Cornell scientists have applied genetic programming to experimental data in order to elicit fundamental physical laws governing the data. These physical laws include Lagrangians and laws relating to momentum conservation. The article is available here. This is a triumph for genetic computing. If one imagines the evolution of scientific discovery in human lifetime terms, the ability to speed this process via evolutionary simulation is quite interesting. However, these types of algorithms can never replace human inspiration and creativity in developing theoretical models.
Comments
Not a lot.
it got hyped up quite a bit with the talk of robots and such when it
was really a computer doing something we know they can do with enough
parameters
A computer program - which is what this robot boils down to - is generally an implementation of an algorithm or heuristic designed to achieve a highly specific goal. If the machine that follows such a simple logic is a scientist, then I hereby nominate the Jacquard loom for an award in fabric engineering.
Note: a program with rich interconnections amongst its functions and a highly unspecified goal could come up with a new theory and explain it in plain language.
Patrick Lockerby | 04/04/09 | 11:16 AM
If biology has given anything to the "more fundamental" physical sciences, it is the evolutionary paradigm. The notion that you can solve an optimization problem by the combination of mutation and selective recombination operations is a very powerful one, which in some cases enables the solution of optimization problems that could not be readily treated with other methods. I have actually done some research in this area (although ultimately it turned out that another method that is more specific to the problem at hand is usually much more powerful, in particular because it gives control of systematic errors, which evolutionary methods as a matter of principle don't).
Georg von Hippel | 04/05/09 | 14:01 PM
Everyone is excited about the King automation, though as always it got hyped up quite a bit with the talk of robots and such when it was really a computer doing something we know they can do with enough parameters; not exactly Skynet.
I think the big value for researchers will be in using the Adam Robots of the world to parse all the existing data and weed out already unsuccessful ideas that have been tried and ones that just can't be realized.
Not everyone gets Science here (though I assume a lot do through their schools) so they can't read both those articles. Maybe someone should do an article on what this means beyond the hype. Errrr, you?? :)