Today in my Twitter feed I was pointed to this article (thanks Errol!) about how computers in telescopes are doing most of the grunt work for astronomers. The story leads with the idea that, “Astronomy could be the first discipline in which the rate of discovery by machines outpaces humans’ ability to interpret it.” That is, the machines are coming up with data in vast quantities, all of which needs to be looked at and interpreted. That’s quite impressive. But the thing that caught my eye was this:
Last year, Hod Lipson and pals at Cornell University developed a genetic algorithm capable of sifting through data looking for the laws of physics behind it.
And it seems to work. These guys generated a load of data by tracking the motion of things like simple harmonic oscillators and chaotic double-pendulums. They then set their algorithm loose on the raw data–not the manicured stuff but the warts’n'all measurements.
Their jaw-dropping result is that their algorithm derived Newton’s laws of motion from this data, without outside help.
Now that is one smart algorithm. But the article goes on to say that the software has come up with other relationships as well, some of which are not known physical laws. They could be false positives, but they could almost be brand new scientific discoveries.
Smart things, these computers.