The moment being the thought that natural selection is just an unnecessarily complicated version of searching, the sort of thing that google is so good at. Apart from the thought being striking, I suppose the value add is that the analogy suggests new ways of thinking about both natural selection and searching.
We suppose, like google, that we have a very large space to search. There is lots of stuff out there, it does not really matter what, organised as some sort of high dimensional metric space. We have the notion that one point in the space might or might not be near another point, might or might not be a neighbour.
We also have value, so we can associate each point with a real number and this point is better than that point if it has a higher value. Computing the value of a point might be more or less complicated; it might, for example, not be enough just to look at the design and one might actually have to bring the thing to life and give it a go. Part of Google’s success is the trickery it deploys to do all this. The people at Autonomy, the people embroiled in an alleged, large scale accounting fraud, used to claim to have an even better trick, the details of which they were always very coy about.
To speed things up we have a number of search processes, which we will call dogs. Possibly a large number, a regular pack of attack dogs.
At any one point in time, each dog is at a particular place, or point, in the search space. Each dog is trying to get to the best point in the space and to that end, being rather short sighted and unable to take a view of the space as a whole, at each move he (or she) moves to some neighbouring point which looks better than the one he is standing on. There will almost always be such a better point.
There is some kind of higher authority, a divinity if you will, a divinity which beats out the time of the process as a whole, and dogs which seem to be making progress are replicated. Dogs which are not doing so well are killed off. The divinity might, of course, have other things on his (or her) mind. He might have decided up front, without bothering to do any proper testing, that blue eyes is a good thing, and so bears down on any tendencies in other directions. Interferes with the supposedly objective value system. And sometimes he will get this wrong, he will decide on, for example, blue eyes, with it turning out in the end that blue eyed humans have other traits which he does not care for at all.
Eventually, some of the dogs arrive at very good places. Maybe, but this is unlikely, summits from where there is nowhere else to go. The end of the line.
Then, moving onto natural selection proper, let us suppose that the space we are searching is the space of possible humans and the dogs are searching for decent people to belong to.
At each move, the dog brings to life a human corresponding to the spot or point that it is standing on and takes it out for a run. Good run then good value. In which we must remember that the space is a space of possibilities only, with the number of actual humans being the same as the number of dogs. The dog then looks about for a neighbouring point which will do better and when he finds one that looks promising, he jumps ship. Abandons first human in favour of second human. Takes this new one out for a run. And so on and so forth.
Eventually we get to a fairly stable population of dogs with very good owners. There is a bit of variation ongoing, over time, but not all that much.
Until there is global warming or something that changes the rules of the game and searching moves into another more active phase.
The message from Fernando and his colleagues seems to be, in so far as I understand it, that sometimes the natural selection sort of searching is a good way forward, sometimes other kinds of search techniques are better. This in the context of brains trying to organise themselves.
PS: it seemed appropriate to use an illustration selected for me by google. Whereas autonomy, had it been available to me, could have generated a search term from the text of this post and got me to a picture that way.
No comments:
Post a Comment