Researchers create first robot scientist?
from Boston.com?

Researchers said yesterday that they have created the world's first robotic scientist, a system that can form theories, devise experiments, and then carry out the experiments almost entirely without human help.

The system, say its British creators, did just as well as biology graduate students in solving a problem in genetics, according to an article in today's issue of the journal Nature. Although the system uses robotic equipment common in modern laboratories, this is the first time that a machine has carried out so many of the roles traditionally done by scientists.

Some scientists questioned whether the system, dubbed the "Robot Scientist" by its creators, deserved the title of scientist. For human scientists, some of the most interesting discoveries happen when researchers notice something they weren't looking for and suddenly change course, said Stuart Schreiber, a Harvard professor who is one of the intellectual leaders in automating aspects of modern biology. And breakthroughs often come when a scientist discards a basic assumption of the experiment. The Robot Scientist, on the other hand, is bound by the set of rules programmed into it.

"When Deep Blue recognizes a seemingly incomprehensible move by Kasparov, it doesn't have to consider that the rules of chess are incorrect," said Schreiber, comparing the system to IBM's chess-playing computer.

I, for one, welcome our new robot overlords.

More

Researchers create first robot scientist

By Gareth Cook, Globe Staff, 1/15/2004

Researchers said yesterday that they have created the world's first robotic scientist, a system that can form theories, devise experiments, and then carry out the experiments almost entirely without human help.

The system, say its British creators, did just as well as biology graduate students in solving a problem in genetics, according to an article in today's issue of the journal Nature. Although the system uses robotic equipment common in modern laboratories, this is the first time that a machine has carried out so many of the roles traditionally done by scientists.

With technology becoming increasingly sophisticated, the announcement adds to a sprawling debate seen in fields as diverse as drug discovery and the exploration of Mars about what role humans will play in the future of scientific discovery.

In biology, as in many sciences, automation has given scientists powerful tools to find masses of information, but not the tools to make sense of it all, a problem the new system is aimed at helping to solve.

"In a number of areas scientific data is being generated at enormous rates, creating the need for the automated analysis of the data," said Ross D. King, the system's co-inventor and a professor at the University of Wales, Aberystwyth.

Though a robotic scientist may evoke images of a gleaming steel humanoid, clad perhaps in a starched white lab coat, the new invention is instead a modest collection of off-white boxes connected by wires in King's Welsh laboratory.

One computer acts as the brain, formulating experiments and interpreting the results, while other pieces of equipment carry out the experiments, drawing on a large library of materials.

Some scientists questioned whether the system, dubbed the "Robot Scientist" by its creators, deserved the title of scientist. For human scientists, some of the most interesting discoveries happen when researchers notice something they weren't looking for and suddenly change course, said Stuart Schreiber, a Harvard professor who is one of the intellectual leaders in automating aspects of modern biology. And breakthroughs often come when a scientist discards a basic assumption of the experiment. The Robot Scientist, on the other hand, is bound by the set of rules programmed into it.

"When Deep Blue recognizes a seemingly incomprehensible move by Kasparov, it doesn't have to consider that the rules of chess are incorrect," said Schreiber, comparing the system to IBM's chess-playing computer.

The Robot Scientist works in an area of biology known as functional genomics, which is concerned with uncovering the roles that different genes play in the machinery of life. As a test, the system was told to discover how certain genes affect a complex chemical pathway inside yeast cells. The task for the computer, and a common one in biology, was to figure out which genes are involved in which steps of the pathway by testing yeast cells with different genes removed.

The mind of the Robot Scientist is a piece of software, created by King with his colleagues in Manchester, London, and Aberdeen, which forms a hypothesis about which gene is involved in what step of the pathway and then devises an experiment to test the hypothesis.

This computer then sends these commands to a piece of robotic lab equipment, which can select all the appropriate ingredients, including a yeast cell with the appropriate gene removed. The robotic lab equipment can then observe the outcome of the experiment -- whether the yeast cell grows successfully -- and feed the information back to the Robot Scientist's main software, which decides whether the experiment vindicates the hypothesis and then selects a new experiment to learn more.

The Robot Scientist was able to determine the functions of the genes quickly and accurately, according to the paper. When a group of graduate students were asked to choose and design experiments with the same aim, only the best performers did as well as the robot, said King. Scientists who have seen the paper said that the system did not represent a major advance in either robotics or artificial intelligence, but rather a milestone in combining the two.

"I don't think it is a surprise that this is possible," said Gad Shaulsky, a Baylor College of Medicine genetics professor who was not involved in the research. "What is really wonderful about this is that they did it."

Pat Langley, a specialist in the use of artificial intelligence for scientific discovery at Stanford University, said that credit for the first robot scientist should go to a more modest experiment run by another researcher in 1990. That work is cited in today's paper, but King said that system used only very limited amounts of background knowledge, which restricted it to much simpler lab work, and was not able to consider multiple hypotheses -- a hallmark of the scientific process. Robots are already doing the work of scientists in a wide array of areas. Since the Apollo program, most space exploration has been done by robotic satellites that report back to human controllers. In biology, gene sequencing that once had to be done manually is now done by machines. And some mathematicians have even turned to computers to help them prove theorems, a development that some view as even more humiliating than a computer program beating a chess grand master.

The Robot Scientist isn't likely to be used in laboratories unless it becomes more sophisticated and cheaper to build, Shaulsky and others said, but it could be of interest to pharmaceutical companies, which do large amounts of automated research in their search for new drugs. Shaulsky said that he already uses an artificial intelligence software program, called "GenePath," to help design genetics experiments.

In the work reported in Nature, King and his colleagues used the Robot Scientist on genes whose function is already known, so they could be sure of their results. Now, though, they plan on using it to look at yeast genes that are not understood -- about 30 percent of the total. If that works, and the system discovers genuinely new information, it will earn the Robot Scientist a new level of respect from its warm-blooded colleagues, King predicted.

Gareth Cook can be reached at cook@globe.com.
© Copyright 2004 Globe Newspaper Company.

posted by prof_booty | 12:24 PM