We were able to show, for the first time, that invertebrate animals like fruit flies can sleep. Now that is a remarkable achievement – not only because you can now do the genetics related to sleep and its control, but also because it simply had never been surmised. And what it's done is provoke an extraordinary set of experiments of Greenspan and his associates, particularly Bruno Van Swinderen, in which you actually can look at selective states that really resemble attention, even if it isn't quite like attention, in which selective effects can be seen in perception of fruit flies in virtual reality mazes. So that's an example outside of my own interests but deeply arresting to my interests, and I'll come back to that. But the one I want to pay attention to has to do with a remarkable problem I think probably unique to the nervous system.

I mentioned the complexity and individuality of every nervous system. I mentioned the number of connections which are simply staggering. I did not mention the fact that these connections are degenerate – namely that the same structure can give rise to different functions and vice versa; that the same function can be connected or correlated with a whole variety of different structures. This is a very important concept in biological science but before I deviate and describe that, I want to stay on the ball and really describe how we attack this problem of originality and uniqueness and degeneracy in the nervous system. You see the immune system was sort of easy because one cell made one antibody. When you got the antibody and you got the cell you could follow things pretty well. But in the nervous system you have these immensely connected graphs; you have these circuits which have innumerable connectivities and paths – a highway system the likes of which just absolutely even probably is greater than what you see on the internet in terms of connectivity. How are you going to deal with that? Well, one problem is this: that if you're even the most advanced kind of neurophysiologist who's measuring the electrical activity of each neuron in a particular part of the brain... in a particularly defined task, you can only sample hundreds of neurons at one time, but you're sampling out of populations of billions, and so you're never sure you're covering the waterfront. Second of all, the system is very changeable, so from animal to animal, even though you get the same function, you're not sure you're testing the same circuit. There's a push, of course, to make everything look very uniform but it really isn't.

And so that forced us to think very hard about this problem: if you have the fact that the brain is embodied and the body's embedded, you have this complexity but you can't sample them all, how are you going to write a model that would only look at, say, synaptic strengths, or a model that only looked at the body/brain interaction, or a model that looked at this environment by modeling it? Well, we came to the conclusion that we'd have to invent a whole new approach, and that approach falls under the rubric of brain-based devices.