One of the things I did during that time but never published, was to try and calculate what was the reliability of biological machines. And I was very interested in this because there was a paper by von Neumann showing how you would make computing... reliable computing systems with unreliable elements. What he suggested was in fact later used in all the space missions. Basically if you have unreliable elements like computers and you have to have reliable performance, you put three of them in the machine, and of course you take the majority vote of them. And so even if one blanks out it's very unlikely that all three will blank out on the same thing. And if you really want to be really reliable, you do four of them, because the probability that they'll all go wrong is of course P⁴. So that kind of thing is well known in... engineering, and I was very stimulated by that, and also by what living cells could or couldn't do. And many of these theories really required that cells could count molecules exactly. That is, they could say, I will make 20 molecules and somebody else would make 21. And you see, it's a beautiful way of getting a gradient. So if you just have a rule, I'll make one more molecule than my neighbour, that's all you need, then of course you get a beautiful linear gradient going from one to whatever you like. Now, the point is this: can cells actually make... know the difference between 20 and 21? I mean, do they have reliable counting machinery at the molecular level? And you can show quite easily that they can't. That is, they don't work accurately at that level, therefore everything in development that you posit... that's done by dead reckoning won't work. As indeed was discovered in Stanford Research Institute when they first tried to program robots by dead reckoning. These robots were given a problem of moving around in a room with obstacles, and what they had to have, their wheels were powered by stepping motors, so after making a calculation of the geometry of the room, they then worked out how many little steps they have to give to each motor so that the whole machine would move on a path, avoiding the obstacles, to get to that end. And of course there are errors in this. And so that meant that after sufficient of these had been added together, these robots launched themselves out of windows into the street below. The windows were closed so they never landed there. And it's well known that dead reckoning isn't the way to do it, you can't navigate by dead reckoning and what you have to do is to correct your position at each point. That's why people get out - on ships - get out and have a quick look at the stars through all their apparatus, so they can correct it. And that's what you need, you need feedback, you need to say: I've got so far, am I in the right place, if not, correct it. That means an input to the system. So the system cannot work in general by just saying: I've got it in my genes and I'll just read it out. So this will move so many microns, turn right, do another thing and then hook up. That won't work. What you've got to do is say: go in that general direction, when you get there, check it out. That is, if you receive a signal there, make the adjustments. You need the feedback. And so that showed that one had a much more elaborate mechanism in which environment and... and environment's used in the most general sense, because of course in development the environment isn't, you know, the sea or the field, the environment is other cells, the environment would have to make an input. Now we have a system in which we can have this feedback, we can have a... a path being set and we can switch it. And that began to make sense rather than this rather mechanical idea that you look up in a table where you are and what you're going to be. And as matters have turned out, that's the correct way of viewing it. And I think quite a lot of confusion was caused by people who kept on saying it's gene regulation and had a picture in their heads, if not a model, but at least a picture of, you know, the top gene who would instruct you know, the number two genes, you guys get out and do this, and that they would instruct other genes. And of course one knows that this is the great paradox of all administrative systems, is why have people tell other people to do things if they could just as well go out and do them themselves, if that's all they're doing. And in fact, this kind of hierarchical command system seems to be favoured by humans, at least intellectually, as a way of describing complex systems, because maybe that is the only way they can conceive of classifying them in some way, by departments. But I always feel that that... we should suspect these easy analogies because they are likely to be wrong, because these analogies operate in our conscious minds which are very restricted and not necessarily... and systems that I think are capable of multiple interactions can do much better than this one... this command hierarchical systems.