I think the genetic algorithm is seriously cool. Like an absolute triumph of human engineering. They took the evolutionary fairy tale, and boiled down its best bits, and put them into a computer program… suitably modified.
Where it goes off the rails is when someone thinks it somehow proves, or helps prove, the idea of evolution. Or, worse, when people think that what it does somehow reflects what evolution is said to do, and have done. In this post I wish to deal with one aspect of the absolute failure of the algorithm to reflect evolution: the Bootstrap Problem.
Lifting in Vain
Hopefully everyone is already familiar with the ‘bootstrap’ problem, but in case some aliens from Mars (or non-native English speakers) have wandered in: the expression is meant to ridicule someone who thinks that the poor and downtrodden can ‘lift’ themselves out of poverty.
Now, as to poverty, the expression is, perhaps, not all that true. Lots of poor people have left poverty, and many of them have done so with a lot of their own effort. But as far as the physics of the expression is concerned, it is spot on. The man who reaches down and pulls on his own bootstraps is not lifting himself up; he is merely pulling his upper body and his feet together.
Thus the expression, “You can’t lift yourself up by your bootstraps”. And yet that is, figuratively speaking, what evolution is meant to do: for an organism to change its own genetic code, right in the middle of a functioning organism, which will go on to produce more functioning organisms.
A Prominent Difference
Which is not what the genetic algorithm does. At all. Quite different, actually:
The genetic algorithm is operating on a functioning computer
The proposed changes make no difference to the operation of the computer. The operating system is not changed; the form and function of the mouse remains constant.
The computer, along with the operating system and running program, does not reproduce itself, keeping everything working, into a downstream population which keeps on running the genetic algorithm.
IOW no bootstraps are involved.
Starting in the Wrong Place
The genetic algorithm starts in the wrong place: it does not produce the genetic algorithm. If you say that is a silly objection, I will reply that that is because it is a silly idea. No one envisions that an operating system will write itself, still less that it will help design and build the machine it runs on, because that is a truly stupid idea.
And the designers of the GA knew that. They did not sit around and stare at a pile of spare computer parts sitting in some warm pond. They took a fully functioning computer built in China or something, and used its fully functioning features to program in the GA.
Which then spit out… data. Not instructions for the rebuilding of the computer and operating system, but… data.
The Difference Makes a Difference
Now I want to point something out. Suppose that the designers had found the wait by the warm pond to be a little boring, and wanted to short-circuit the process. They could (well, they couldn’t, but that comes later) have instead set up a self-replicating computer (I guess it would be a robot), and put the GA into its operating system.
And then they could have made it so the operating system and computer were modified by the instructions rewritten by GA over generation after generation.
Except… they couldn’t. Because while they were able to invent and use the GA on a fully functioning computer, and I’m sure will redesign and refine it; and while others are able to produce a functioning computer capable of running the GA… no one on Earth can yet create a self-replicating computer.
You Bring Your Own Dirt
I will leave it to my audience to look up or remember the ‘You bring your own dirt’ joke. What I want to caution here is a false view of what I am demanding. Even if we allow the designers to skip the ‘warm soup’ steps… to truly imitate evolution, they would not only have to produce a self-replicating computer, but one that did so from native material, competing with others for it.
The temptation would be to create one in some nice clean room, with spare parts and the like helpfully lying around, all organised. No, to truly imitate evolution (after skipping the most difficult steps), the GA would need to be running the OS and reproduction system on a self-replicating, self-feeding, self-repairing machine.
Conclusion
Which, of course, these brilliant designers are utterly incapable of building. Utterly. Perhaps in a hundred years advanced AI might make a shot at it. But for the moment the particular problem that I wish to point out is that the GA is not actively reprogramming the operating system it depends on to run. The value, or lack, of its output makes no difference to the running of the computer.
The GA algorithm is head and shoulders above the “Methinks it is a weasel’ simulation, although it suffers from many of the same flaws… just fancied up a good deal. But neither of them, nor any other proposed algorithm, is willing to try lifting itself by its own bootstraps.
The basis of my faith is probably pretty stupid to some people, but if I had to put it down to a gross oversimplification it pretty much stems from a refusal te believe that I came from rocks, or rock soup or whatever. I suppose dust isn't that far removed if you think about it...