So, something like natural selection can work in a computer to produce artificial webs that are more efficient in catching flies than the original webs. This is still not quite true natural selection, but it is a good step closer to natural selection than the pure artificial selection of the biomorphs. But even NetSpinner still is not true natural selection. NetSpinner has to make a calculation to decide which webs are good enough to breed from and which are not. A decision has to be made by the programmer about how costly a given length of ‘silk’ is, in the same currency as the value of a ‘fly’. The programmer could, at will, change the currency conversion rate. He could, say, double the {64} ‘price’ of silk. This could lower the breeding success of larger or denser webs which, for the sake of catching a few extra flies, are a bit extravagant with silk. The programmer has to decide the currency conversion himself and he could choose any conversion factor, at will. This is only one of many such currency conversions that are going on under the surface. The rate at which fly ‘flesh’ is converted into baby spiders is also decided by the programmer. It could be different. The extent to which spiders die for other reasons, having nothing to do with how good their webs are, is also implicitly decided by the programmer. The decision is arbitrary, and a different decision might produce a different evolutionary result.
In real life, none of these decisions is arbitrary. None of them is really a decision at all, and no computational machinery is used to make them. They just happen, naturally and without fuss. Fly flesh just is converted to spider offspring flesh, and the currency conversion factor just is. If we come along afterwards and calculate it, that is our business. The conversion happens automatically, whether anybody puts it into mathematical economic terms or not. The same goes for the conversion of insect flesh into silk. NetSpinner, in effect, assumes that all flies are the same as each other. In real life there may be formidable complications of detail and these, too, emerge simply and without fuss. Quite apart from the fact that some insects are larger than others, there could be subtle qualitative differences. Suppose that, in order to make silk, a particular amino acid which is in short supply is necessary. Different kinds of insects vary in how rich they are in this particular amino acid. The true calculation of the value of an insect, then, has to take account of what kind of insect it is as well as how big it is. NetSpinner could compute this kind of effect, but it would be another arbitrary calculation.
Figure 2.13 (overleaf) Fifty generations of evolution of three sexually reproducing demes of computer webs bred by ‘natural’ selection in NetSpinner. In the eleventh generation, two web genotypes from Deme 3 migrated into Deme 2 where they were available for crossbreeding (indicated in this illustration by the solid arrows). {65}
In real life it just happens, automatically and without contrivance. Here's another complication. Presumably the value of one extra fly is less when a spider is nearly full than it is when she is nearly empty. NetSpinner ignores this, real life does not. NetSpinner could make an arbitrary calculation to allow for the complication of satiation. In real life it just happens, willy-nilly. No explicit calculation has to be done.
The point I am making is so obvious that it hardly needs making, yet so important that it must be made. Every time an additional and complicated point of detail is incorporated into NetSpinner, extra pages of difficult computer code have to be written by a clever human programmer. Yet in real life there is, by contrast, a marked lack of explicit computation. The currency-conversion factor between fly protein and silk protein is just automatically there. The fact that a fly is more valuable to an empty spider than to a full one needs no imported computation. It would be surprising if food were not more valuable to an empty spider. We are accustomed to seeing a computer model as a simplification of the real world. But there is a sense in which computer models of natural selection are not simplifications but complications of the real world.
Natural selection is an extremely simple process, in the sense that very little machinery needs to be set up in order for it to work. Of course the effects and consequences of natural selection are complex in the extreme. But in order to set natural selection going on a real planet, all that is required is the existence of inherited information. In order to set a simulated model of natural selection going in a computer, you certainly need the equivalent of inherited information, but you need a lot else besides. You need elaborate machinery for calculating lots of costs and lots of benefits and the assumed currencies for converting one to another.
More, you need to set up a whole artificial physics. We chose spider webs for our example because, of all devices in the natural world, they are among the simplest to translate into computer terms. Wings, backbones, teeth, claws, fins and feathers: in principle we could make computer models of all of them and the computer could be programmed to judge the efficiency of variant forms. But {68} it would be an aggravatingly complicated programming task. A wing, a fin or a feather cannot show its quality unless it is placed in a physical medium — air or water — with properties such as resistance, elasticity and patterns of turbulence. These are hard to simulate. A backbone or a limb bone cannot show its quality unless placed in a physical system of stresses, leverages and frictions. Hardnesses, brittlenesses, elasticities of bending and compression — all would have to be represented in the computer. To simulate the dynamic interactions among lots of bones, strutted at various angles and roped together by ligaments and tendons, is a formidable computational task involving arbitrary decisions at every turn. To simulate air flow and turbulence around a wing is such a difficult problem that aero-engineers frequently resort to models in wind tunnels rather than attempt to simulate it on a computer.
I must not underestimate the work of computer modellers, however. The discipline of Artificial Life’ was named in 1987 and I was honoured to be invited to the christening in Los Alamos, once the home of the atomic bomb, now turned to more constructive purposes. Christopher Langton, the inspiration and convenor of the original 1987 conference and its successors, has now founded a journal of artificial life, the first volume of which has just arrived. It contains articles that already lighten the pessimism of the previous paragraph. For example, three North Americans called Demetri Terzopoulos, Xiaoyuan Tu and Radek Grzeszczuk have written a spectacular simulation of computer fishes which behave like real fishes and interact with each other in simulated computer water. The computer world in which these fishes swim has its own simulated physics, based upon the real physics of water. Much of the programming effort goes into the simulation of a single fish, getting its behaviour right. Then this working fish is reproduced many times with variations and they are all released into the ‘water’ where they ‘notice’ each other and interact with each other. For example they avoid ‘colliding’ with each other and they associate with each other in ‘schools’.
Each computer fish has an anatomy built up of twenty-three nodes, arranged in simulated three-dimensional space and linked to {69} their neighbours by ninety-one ‘springs’ (Figure 2.14). Twelve of the springs are capable of contracting: they are the ‘muscles’ of the artificial fish. The sinuous swimming movements of a real fish, including turns, are simulated by controlled waves of contraction passing among the ‘muscles’. The fish can learn from experience to improve the sequencing of muscular contractions to swim, turn and follow targets. Fishes have three ‘mental state variables’ called ‘hunger’, libido’ and ‘fear’, which combine to generate ‘intentions’. Intentions include ‘eat’, ‘mate’, ‘wander about’, leave’ and ‘avoid collision.The fish has two sense organs, one that measures the ‘temperature’ of the water and one that acts as a crude ‘eye’, detecting the position, colour and size of objects out there in its world. For cosmetic purposes, the skeleton of nodes and springs is clothed in a solid-seeming, fish-coloured envelope. Different kinds of fish, for example predators and prey, are distinguished, not only by different external cosmetic renderings, but also by differences in behaviour (Figure 2.15). Pr
edators differ from prey, not just in their size but in their behavioural predispositions, the weightings given to the three mental state variables and the various ‘intentions’. Even with today's fast computers, by the way, simulations
Figure 2.14 Artificial fish. The skeleton of springs. {70}
Figure 2.15 Artificial shark stalking a school of prey fish.
of this kind are so costly in computer time that an artificial world containing many interacting fish cannot provide a plausible illusion in real time. Fish swim and chase each other, flee from each other and court each other on a slower time scale than in the real world, and we have to resort to the equivalent of time-lapse photography if we want to be entertained at life speed. This, however, is a detail, of no great theoretical importance: a problem that will disappear with future generations of computers.
Terzopoulos, Tu and Grzeszczuk's artificial world of fish in computer water is rich enough to be a good candidate for evolutionary simulation. At present, although their fish ‘mate’, this is limited to courtship behaviour: they do not actually reproduce. An obvious next step, of which the authors are well aware, is to set up ‘genes’ for the quantitative weightings of the various behavioural variables governing the muscle springs, and, at a higher level, the mental state variable and intentions. Males and females who mate could recombine their genes, with occasional mutation, to produce new generations of differing genetic constitution. Evolution by natural selection, albeit in the {71} ultimately artifical environment of the computer, would then follow. There might be no need to define two kinds of fish called predators and ptey. You might start with two species that differ only in size and mating compatibility but not in habit, and natural selection might naturally lead the larger species to evolve, over many generations, the habit of preying on the smaller. Who knows what intriguing quirks of artificial natural history might emerge before our eyes?
I foresee, and look forward to, a burgeoning field of research to which one might give the oxymoronic title Artificial Natural Selection. Nevertheless, there is a sense in which the easiest ‘simulation’ of the real world of natural selection is the real world itself. Bones actually do vary in breaking strain, compression elasticity, hardness, lines of force and expense in calcium consumption. You can calculate the details if you want to but, whether you calculate them or not, the fact remains that some bones break and others don't; some bones consume lots of precious calcium while others leave calcium spare to put in milk. Real life is starkly simple in this sense. Some animals are more likely to die than others. The fastest computer in America could spend a year costing out and calculating the details. But in nature the brute fact is, that some do die and others don't. That's all.
You can, if you wish, think of the genes in all the populations of the world as constituting a giant computer, calculating costs and benefits and currency conversions, with the shifting patterns of gene frequencies doing duty for the shuttling 1s and 0s of an electronic data processor. It is quite an illuminating insight to which we'll return in the closing pages of this book. But now it is time to illuminate the title. What is Mount Improbable and what shall we learn from it? {72}
<<
* * *
>>
CHAPTER 3
MOUNT IMPROBABLE REARS UP FROM THE PLAIN, LOFTING its peaks dizzily to the rarefied sky. The towering, viftical cliffs of Mount Improbable can never, it seems, be climbed. Dwarfed like insects, thwarted mountaineers crawl and scrabble along the foot, gazing hopelessly at the sheer, unattainable heights. They shake their tiny, baffled heads and declare the brooding summit forever unscalable.
Our mountaineers are too ambitious. So intent are they on the perpendicular drama of the cliffs, they do not think to look round the other side of the mountain. There they would find not vertical cliffs and echoing canyons but gently inclined grassy meadows, graded steadily and easily towards the distant uplands. Occasionally the gradual ascent is punctuated by a small, rocky crag, but you can usually find a detour that is not too steep for a fit hill-walker in stout shoes and with time to spare. The sheer height of the peak doesn't matter, so long as you don't try to scale it in a single bound. Locate the mildly sloping path and, if you have unlimited time, the ascent is only as formidable as the next step. The story of Mount Improbable is, of course, a parable. We shall explore its meaning in this and the next chapters.
The following is from a letter that The Times of London published a few years ago. The author, whose name I have withheld to spare embarrassment, is a physicist, regarded sufficiently highly by his peers to {73} have been elected a Fellow of the Royal Society, Britain's most distinguished learned institution.
Sir, I am one of the physical scientists ... who doubt Darwin's theory of evolution. My doubts arise not from any religious motive or desire to add fuel to either side of any controversy but merely because I think that Darwinism is scientifically indefensible.
...We have no option but to accept evolution — all the fossil evidence points to it. The contention is only about the cause. Darwin maintains that the cause was chance: as generation succeeded generation there would be minor variations at random, those that gave some advantage would persist and those that did not would disappear. Thus living beings would gradually improve with, for example, enhanced powers of obtaining food or of destroying their enemies. This process Darwin called natural selection.
As a physicist, I cannot accept this. It seems to me to be impossible that chance variation should have produced the remarkable machine that is the human body. Take only one example — the eye. Darwin admitted that this defeated him — he could not see how it could have evolved from a simple light-sensitive organ ... I myself can see no alternative to the hypothesis that living matter was designed. The origin of life is not explainable in terms of standard science nor is the wonderful succession of living creatures formed throughout the thousands of millions of years of this planet's existence.
But who was the Designer?
Yours faithfully,
The author is at pains to let us know, twice, that he is a physicist, which gives special weight to his views. Another physical scientist, a professor of chemistry at San Jose State University, California, has burst into biology with a publication called ‘The Smyrna Fig requires God for its Production. He describes the remarkable complexity of the relationship between figs and their wasp pollinators (see Chapter 10) and he comes to the following conclusion: ‘A young wasp lies dormant in a caprifig all winter, but hatches at the exact time to lay her eggs in the summer crop of caprifigs which is necessary to pollinate the fruit. This all requires exact timing which means God {74} controls it’! (The exclamation mark is mine.) ‘To think that all of this exact pattern resulted from evolutionary chance is preposterous. Without God nothing like the Smyrna fig could exist... Evolutionists pretend that things arise by chance without a definite purpose or a completely thought out plan.’
One of Britain's most famous physical scientists, Sir Fred Hoyle (incidentally the author of The Black Cloud, which must be among the best science-fiction novels ever written), frequently expresses a similar view with respect to large molecules such as enzymes, whose inherent ‘improbability’ — that is the probability that they'd spontaneously come into existence by chance — is easier to calculate than that of eyes or figs. Enzymes work in cells rather like exceedingly numerous machine tools for molecular mass production. Their efficacy depends upon their three dimensional shape, their shape depends upon their coiling behaviour, and their coiling behaviour depends upon the sequence of amino acids which link up in a chain to make them. This exact sequence is directly controlled by genes and it really matters. Could it come about by chance?
Hoyle says no, and he is right. There is a fixed number of amino acids available, twenty. A typical enzyme is a chain of several hundred links drawn from the twenty. An elementary calculation shows that the probability that any particular sequence of, say 100, amino acids will spontaneously form is one in 20 × 20 × 20 ..
. 100 times, or 1 in 20100. This is an inconceivably large number, far greater than the number of fundamental particles in the entire universe. Sir Fred, bending over backwards (unnecessarily, as we shall see) to be fair to those whom he sees as his Darwinian opponents, generously shortens the odds to 1 in 1020. A more modest number to be sure, but still a horrifyingly low probability. His co-author and fellow astrophysicist, Professor Chandra Wickramasinghe, has quoted him as saying that the spontaneous formation by ‘chance’ of a working enzyme is like a hurricane blowing through a junkyard and spontaneously having the luck to put together a Boeing 747. What Hoyle and Wickramasinghe miss is that Darwinism is not a theory of random chance. It is a theory of random mutation plus non-random cumulative natural selection. Why, I wonder, is it so hard for even sophisticated scientists to grasp this simple point? {75}
Darwin himself had to contend with an earlier generation of physical scientists crying ‘chance’ as the alleged fatal flaw in his theory. William Thomson, Lord Kelvin, was perhaps the greatest physicist of his day and Darwin's most distinguished scientific opponent. Among his many achievements he calculated the age of the Earth based on rates of cooling, assuming that it had once been a part of the ‘fires’ of the Sun. He concluded that the Earth was some tens of millions of years old. Modern estimates put the age up in the thousands of millions of years. It is no discredit to Lord Kelvin that his estimate was one hundredth part of the right answer. Dating methods using radioactive decay were not available in his time, and nuclear fusion, the true ‘fire’ of the Sun, was unknown, so his cooling calculation was doomed from the start. What is less forgivable was his lofty dismissing, ‘as a physicist’, of Darwin's biological evidence: the earth wasn't old enough; there hadn't been enough time for the Darwinian process of evolution to have achieved the results we see around us; the evidence of biology must simply be wrong, trumped by the superior evidence of physics. Darwin might just as well have retorted (he didn't) that the biological evidence clearly indicates evolution, therefore there must have been time for evolution to occur, therefore the physicist's evidence must be wrong!