In just over a day, a computer program at Cornell has extrapolated Newton’s laws of motions from a pendulum’s swings. The program starts with random combinations of addition, subtraction, multiplication, division and a few algebraic operators. Then, via a genetic algorithm, the program refines the formulae to pick out ones that fit the data better. It came up with the law of conservation of momentum and Newton’s second law of motion.
This raises some serious questions about the nature of science. It took physicists centuries to develop laws of nature to fit observed data; it took Hod Lipson’s program a day. We could worry about machines replacing human ingenuity. Or — a view I find more compelling — we see these algorithms as an aid to deriving the laws that explain other agglomerations of data, such as the genome or proteome. We don’t know the “laws of biology” in the way we know the laws of physics; and still less do we know the “laws of social science.” Maybe artificial intelligence is the best way to make progress on that.
Here’s Lipson giving a TED talk on robotics.