In recent decades, robots have replaced millions of manual laborers; now they're moving in on scientists, too. A fully automated robotic laboratory can design its own molecular biology experiments and has even made its first discoveries, a multidisciplinary team reports this week. Meanwhile, a team of computer scientists has developed a robot that can independently come up with the “laws of motion” for a dynamical system such as interconnected pendulums.
Robots are doing ever more of the physical labor in laboratories--from analyzing DNA samples to handling data tapes from massive particle-physics experiments. And scientists increasingly rely on computers to analyze their data. But the highest-level thinking--the formulation of hypotheses and designing of experiments to test them--has remained the preserve of humans.
That's starting to change with the efforts of computer scientist Ross King of Aberystwyth University along with systems biologists at the University of Cambridge, U.K., who have developed a robot named Adam to identify genes involved in yeast metabolism. Adam doesn't look so much like an android as a huge box of Rube Goldberg–type equipment. But it does far more than just analyze cells.
Using algorithms programmed by scientists, Adam formulates hypotheses about the origins of "orphan enzymes": enzymes for which scientists have been unable to identify the encoding genes. The robot then plans and executes experiments to test its hypotheses--selecting yeast mutants from a collection, incubating cells, and measuring their growth rates. As King's team reports this week in Science, Adam came up with 20 hypotheses about genes encoding 13 enzymes, 12 of which it confirmed.
The second paper, also in Science, reports a similar feat in physics. Cornell University computer scientists Michael Schmidt and Hod Lipson devised an algorithm that will deduce laws about the motion of a nonlinear dynamical system--for example, a pendulum suspended from the end of another pendulum. The dynamics of such a system can be captured in a mathematical function called a Hamiltonian, which is essentially an expression of the system's energy. Schmidt and Lipson's robot can deduce the Hamiltonian and other key mathematical quantities for a system by observing its motion.
Both papers show how to "automate parts of the science enterprise that haven't been automated much before," says computer scientist David Waltz of Columbia University. In both cases, he points out, the system not only will generate a hypothesis but then can go on to test it and revise it as a result of the testing. "There's an interesting message from this," says Waltz--and perhaps a daunting one for some scientists. In the future, he says, scientists, in order to carry out their work, "might have to learn how to program computers and express knowledge about the world the way people in artificial intelligence have done."
At the moment, neither robot is likely to bag a Nobel Prize. Schmidt and Lipson's system only "discovers" Hamiltonians and other functions that a sharp graduate student might quickly figure out. Adam has traced the previously unknown origins of several enzymes but hasn't made a conceptual breakthrough. "Qualitatively, humans are still winning hands down because of the amount of knowledge any human scientist brings to a question," says computer scientist Bruce Buchanan of the University of Pittsburgh in Pennsylvania. But he predicts that as programs continue to improve, automated systems may come up with discoveries humans would never have imagined.