ENGINEERING BY EVOLVED NATURAL SELECTION:
The marvelous Richard Dawkins gives a short account of contraptions made by a German engineer that gradually 'evolved' from crude starting materials. Include a wind harnessing system of propellors inspired by the principles of bird flight, and an 'intelligently designed' (!) foil that is engineered by 'natural selection' to minimize air drag. This presentation should put to rest one of the many misconceptions about evolution- that it is so 'random' and hinges upon such a fortuitous happenstance, that it simply could not have taken place naturally in such a 'short' time. An illuminating presentation lasting only a few minutes
These days, I have strongly started to think that it is relatively easy to harbour doubts about evolution if one has not read into it in some detail. For example, questions about the improbability of complex life arising 'randomly' by mutations from simple chemical precursors in a short period of time, is what can be only called the "argument from incredulity"- we find it hard to believe simply because we cannot readily imagine it. No wonder the 'intelligent designees' can hoodwink people in a hardbeat. A more incredulous stance would actually be the belief that simply because something is incomprehensible to us human beings means that it cannot have taken place in nature. Talk about smug and self-centered satisfaction!
A little glance into some of the details of evolution should be enough to convince us of the utter beauty, logic and very much probable simplicity of the process. In retrospect, evolution should look infinitely more simple and beautiful than the transcendental and much more complex process of some inconceivable super-designer designing such a complex world.
posted by Wavefunction at 7:23 PM
1 Comments:
Interesting stuff.
I have a couple of questions though: anyone who has programmed stochastic gradient descent algorithms, self organizing maps and such would know that the convergence of the algorithm depends on several factors, like the initial conditions, update rate etc, to name a few. I do not have much experience with evolutionary algorithms, but I do not know if the algorithms are guaranteed to converge or not. Similarly, many of the above mentioned algorithms (stochastic etc.) converge under fairly specific conditions, like the function being a quadratic form. Hence, it is not clear to me that evolution will always have a solution to a problem. Of course, if a problem does have a minimum, evolution is likely to find it, but I do not know how the mutation rate etc. was set in the case of these experiments and how that would affect the final solution, if at all. It is quite possible that we do not understand evolutionary problem solving and hence cannot prove convergence with present means. If you know the answer, I would like to hear it. The other problem is that of beginning with arbitrary initial conditions. Would the model always converge for different initial conditions to the same final solution?
It would be really interesting to know if these algorithms suffer from such problems, and if so, what kind of solutions emerge.
On a somewhat unrelated note: there was a brief report in last year's Nature around this time, of self replicating robots. That was interesting too.
Aniket
Post a Comment
<< Home