"Robot Scientist" Said to Equal Humans at Some Tasks

John Roach
for National Geographic News
January 14, 2004
It looks nothing like C-3PO of Star Wars fame, but a team of British scientists have created a "robot" that can formulate hypotheses, design experiments, and interpret results on par with the best of their human counterparts.

Called the "robot scientist," the technological marvel proved its intellectual prowess by correctly determining the function of specific genes within yeast.

"There was no difference between the robot and the best humans on this task," said computer biologists Ross King at the University of Wales in Aberystwyth and Stephen Oliver at the University of Manchester in an e-mail interview.

King and Oliver describe the apparatus as rather boring in appearance—a liquid-handling machine networked to several computers—but packed with sophisticated robotics and advanced artificial intelligence (AI) software.

Together with colleagues at the University of Wales, University of Manchester, Robert Gordon University in Aberdeen, and Imperial College in London, the researchers report on the robot in the January 15 issue of the science journal Nature.

According to the paper, the robot provides important insight to the scientific process, a process in which automation is both inevitable and desirable.

"It is inevitable because it will be required to deal with the challenges of science in the 21st century. It is also desirable because it frees scientists to make the high-level creative leaps at which they excel," the researchers conclude in Nature.

James Collins, a biomechanical engineer at Boston University in Massachusetts, said he would be surprised if the robot scientist replaced humans, but that it would certainly be useful in sorting through the mountains of data in areas such as gene research.

"It's a great demonstration of how data analysis and experimental techniques can be automated and integrated to address large-scale exploratory science projects," he said. "But we are a long ways away from using these types of approaches to replace scientists."

Yeast Genes

The robot scientist was tasked to figure out the function of different genes in baker's yeast (Saccharomyces cerevisiae) using so-called "knockout" strains of the yeast that have one gene removed.

By observing how the knockout strains grow, or don't grow, it is possible to devise various theories for the function of the knocked-out gene. The researchers liken the process to trying to figure out how different parts of a car work by removing them one at a time.

The robot scientist, equipped with a wealth of information about biochemistry and sophisticated AI software, watches the yeast grow, generates a set of hypotheses concerning the function of the gene in question, and then plans an experiment that will eliminate as many of the hypotheses as quickly and cheaply as possible.

The robot then conducts experiments by dispensing and mixing liquids using one machine and measures the growth of yeast with a second machine that feeds the results back into the system. It then evaluates the results against the set of hypotheses, generates new hypotheses, and the process starts again.

The researchers say this is the same type of scientific process that humans use to understand the world.

In fact, King and Oliver said the robot performed as well as a group of graduate students and staff tasked with the same problem and "was both significantly cheaper and faster than just choosing random experiments or the cheapest experiments."

New Knowledge?

In the baker's yeast test, the robot scientist proved it could figure out the function of specific genes. The researchers know the robot's results were correct because they had already independently gotten the same answers.

"Our main focus now is trying to demonstrate that we can generate new scientific knowledge," said King and Oliver.

For example, about 30 percent of the 6,000 genes in baker's yeast are unknown but many are thought to be common to the human genome. If the robot scientist can figure out the function of those genes, the results could prove medically important.

If the robot scientist succeeds, the researchers said any new knowledge would need to be treated as provisional until independently verified, as is protocol with the scientific process. "There is no reason to treat the robot's results any differently," they said.

Regardless of how successful the robot scientist becomes, Collins said human imagination and creativity would continue to play an important role in the scientific process.

"There is still a significant role for imagination and creativity that is overlaid, coupled with, the scientific knowledge in front of you for coming up with a non-obvious hypothesis," he said.

© 1996-2008 National Geographic Society. All rights reserved.