Describing the basic laws of physics occupied Sir Isaac Newton for decades. In April scientists unveiled a computer program that can analyze data and independently derive those fundamental physical laws within a matter of hours. This program could relieve major logjams in scientific research. Modern instruments like space observatories, particle colliders, and gene chips produce vast amounts of data, and mining that data is a slow, laborious process. Smart software—a synthetic scientist, in essence—could greatly speed it up.
Cornell University roboticist Hod Lipson and his Ph.D. student Michael Schmidt developed their system to analyze data from the kinds of mechanics experiments that college students encounter in introductory physics courses: observing the motion of a swinging pendulum or of two weights bouncing on connected springs, for instance. An automatic camera fed data directly to their computer program, which then tried millions of mathematical expressions to identify which ones held true from one experiment to the next. Using an evolutionary algorithm, the program randomly varied the winning equations to match the data more closely. In this way it “discovered” a handful of natural laws, including conservation of energy and momentum. Complex experiments required as much as 40 hours, simple ones as little as 10 minutes.
The Cornell program “won’t replace scientists anytime soon,” Schmidt says. “But it will let them look in a more efficient way at what might be interesting.” Gene chips, for instance, can measure the expression of thousands of genes at a time, but the important question is how one gene regulates others within that incredibly complex web of relationships. Smart software could rapidly flag interesting patterns—such as the way that levels of one protein depend on six others—so that researchers could then follow up with targeted experiments.douglas fox
