I was intrigued to see William Messinger's review of Richard Dawkins' book, The Blind Watchmaker, in the May-June issue of Contrast. As a computer scientist, I was attracted to Dawkins' book recently because of his use of computers, and in particular by his "biomorph" program. [See Note 1]
Richard Dawkins is a biologist who believes in incremental evolution. He is convinced that the accumulation of small changes in the genetic code of living organisms throughout the history of the earth can explain all of the order and variety visible today. Replying to William Paley's argument for the existence of God by the analogy with order and complexity in a watch requiring a watchmaker, Dawkins says that,
"Natural selection, the blind, unconscious, automatic process which Darwin discovered, and which we now know is the explanation for the existence and apparently purposeful form of all life, has no purpose in mind. It has no mind and no mind's eye. It does not plan for the future. It has no vision, no foresight, no sight at all. If it can be said to play the role of the watchmaker in nature, is is a blind watchmaker." [page 5]
Dawkins wrote his book to promote his point of view. In it he describes a computer program that simulates evolution by drawing tree-like structures called "biomorphs" (see Figure 1) on the computer screen. Each drawing is controlled by sixteen "genes" -- twelve of them numerical values that control particular aspects of the drawing. The program randomly "mutates" one or more genes each generation, presenting the user with 14 mutated offspring from which to select a new parent for future breeding. The results are a surprising variety of shapes, resembling variously trees, flowers, bugs, and so on. Not all at once, but the accumulation of random mutations, selected for "survival" by the user, can lead to many different kinds of shapes depending on the sequences of selection choices.
Figure 1. Breeding Biomorphs in the Blind Watchmaker program.
Undirected random mutation leads nowhere, as Dawkins recognizes. Natural selection, however, provides direction. Supposing it to be somewhat like a target against which each mutation is tested for similarity, he contrived a computer experiment to simulate a monkey randomly typing on a typewriter, but only changing a previously "fittest" sentence, one letter at a time. Each change that more closely resembles the target line from Shakespeare survives, while each change that does not is discarded. Not surprisingly, the computer program very quickly reproduced the correct line -- Dawkins reports times well under a hundred generations. As a mathematician and computer scientist I found these figures incredible so I wrote a program to reproduce the results. I was unable to attain matches in less than 600 generations without constraining the mutations to letters not yet matching the target -- a kind of Maxwellian "demon" that allows only anti-entropic evolutionary changes. This model is biologically incorrect, but that error does not significantly affect Dawkins' point.
The fundamental flaw in Dawkins' reasoning lies in anti-entropic assumptions about natural selection. I will save that discussion for another time and concentrate here on the time considerations in the evolutionary model. In other words, for our purposes now we will assume that natural selection makes no errors, is not blind, and inexorably advances evolution toward ever greater complexity and order. This assumption cannot be sustained on entropic grounds, but even if it were true, we find that evolution is not possible in the time scale allotted to it.
A human being has a genetic makeup containing something like 100 billion codons or DNA radicals [see Note 2]. At each locus any one of the four possible DNA radicals could be encoded, but with the exception of perhaps a few thousand genes that determine variability such as physical features and genetic diseases, most of them must be a specific codon for the organism to be a viable human. It is widely believed that there are many "junk" codes in the DNA that do not contribute to the viability of the organism, but that has not been established. Let's assume that only one billion codons (1%) are required to specify a viable human; the other 99% are assumed to be absolutely useless, encoding no proteins and contributing nothing to the survival of the species.
For natural selection to work, each generation of evolutionary improvement must contribute only improvements to the organism (or else leave the genetic code unchanged); changes away from the target do not contribute to the evolutionary progress and are killed off by being less fit for survival. Since the chance of a mutation in a particular DNA locus has only 25% probability of matching the target, multiple simultaneous mutations render evolutionary progress increasingly improbable, a fact that Dawkins recognizes. We assume, therefore, that each generation in the evolutionary tree from the dawn of life to modern humanity represents only one incremental genetic step. This means that the evolutionary tree has one billion separate (genetically different) organisms in the line from protozoan #1 to modern man. Since the current theory places the dawn of life at something like one billion years ago, we find that evolution necessarily advances on the average, one step every year. That is a full cycle, parent giving birth to a mutated child who reaches reproductive maturity and gives birth to a new mutated child, in one year. It is clear that later in the chain the reproductive cycle is many years, but early on (protozoan through early segmentation stages) it can be reasonably assumed that it was less than one year. We conclude therefore, that for evolution to have occurred, natural selection must be able to identify and select for improvements immediately in the first generation of a mutation's appearance, a supposition so ridiculously optimistic that not even the most zealous evolutionists argue for it.
Note that we are concerned only with the single line of heredity from the common ancestor to the first man. Each organism in that line must bear at least one offspring with a beneficial mutation and no harmful mutations. Other siblings in the family, and other organisms in the then-current species may have beneficial or harmful mutations, or no mutations at all, since they do not contribute to the genetic heritage of humans. If the chance of a single DNA locus mutating is about 10-9 and if each organism bears four or more offspring, then the chances are good for an evolutionary improvement in that generation. If the mutation rate is significantly higher (say for example, 10-6 as suggested by some) then each offspring is likely to suffer more harmful mutations than good, and the probability of successful transmission of progress dwindles to zero -- unless there is a little demon guiding the mutations the way Dawkins' Shakespeare program did, so that mutations only occur on the DNA that has not yet matched its target. If the chance of mutation is significantly less than our assumption, then we cannot sustain the necessary evolutionary rate.
We assumed for our calculations that natural selection never made any mistakes, never deviated from its distant target. Dawkins recognizes that "life isn't like that. Evolution has no long-term goal. There is no long-distance target, no final perfection to serve as a criterion for selection." [page 50] If we take this view then of course we need much more time for the evolution of man.
Dawkins recognizes that the real-life genetic codes have unobvious effects
on the organism, and he designed his biomorph program to have similarly
unobvious connections between the genes and the shape of the biomorph.
And while it is important to him to continually deny a long-term objective,
it seems to creep back into his discussion from time to time. At the end
of the paperback edition of his book he offers a $1000 reward for the genetic
formula for a particular biomorph: since it resembled a jewel-encrusted
chalice, Dawkins called it "the Holy Grail" (see Figure 2). The reward
and the printed image serve as a long-term target to which the reader is
thereby encouraged to guide the random evolution of the program. If the
shape of the biomorph were related to the genes in some obvious way, there
would be no contest and no prize. But not only is the relationship somewhat
unobvious, the genes interact between themselves in unobvious ways, so
that a biomorph that looks similar to the target may not necessarily be
genetically related to it. Dawkins designed this property into the program
to mirror his perception of life.
|Figure 2. The "Holy Grail"||Figure 3. Applying scientific principles of
design to deduce its "genetic code"
If a person were to take Dawkins up on his challenge, using the program to try to evolve the holy grail by the accumulation of small chance mutations and selecting the best fit each time could take millions of years. Or perhaps millions of computers. On the other hand, Dawkins' program, like nature and unlike the holy grail biomorph, is the product of careful design to achieve a particular purpose -- namely the simulation of evolution -- and like all products of design, the program is susceptible to scientific investigation and analysis. Where there is no design, the results of analysis yield no coherent results and no comprehensible theory. Taken on its own terms, the holy grail was not open to analysis: it was just a pattern of pixels. But the program was something else.
I am a programmer, and I understand what a program has to do to accomplish its purposes; I made a living doing that kind of design. I am also an educator now, and I see a lot of student programs that lack coherent design. I read about Dawkins' program about a year after it was published, and after asking to make sure the prize had not yet been claimed, I sent away for the program. I fully expected to disassemble it to see how it related the genes to the drawing; that proved unnecessary. Although it is not without a few bugs, the program was sufficiently well-designed that I could apply the scientific method to its functioning and infer the complete laws of genetics for its artificial world. I fiddled with the genes for a while, then began formulating hypotheses and running experiments to verify or falsify them. Within a couple hours I understood the rules well enough to write my own program by the same rules.
I knew I would be busy for the next two or three weeks -- too busy to tinker with finding the holy grail -- so I wrote a program to try directed evolution. You might think of my program as a kind of theistic evolution: the mutations were systematic and directed toward the target. For three weeks my program ran continuously on two computers, churning out biomorphs at the rate of 1000 per second and testing them for similarity to the target. If I had let it go, it might have found the grail in a couple years. I chose not to wait.
Taking pencil and paper in hand, I then approached the problem from a designer's perspective: What combination of genes would give the particular pattern of pixels along the top of the cup? I drew colored lines on the paper representing the computer graphics screen drawing effects, and calculated what the interactions would yield (see Figure 3). Within three days I had calculated appropriate values for all twelve genes, and breathlessly entered the numbers into Dawkins' program. The chalice appeared! Well, not quite. I often tell my clients and students, "God gets His creations right on the first cut; the rest of us have to debug our work." In this case I had misunderstood how one of the genes worked. A little more analysis and a couple more hours of testing solved the problem. I won the prize [See Note 3] in three days of creative analysis.
What does all this prove? Dawkins intended his program to prove the
viability of incremental evolution as a means of explaining order and diversity
in nature. It does nothing of the sort. It does prove that a well-designed
information system can effect any sort of natural laws that the designer
cares to program into it. Taken with the results of my efforts to win the
prize, it suggests that the creation hypothesis generates better scientific
analyses and better engineering solutions than the evolution hypothesis.
Finally, considering the timescale requirements for incremental evolution,
we can conclude that
The Watchmaker is not blind; the biologist looking over His shoulder is blind.
Department of Computer and Information Sciences
Kansas State University
Manhattan, KS 66506
2. This was written before the Human Genome Project had produced any results. The current estimate is closer to 6 billion, but that does not significantly impact the force of the time argument.
3. Actually, I barely missed winning the
prize. When I didn't hear from the publisher for several weeks, I called
to ask what happened. The woman who spoke to me told me I was the third
to submit a solution. When Richard Dawkins was informed of the winning
entry, he reportedly said "So soon? I thought my money was safe." I persuaded
her to give me the name and phone number of the winner, and I spoke to
him at length. I asked if he used the evolve mode of the program, and he
said he tried it, but it kept undoing his progress. Instead it took him
three months, fiddling with the "genes" manually.