search 1034 samples. But he then goes on to say “It is estimated that 8 t the total number of organisms, both single-celled and multi-celled, that have existed on the earth over its duration (circa 4.5 billion years) is m = 1040. Thus it would take a million Roadrunner supercomputers running the duration of the universe to sample as many “life events” as have occurred on the earth.”. Wow, way to undermine your own argument, Bill. You’ve just admitted that a million of the fastest computers ever, running since the creation of the universe, couldn’t search as much space as the biological history of the earth.
As I mentioned, then he launches into another of those idiotic probability arguments. (It’s like he doesn’t want to miss the chance to include a single stupid argument!):
Most search spaces that come up in the formation of biological complexity are far too large to be searched exhaustively. Take the search for a very modest protein, one that is, say, 100 amino acids in length (most proteins are several hundreds of amino acids in length). The space of all possible protein sequences that are 100 amino acids in length has size 20100, or approximately 1.27×10130, which exceeds Lloyd’s limit. For this space, finding a particular protein via blind search corresponds to a 1 in 10130 improbability. Exhaustively or blindly searching a space this size to find a target this small is utterly beyond not only present computational capacities but also the computational capacities of the universe as we know it.
Why do we need to see this bullshit time and time again? No one with the slightest shadow of a clue claims that life requires exactly one possible protein out of all of the incredible number of possibilities. No one with the slightest shadow of a clue claims that all possible proteins are equally likely. No one with the slightest shadow of a clue would claim that the production of any protein is the result of a targeted search for that protein. This is pure stupidity – a pathetic strawman argument that’s been discredited literally hundreds of times. But Dembski trots it out and babbles about it, at length.
Finally, in section 5: “Active Information”, Dembski finally gets to the point: that any search that creates new information must actually have that information somehow included in the search function. And of course, to make that argument, once again he trots out “Weasel”. Yes, once again – the argument comes down to “The reason that weasel works is because the selection function knows its target.” Well, duh. Yeah, Bill? Weasel is a stupid throwaway example. Care to take on something actually real? Like, say, some of the fantastic e.coli experiments? No, I didn’t think so.
Then it’s time for obfuscatory mathematics. In case you haven’t seen the term before, it’s what I call the use of pointless equations that do nothing except look really complicated, and give you the appearance of having actually done something deep. One of the basic facts of math, like programming, is that it’s garbage-in, garbage-out. You can derive really incredibly looking equations using perfectly valid proofs, to demonstrate any conclusion that you want. But your conclusion is only valid in a setting where you’ve accurately modeled the reality that you’re trying to describe.
Search is a lousy model for evolution; general search is a particularly lousy model. I’ve discussed it plenty of times before – for example here – but the problems basically come down to a few simple points:
- As a search, evolution is a multidimensional search. Most of our intuitions about search landscapes is based on two or three dimensions. But evolution as a landscape has hundreds or thousands of dimensions; our intuitions don’t work.
- Evolution is a dynamic landscape – that is, a landscape that changes in response to the progress of the search. Pretty much everyargument that Dembski makes can be thrown out on the basis of this one fact: all of his arguments are based on static landscapes. Once the landscape can change, every single one of his arguments become invalid – none of them work in dynamic landscapes.
- As a search, evolution doesn’t have to work on all possible landscapes. It doesn’t even need to work on most landscapes. It works on landscapes that have a particular kind of structure. It doesn’t matter whether evolution will work in every possible landscape – just like it doesn’t matter that fraction notation doesn’t work for every possible real number. What matters is whether it works in the particular kind of landscape in which our theory says it works. And on that question, the answer is quite clear: yes, it works.
Anyway – using his lousy model of evolution as search, he comes up with what he calls the “Law of conservation of information”. We’ll ignore, for the moment, the fact that it’s not actually a conservation law at all. What it says, basically, is that if a search algorithm performs better than blind search by some factor f(I), that algorithm must have been “purchased” at an information cost I.
What does that “purchased” mean? Basically, that the search function must include, in some form, an amount of information about the search space no smaller that I.
Which comes back to one of the whole objections to the whole NFL-based approach: Dembski insists on searches that perform equally well over all possible landscapes. Saying that a search algorithm performs only on some set of search spaces is exactly the same as saying that the search algorithm contains information about those search spaces!
Dembski’s argument comes down to “Where did that information come from?”; and his answer is that it must have been put there by someone. How do we know that? Well, because this whole argument just showed that for a process to generate information, that process must have contained the information to begin with. So because information doesn’t come from nowhere, it must have been put there. It’s actually a subtly circular argument: we’ve just gone through this mass of obfuscatory mathematics to derive a set of theorems that prove that successful searches must in some sense encode information about the landscapes that they search. But that doesn’t really matter to the argument. Because the real argument is that only an intelligent agent can create information. The whole exercise of deriving the so-called conservation of information laws was based on the premise that you can’t create information; and now after all of that noise, we’ve come full circle to show that the conservation of information laws can be used to prove that you can’t create information. The whole argument is ultimately concluding one of its premises. We can eliminate all of the faux math, and reduce the argument to its simplest form: “Only an intelligent agent can create information, therefore only an intelligent agent can create information”.
Back at the beginning of the paper, I said that Dembski actually manages to basically refute his own argument – that he shows how evolution can actually work. By now, you should see how that happens: this whole argument comes down to asking what it means to drop the “over all possible landscapes” part of NFL. If you do that, then you end up with a search algorithm that can perform very well on some set of landscapes. Which is exactly what us lousy evolutionists have been saying all along.