Rise of the Machine (Scientist)

Back in the day, scientists and engineers used to spend years analyzing data in a painful and tedious process of better understanding nature. Fast-forward to today, and computers can do it in one day. Correction, one computer at Cornell University can do it in a day. This computer is running a program developed by Hod Lipson and Michael Schmidt that incorporate genetic programming along with basic mathematics rules to extrapolate laws of nature. They have already derived the laws of motion for a pendulum in a day.

Scientists and engineers have been down this road before. Half a century ago, IBM’s Herbert Gelernter wrote a program that independently rediscovered Euclid’s rules of geometry, but some think that the solutions were contrived. Then in the 1970’s, Douglas Lenat’s wrote Automated Mathematician, but it didn’t gleam any new mathematical theorems that were useful. Stanford University also ran a project for two decades known as Dendral to extrapolate the structures of organic molecules from measurements by NASA spacecraft. It ended up generating more questions than answers.

Now Genetic Programming works by extracting invariant (constant) data from the database and deriving possible relationships (equations) between it and the rest of the data. The equations are tested again against the data, and the process is begun anew. It is very much an evolutionary program, in that it just plays with the data until it arrives at a reasonable conclusion.

While it is yet another step forward, these artificial intelligence programs are still decades away from replacing a competent scientist. It may never happen! We humans still have the advantage that we can daydream on the problem, and write up something completely new. True or not, Superstring Theory is one beast that I doubt a computer program could come up with. In any case, this is still another piece of the puzzle towards building a thinking/learning Terminator machine someday that will kill us all.