Robot scientist proves its worth

By Paul Rincon
BBC News Online science staff

It's no C-3PO, but the robot scientist is a star performer

Enlarge Image

UK researchers have built a "robot scientist" that performs just as well in the lab as its human counterparts.

The mechanical part of the robot scientist consists of an automated arm that squeezes out and mixes liquid on special plates for scientific analysis.

The process requires three computers to control and is driven by an intelligent software program that tries to explain observations in the experiments.

Details of the research are published in the scientific journal Nature.

The scientists set the robot the task of reconstructing the stages of a well-known metabolic pathway used by yeast.

The robot scientist had to observe the growth of yeast with gene deletions on a growth medium - a classic way of inferring metabolic pathways, the chains of biochemical reactions that occur in living things.

Top marks

Researchers "examining" the robot then compared its performance with that of nine computer scientists and biologists and found little difference between them.

Although existing scientific knowledge was used to programme the computer system with possible hypotheses, it can then use this basic knowledge base to generate new ones based on its observations.

If you opened it up I think humans would have the advantage. We tend to be more flexible Prof Ross King, University of Wales

The robot scientist designed the experiments and interpreted the data entirely on its own.

"It's a simple area of science. In that restricted world the computers compete well with scientists," co-author Professor Ross King of the University of Wales, Aberystwyth, told BBC News Online.

"But if you opened it up I think humans would have the advantage. We tend to be more flexible."

The researchers propose that a system like the robot scientist could be used in the field of genetic analysis, where, they claim, it would be faster than existing automated systems and more cost-effective.

"Another possible application is in drug design. [drug designers] test tens of thousands of different compounds to see whether just one will work.

"You could imagine that some system that uses some intelligence would easily learn from its experience," said Professor King.