What we did is we invented a system in which you could develop a device – it looks more or less like a robot in fact but it's not a robot – and instead of running it by a computer with logic and programs, you simulate a brain in the computer and have the brain controlling the behavior of this device in the real world. You don't simulate the real world; you actually have it in a platform where it searches in space for different objects and what have you. Well, this seemed like a sort of quixotic episode because, first of all, you're not going to be able to get 30 billion connections, even with the most powerful computer. Second of all, it hasn't got a body like ours. But third of all it does have a brain that's very similar to ours in the sense that we use the same principles and we never cheat. We never do a computer program with written algorithms or anything like that to control things. Now that's radically different from what's called artificial intelligence. The artificial intelligentsia actually control robots by writing logical programs and feedback circuits and things of that kind. In the case of brain-based devices, or BBDs as we call them, you have a brain which actually has synapses, real neuroanatomy of this kind, resembling this species or that species, and an environment in which it moves and samples and gets input, and finally it has a built-in value system just like you do, which you inherited from evolution. And what we've been able to show is that these devices... where we can look at every single neuron, these devices will actually learn. They will condition, they will do just what Pavlov did with his dog: get... get particular responses and correlate their visual responses, say, with their taste – I'll come to that in a moment – and they can actually even locate themselves in a room. So let me use the example starting with so-called Darwin 4. By the way, we name these things after the great man; we're up to Darwin 10 now and working on Darwin 11.

Darwin 4 was the first time we had a real-world device; looks a bit like R2D2, in which it had cameras for eyes, it had microphones for ears, it had a value system built into its nervous system, it had a nervous system with definite anatomy that controlled its wheeled movements right, left, sideways, it had reflexes just like an animal, but it was given a scene in which for instance blocks could be picked up by something called a gripper at the bottom of its structure, and when it grabbed the block, it would feel the resistance or the impedance of that block or the conductance. And things that had a low conductance, these metallic blocks, would be bad-tasting, and things with a high conductance would be good-tasting – could as well of been the other way around. Now the fact is each block had a different design on its top, for instance stripes or blobs, and it turned out we assigned blobs to be bad-tasting. Well, what would happen is Darwin 4 would patrol around, look at things that reflected light, go toward them, pick up a block, say, that had blobs on it and was bad-tasting, and that would change its neuronal circuits. It would then patrol to another block that was stripes, parallel stripes, and it would... that would be good-tasting and that would change positive circuits. And pretty soon you would notice that what it would do is avoid blocks that were blobbed and pick up blocks that had stripes, all on its own without instruction, and we could find out everything that was going on in that case. And we found out that we could even get it to condition secondarily; once it was visually conditioned to like stripes and not like blobs, we could hook that up to high tones and low tones that the blocks emitted, and it would condition to those too.