I participate in a discussion forum called the Online Philosophy Club. In the context of one of the conversations I came across a link to The Evolutionary Informatics Lab, run by proponents of Intelligent Design. What follows is a response to one of the papers on the site.
The paper is called LIFE’S CONSERVATION LAW: Why Darwinian Evolution Cannot Create Biological Information, by William A. Dembski and Robert J. Marks II. It promised to be a robust opposition to the view that natural selection is the principal driver of evolution.
At least half of the paper is a kind of mathematics which goes completely over my head. However there are also words, and these are what I will focus on. I am not even attempting to question the maths. But I do want to dig a bit into what assumptions there might be behind the maths.
The paper attempts to establish the ‘Law of Conservation of Information’, and draws out possible implications for natural selection:
Though not denying Darwinian evolution or even limiting its role in the history of life, the Law of Conservation of Information shows that Darwinian evolution is inherently teleological.
A combative tone is struck early on:
Nature, as conceived by Darwin and his followers, acts without purpose—it is nonteleological and therefore unintelligent. As evolutionary geneticist Jerry Coyne puts it in opposing intelligent design, “If we’re to defend evolutionary biology, we must defend it as a science: a nonteleological theory in which the panoply of life results from the action of natural selection and genetic drift acting on random mutations.” [Emphasis added.] But why do Coyne and fellow Darwinists insist that evolutionary biology, to count as science, must be nonteleological. …Where did that rule come from? The wedding of teleology with the natural sciences is itself a well established science—it’s called engineering. Intelligent design, properly conceived, belongs to the engineering sciences.
But this analogy is false. Engineering is applied science. We know some scientific facts, and we apply those facts intentionally to solve a practical problem. There is no question of assuming or speculating that those facts were what they are because of any prior intelligent design or intentionality. If we apply the analogy of engineering, then the wedding of teleology with the biological sciences is called agriculture, not intelligent design. A better parallel would be the science of meteorology when the received wisdom was that gods were behind changes in the weather. This is regardless of the merits of intelligent design as an evolutionary hypothesis.
I would also quibble with the described role of ‘information’ in living things:
The emergence of life constitutes a revolution in the history of matter. A vast gulf separates the organic from the inorganic world, and that gulf is properly characterized in terms of information. The matter that makes up the bricks of your house and the matter that makes up your body is essentially the same. Nevertheless, the arrangement of that matter—the information—differs vastly in these two cases.
I would agree that living things possess and embody information. But so do non-living things. If there was a crucial distinction characterising the gulf between the organic and the inorganic world, it is that living things replicate autonomously whereas non-living things do not. Certainly living things use information to do that replication, and therefore their information is different. But the crucial difference is surely autonomous replication, not possession of information per se?
More significant is the attack on Richard Dawkins’ demonstration (in The Blind Watchmaker) of how a random string of 28 characters could evolve into the phrase METHINKS IT IS LIKE A WEASEL in less than 100 random mutation cycles, as long as directed selection was applied after every cycle. Dembski and Marks seem to misunderstand Dawkins’ point – I don’t know whether or not this is deliberate.
Dawkins’ point was to demonstrate incremental selection over single-step selection, which in this example would have had odds of 1 in 1040 against it. Dawkins was not claiming that a person selecting matches after each step was not the application of intelligence.
Dembski and Marks however seem to think Dawkins was trying to demonstrate natural selection, not incremental selection:
So does Dawkins’s evolutionary algorithm demonstrate the power of the Darwinian mechanism to create biological information? No. Clearly, the algorithm was stacked to produce the outcome Dawkins was after. Indeed, because the algorithm was constantly gauging the degree of difference between the current sequence from the target sequence, the very thing that the algorithm was supposed to create (i.e., the target sequence METHINKS•IT•IS•LIKE•A•WEASEL) was in fact smuggled into the algorithm from the start. The Darwinian mechanism, if it is to possess the power to create biological information, cannot merely veil and then unveil existing information. Rather, it must create novel information from scratch. Clearly, Dawkins’s algorithm does nothing of the sort.
Note the words ‘stacked’, ‘smuggled’, ‘veil’ and ‘unveil’.
In their very next paragraph:
Ironically, though Dawkins uses a targeted search to illustrate the power of the Darwinian mechanism, he denies that this mechanism, as it operates in biological evolution (and thus outside a computer simulation), constitutes a targeted search. Thus, after giving his METHINKS•IT•IS•LIKE•A•WEASEL illustration, he immediately adds: “Life isn’t like that. Evolution has no long-term goal. There is no long-distant target, no final perfection to serve as a criterion for selection.”
There is no irony, except perhaps in Dembski and Marks’ disingenuousness. Yes Dawkins uses a targeted search, but to illustrate incremental selection rather than ‘the Darwinian mechanism’ (presumably natural selection). Yes Dawkins denies that biological evolution is a targeted search.
My point is this. Dembski and Marks may be right about intelligent design, natural selection and the Law of Conservation of Information. But I don’t think they’re right in their interpretation of Dawkins’ intention behind his METHINKS IT IS LIKE A WEASEL example.
They go on:
Dawkins here fails to distinguish two equally valid and relevant ways of understanding targets: (i) targets as humanly constructed patterns that we arbitrarily impose on things in light of our needs and interests and (ii) targets as patterns that exist independently of us and therefore regardless of our needs and interests. In other words, targets can be extrinsic (i.e., imposed on things from outside) or intrinsic (i.e., inherent in things as such).
Not sure I see the difference between these two. Seems a bit like talking about flags and failing to see the difference between red flags and green flags. In both cases a target is set by someone or something capable of setting a target. I wonder whether a target set by a dog (eg the stick he has decided to chase) counts as (i) or (ii). It’s not a ‘humanly constructed pattern’, and it does ‘exist independently of us’ – but not of the dog. Also if I had a target to lose weight would that be instrinsic (it concerns my weight, and I am the one setting the target) or extrinsic (I might have written my target weight on a notice board)?
Earlier in the paper Dembski and Marks use the example of an acorn growing into an oak tree. Would that qualify as a target, and if so as type (ii)? If so that would beg a huge question. There is a conceptual difference between saying an acorn happens to grow into an oak tree and saying an acorn has the target of growing into an oak tree.
But the fact that things can be alive and functional in only certain ways and not in others indicates that nature sets her own targets.
We need to be careful with this sort of talk. It is OK metaphorically but we are not justified in saying this literally. If intelligent design is true, then perhaps so. But until then to say that ‘nature sets her own targets’ can only be metaphorical, as is:
The targets of biology, we might say, are “natural kinds” (to borrow a term from philosophy). There are only so many ways that matter can be configured to be alive and, once alive, only so many ways it can be configured to serve different biological functions. Most of the ways open to evolution (chemical as well as biological evolution) are dead ends. Evolution may therefore be characterized as the search for alternative “live ends.”
It’s a search in the logical sense, but it doesn’t imply any intentional target-setting, any more than talk of ‘water finding its own level’ implies any intentional targeting on the part of the water.
…In other words, viability and functionality, by facilitating survival and reproduction, set the targets of evolutionary biology. Evolution, despite Dawkins’s denials, is therefore a targeted search after all.
Absolutely not. It might be, if intelligent design is correct. But we cannot assume it is.
The next section, 4 Computational vs. Biological Evolution, argues from big numbers. To arrive at a modest-sized protein by chance (blind search) corresponds to a 1 in 10130 improbability. Indeed. This is the equivalent of arriving at METHINKS IT IS LIKE A WEASEL in one step, which we said was odds of 1 in 1040. But evolutionary theory assumes incremental selection, not single-step selection.
Dembski and Marks continue using the example of Dawkins’ METHINKS IT IS LIKE A WEASEL algorithm in the next section 5 Active Information. The question is whether a computer simulation, in this case not the METHINKS IT IS LIKE A WEASEL algorithm but another program, ev, can generate an increase in information just from ev’s selective process itself.
Dembski and Marks are firmly of the view that ‘algorithms cannot create information but [can] only shuffle it around’. They quote supporting references from physicist Léon Brillouin and biologist Peter Medawar, and then demonstrate the ‘basic idea’ in a ‘straightforward’ way using Dawkins’ WEASEL algorithm:
What allowed his evolutionary algorithm to converge so quickly on the target phrase METHINKS•IT•IS•LIKE•A•WEASEL is that a fitness function gauging distance from that phrase was embedded in the algorithm (indeed, the very target phrase was itself stored in the algorithm). But in that case, fitness functions gauging distance from any other string of letters and spaces could just as well have been substituted for the one Dawkins used; and with those other fitness functions, the algorithm could have converged on any sequence whatsoever.
I completely agree. The measure of fitness (fitness function) at each cycle was degree of fit to METHINKS IT IS LIKE A WEASEL. The fitness function could have been something else.
So the target sequence METHINKS•IT•IS•LIKE•A•WEASEL initially had very small probability p (roughly 1 in 1040) of arising by pure chance from a single query; and it has probability q (close to 1) of arising from Dawkins’s evolutionary algorithm in a few dozen queries. But that algorithm requires a precisely specified fitness function that gauges distance from a target sequence, and such a fitness function can be built on any sequence of 28 letters and spaces (and not just on METHINKS•IT•IS•LIKE•A•WEASEL). So how many such fitness functions exist? Roughly 1040. And what’s the probability of finding Dawkins’s fitness function (which gauges distance from METHINKS•IT•IS•LIKE•A•WEASEL) among all these other possible fitness functions? Roughly 1 in 1040.
Again, completely sound. The probability of stumbling on the correct fitness function by chance would be infinitesimally low. But, again, Dawkins was not trying to demonstrate getting to METHINKS IT IS LIKE A WEASEL by chance. He was trying to show the difference between these two algorithms:
1 Single-step selection:
1.1 Take a random string of 28 characters.
1.2 Test it against METHINKS IT IS LIKE A WEASEL.
1.3 If string = METHINKS IT IS LIKE A WEASEL, then stop.
1.4 If string is not = METHINKS IT IS LIKE A WEASEL, then randomly change the whole string and loop back to 1.2.
2 Incremental selection:
2.1 Take a random string of 28 characters.
2.2 Test it against METHINKS IT IS LIKE A WEASEL.
2.3 If string = METHINKS IT IS LIKE A WEASEL, then stop.
2.4 If string is not = METHINKS IT IS LIKE A WEASEL, then randomly change only those characters which do not match the equivalent characters in METHINKS IT IS LIKE A WEASEL and loop back to 2.2.
According to Dembski and Marks though,
Dawkins’s algorithm, far from explaining how METHINKS•IT•IS•LIKE•A•WEASEL could be produced with high probability, simply raises the new problem of how one overcomes the low probability of finding the right fitness function for his algorithm. Dawkins has thus filled one hole by digging another.
Simulations such as Dawkins’s WEASEL… and Schneider’s ev … capitalize on ignorance of how information works. The information hidden in them can be uncovered through a quantity we call active information. Active information is to informational accounting what the balance sheet is to financial accounting. Just as the balance sheet keeps track of credits and debits, so active information keeps track of inputs and outputs of information, making sure that they receive their proper due. Information does not magically materialize. It can be created by intelligence or it can be shunted around by natural forces. But natural forces, and Darwinian processes in particular, do not create information. Active information enables us to see why this is the case.
Active information tracks the difference in information between a baseline blind search, which we call the null search, and a search that does better at finding the target, which we call the alternative search. Consider therefore a search for a target T in a search space Ω (assume for simplicity that Ω is finite). The search for T begins without any special structural knowledge about the search space that could facilitate locating T. Bernoulli’s principle of insufficient reason therefore applies and we are in our epistemic rights to assume that the probability distribution on Ω is uniform, with probability of T equal to p = |T|/|Ω|, where |*| is the cardinality of *.50 We assume that p is so small that a blind or null search over Ω for T (i.e., a search for T by uniform random sampling of Ω) is extremely unlikely to succeed. Success demands that in place of a blind search, an alternative search S be implemented that succeeds with a probability q that is considerably larger than p.
Whereas p gauges the inherent difficulty of locating the target T via a blind search, q gauges the difficulty of locating T via the alternative search S. The question then naturally arises how the blind or null search that locates T with probability p gave way to the alternative search S that locates T with probability q. In the WEASEL, for instance, Dawkins starts with a blind search whose probability of success in one query is roughly 1 in 1040. This is p. He then implements an alternative search (his evolutionary algorithm) whose probability of success in a few dozen queries is close to 1. This is q.
Dawkins leaves the discussion hanging, as though having furnished an evolutionary algorithm that locates the target phrase with high probability (which we are calling S), he has demonstrated the power of Darwinian processes. But in fact all he has done is shifted the problem of locating the target elsewhere, for as we showed earlier in this section, the fitness function he used for his evolutionary algorithm had to be carefully chosen and constituted 1 of 1040 (i.e., p) such possible fitness functions. Thus, in furnishing an alternative search whose probability of success is q, he incurred a probability cost p of finding the right fitness function, which coincides (not coincidentally) with the original improbability of the null search finding the target. The information problem that Dawkins purported to solve is therefore left completely unresolved!
This is where the maths starts to go over my head. But I’m not sure it’s relevant, because I think the issue is logical rather than mathematical. Dembski and Marks seem to have completely missed the point. I can’t see Dawkins disagreeing with their estimates of relative probabilities. Where I think he would disagree is with the definition of the problem.
In the theory of evolution by natural selection there is no target. Therefore there is no ‘fitness function’ expressible in terms of a target. Yes the WEASEL algorithm had a fitness function expressed in terms of a target, but the WEASEL algorithm was to demonstrate incremental selection versus single-step selection, not evolution by natural selection.
The theory of natural selection is not saying: If replicator X randomly mutates such that it meets selection criterion C, then X will survive to replicate, else X will not survive to replicate. If it was saying something like that, then C would be a target.
The theory of natural selection is saying something more like this: Replicators X, Y, Z… etc are in competition for survival and replication. Replicators X, Y, Z etc are all capable of randomly mutating such that the mutations (a) will affect the replicators’ viability and replication prospects (compared to the competition); and (b) will be inherited by the next generation. If the effect of the mutation in X (if it happens) is such that it benefits the survival and replication prospects of X relative to the survival and replication prospects of Y, Z etc (which may or may not have also been mutating randomly) then the relative population of X will increase at the expense of Y, Z etc. This same ‘algorithm’ applies at each generation. Favourable variations will therefore be preserved, and may be built on, at the expense of unfavourable variations (where ‘favourable’ and ‘unfavourable’ are to be construed purely in terms of relative survival in relation to the competition).
Dembski and Marks seem to want to see the generation of information as the target or goal or essence of life. Then with the assumption of information as the primal concept, they formulate theorems (on the analogy of the conservation of mass and the conservation of energy) to prove that information (defined in terms of probability) cannot be created except by intelligence.
But the theory of evolution by natural selection does not start here, and does not agree with the assumption that we have to start here. It starts with the concept of a replicator in competition with other replicators. In order for a replicator to be a replicator it may need to possess or embody a quantum of ‘information’, but that is secondary. At this level the replicator is not searching for anything. It is not trying to do anything. It is not even trying to survive or replicate. It just does survive and replicate – or it does not.
To refute the theory of evolution by natural selection what is needed is a demonstration of flaws in the concepts of competing randomly mutating replicators, not a demonstration that ‘information’ cannot be created except by intelligence.
Dembski and Marks define ‘information’ in teleological terms. The first sentence of their paper is:
Any act of intelligence requires searching a space of possibilities to create information.
They then ask the straw-man question: Can anything other than intelligence search a space of possibilities to create information? They conclude that nothing can, therefore natural selection cannot.
But natural selection is not ‘searching a space of possibilities to create information’. It is not searching for anything because it does not care. Key to this kind of ‘information’ is the concept of a ‘right answer’. But in natural selection there is no right answer. The replicator either survives to replicate or it does not. A mutation M in replicator X in context C faced with future F1 might have led to replicator X surviving to replicate. The same mutation M in the same replicator X in the same context C but faced with a different future F2 might have led to replicator X not surviving to replicate. So there was no ‘right answer’ at the time of the ‘search’.
© Chris Lawrence 2010.