In short, the dynamic of evolution is not prescribed in advance: it is `emergent'. It creates its own context, and reacts to that context, as it proceeds. So at any given time we expect to find some sensible directionality to evolutionary change, consistent over many generations, but often the universe itself only finds out what that direction is by exploring what's possible and discovering what works. Over a longer timescale, the direction itself can change. It's like a river that flows through an eroding landscape: at any given time there is a clear direction to the flow, but in the long run the passage of the river can slowly change its own course.
It is also important to appreciate that individual organisms do not compete in isolation, or against a fixed background. Billions of competitions go on all the time, and their outcome may be affected by the results of other competitions. It's not like the Olympics, where the javelin-throwers politely wait for the marathon-runners to stream past. It's more like a version of the Olympics where the javelinthrowers try to spear as many marathon-runners as they can, while the steeplechasers are trying to steal their javelins to turn each hurdle into a miniature pole vault, and the marathon-runners' main aim in life is to drink the water-jump before the steeplechasers get to it and drink it first. This is the Evolympics, where everything happens at once.
The evolutionary competitions, and their outcomes, also depend on context. Climate, in particular, plays a big role. In the Galapagos, selection for beak size in Darwin's finches depends on how many birds have what size of beak, and on what kinds of food - seeds, insects, cactus - are available and in what quantities. The amount and type of food depend on which plants and insects are competing best in the struggle to survive - not least from being eaten by finches - and breed. And all of this is played out against a background of climatic variations: wet or dry summers, wet or dry winters. Observations published in 2002 by Peter and Rosemary Grant show that the main unpredictable feature of finch evolution in the Galapagos is climate. If we could forecast the climate accurately, we could predict how the finches would evolve. But we can't predict the climate well enough, and there are reasons to think that this may never be possible.
That doesn't prevent evolution from being `predictive', hence a science, any more than it prevents meteorology from being a science. But the evolutionary predictions are contingent upon the behaviour of the climate. They predict what will happen in what circumstances, not when it will happen.
Darwin almost certainly read Paley's masterwork as a young man, and in later life he may well have used it as a touchstone for his own, more radical and far more indirect, views. Paley succinctly expressed many of the most effective objections to Darwin's ideas, long before Darwin arrived at them. Intellectual honesty demanded that Darwin should find convincing answers to Paley. Such answers are scattered throughout Darwin's epic treatise The Origin of Species, though Paley's name does not appear.
In particular, Darwin found it necessary to tackle the thorny question of the eye. His answer was that although the human eye appears to be a perfected mechanism, with many interdependent parts, there are plenty of different `eyes' in the animal kingdom, and a lot of those are relatively rudimentary. They can even be arranged in a rough progression from simple light-sensing patches to pinhole cameras to complex lenses (though this arrangement should not be interpreted as an actual evolutionary sequence). Instead of half an eye, we find an eye that is half as effective at detecting light. And this is far, far better than no eye at all.
Darwin's approach to the eye is complemented by some computer experiments published by Daniel Nilsson and Suzanne Pelger [1] in 1994. They studied a simple model of the evolution of a lightsensing patch of cells, whose geometry could change slightly at every `generation', and which was equipped with the capacity to develop accessories such as a lens. In their simulations, a mere 100,000 generations were enough to transform a light-sensing patch into something approaching the human eye, including a lens whose refractive index varied from place to place, to improve its focus. The human eye possesses just such a lens. Moreover, and crucially, at every one of those 100,000 steps, the eye's ability to sense light got better.
This simulation was recently criticised on the grounds that it gets out what it puts in. It doesn't explain how those light-sensing cells can appear to begin with, or how the eye's geometry can change. And it uses a rather simplistic measure of the eye's performance. These would be important criticisms if the model were being used as some kind of proof that eyes must evolve, and as an accurate description of how they did it. However, that was never the purpose of the simulation. It had two main aims. One was to show that in the simplified context of the model, evolution constrained by natural selection could make incremental improvements and get to something resembling a real eye. It wouldn't get stuck along the way with some dead-end version of the eye that could be improved only by scrapping it and starting afresh. The second aim was to estimate the time required for such a process to take place (look at the title of the paper), on the assumption that the necessary ingredients were available.
Some of the model's assumptions are easily justified, as it happens. Light carries energy and energy affects chemical bonds, so it is not
[1] 'A pessimistic estimate of the time required for an eye to evolve', Proceedings of the Royal Society of London B, volume 256 (1994), pp. 53-8.
surprising that many chemicals respond to light. Evolution has an immense range of molecules to draw on - proteins specified by DNA sequences in genes. The combinatorial possibilities here are truly vast: the universe is not big enough, and has not lasted long enough, to make one molecule of each possible protein as complex as, say, haemoglobin, the oxygen-carrier in blood. It would be utterly astonishing if evolution could not come up with at least one light-sensing pigment, and incorporate it into a cell.
There are even some ideas of how this may have happened. In Debating Design, Bruce Weber and David Depew point out that lightsensitive enzyme systems can be found in bacteria, and these systems are probably very ancient. The bacteria don't use them for vision, but as part of their metabolic (energy-gaining) processes. Proteins in the human lens are very similar to metabolic enzymes found in the liver. So the proteins that make the eye did not start out as components of a system whose purpose was vision. They arose elsewhere and had quite different `functions'. Their form and function were then selectively modified when their rudimentary light-sensing powers turned out to offer an evolutionary advantage.
Although we now know quite a lot about the genetics of the human eye, no biologist claims to know exactly how it evolved. The fossil record is poor, and humanoid eyes don't fossilise (though trilobite eyes do). But biologists can offer simple reasons why and how the eye could have evolved, and these alone are sufficient to demolish claims that its evolution is impossible in principle because the eye's components are interdependent and removing any one of them causes the eye to malfunction. The eye did not evolve one component at a time. Its structure evolved in parallel. The instigators of more recent revivals of Paley's doctrine, albeit in less overtly theist tones, have taken on board the message of the eye as a specific case ... but its more generic aspects seem to have eluded them. Darwin's discussion of the eye, and the Nilsson-Pelger computer experiment, are not limited to eyes. Here is the deeper message. When confronted with a complex living `mechanism', do not assume that the only way it can evolve is component by component, piece by piece. When you see a watch, do not think of hooking up springs and adding cogwheels from some standard box of spare parts. Think more of a Salvador Dali `soft watch' that can flow and distort, deform, split apart, and rejoin. Think of a watch whose cogwheels can change shape, grow new teeth, and whose axles and supports evolve along with the cogs so that at every stage the whole thing fits together. Think of a watch that may have started out as a paper clip, and along the way became a pogo-stick. Think not of a watch that does and always did have a single purpose, which was to tell the time. Think of a watch that once held sheets of paper together and could also be straightened out to form a toothpick, and which later turned out to be great for bouncing, and started to be used for measuring time only when someone noticed that its rhythmic movements could chart the passing seconds.