I think there's actually a lot to be done in evolution that hasn't been done. Simulated evolution. I started doing a little bit of it at Thinking Machines. I'd love to get back to it. But the basic idea is right now, if when people describe evolution, they really describe it in one of two ways. They either describe it... the most simple way is the kind of Richard Dawkins way of every gene has some independent selective value and this is the sort of classical way of population biologists describe it. And you can describe how quickly genes get selected. But that sort of suggests that a gene has this independent value. And I think that's way too simple an idea, that really interesting things happen out of combinations of genes. And so the mathematics that says how quickly genes evolve doesn't work very well for combinations of genes. In fact, it gets exponentially slower for combinations of genes. So if you try to do the calculation, you'll see that a couple of billion years is not enough to check out all the combinations of genes that you need to do something complicated. So there must be some other factor at work.

This is something that right now the evolutionists don't like to talk about much, because evolution is under attack from religion. Still. And so evolutionists don't like to admit that there's sort of a fundamental hole in the story. But the timescales don't really work out when you look at combinations of genes. So the only story we really have about combinations of genes is either that you evolve something simple and there's some path of simple improvements that goes from, let's say, a single cell bacterium to a human being by improving it one step at a time. And that is possible, but I think something more is going on and I think that, in fact, nature does things much more like an engineer does things, which is that nature develops modules which interact in limited ways with other modules and there's causes and effects and boundaries, but I can change how this one works, I can improve this one without changing its output behaviour, without changing the interface to the next one, so that there are certain interfaces that evolve more slowly. But what happens inside the modules evolves more quickly.