Slaves No More, `Smart' Robots Invade The Lab

They're fast, reliable, and tireless. Now they want a place at the bench. Can machines adopt the scientific method? PITTSBURGH--Repetitive lab work makes John Lindsey's mind wander. After a week or so of compound making - repetitive lab work at its worst - the Carnegie Mellon University chemist begins to think about automation. Robots. In particular, robots that can do his job. Lindsey has spent five years pursuing the construction of just such a machine - a compact and reliable robot that can perform the repetitive measure-mix-stir-analyze cycle of synthetic chemistry in the lab. "Chemists are still making compounds the way they did 100 years ago," he complains. "Once the chemist has a compound, of course, he can analyze it with a broad array of [automated] tools, but getting there is the problem. I wanted to push automation further up into the process." With help from Carnegie Mellon artificial intelligence expert Tom Mitchell, Lindsey is approaching his goal. New trials with a second-generation prototype at Carnegie Mellon suggest that laboratory robots, the long-promised panacea for scientists burdened with experimental chores, may finally be ready to take their place at the bench. Robots of various degrees of complexity have been available for decades. But the exorbitant cost, substantial programming requirements, and questionable reliability of the machines have kept them from becoming commonplace in the lab. The falling cost of personal computers has helped. And the payoffs are nothing to be sneezed at. At the California Institute of Technology, for example, researchers in molecular biologist Leroy Hood's lab are demonstrating 15-fold improvements in gene-sequencing efficiency using modified commercial robots. Scientists at Caltech have aimed their robots at high-volume gene research labs. Their success is due in large part to a new generation of specialized commercial robots, known as "automated workstations." Combined with new sequencing technologies, the machines are emerging as the key to the Human Genome Project - the 15-year, $3 billion effort to map and sequence the human DNA molecule. More than two dozen U.S. labs are now using a $30,000 robot made by Beckman Instruments for gene sequencing, using techniques developed at Caltech. "A lot of the tasks in sequencing the human genome are going to be repetitive," says Deborah Nickerson, a member of Hood's team. "When you're talking 3.5 billion bases [the building blocks of the human genome], you're going to need automation." Instead of hands, the Beckman robot's arms end in pipettes, either alone or in clusters of six. With another, more conventional, robot supplying trays of chemicals, the robots can run hundreds of reactions simultaneously, without human intervention. The Caltech team has even modified such commercial robots to do polymerase chain reaction (PCR) testing, a new DNA synthesis process that requires rapid temperature changes from 25øC to 100øC. "It took a while to develop a machine where the computer chips wouldn't die under that stress," says Nickerson. A Caltech robot is already doing PCR-based diagnosis of 200 people a day, looking for genetic mutations like sickle cell anemia and cystic fibrosis. The Carnegie Mellon researchers have a different purpose in mind for their robots. They want to prove that robots can serve as a sort of "right-hand man" to scientists doing basic research. And they have already enjoyed some success. A modified commercial robot, starting with little more than a batch of chemicals and the instructions to find the best "recipe" for a certain compound, worked nonstop for 27 hours until it arrived at the most efficient solution. "In principle, you could attempt the same kind of thing yourself, but you probably wouldn't," says Mitchell. Done by hand, the experiment might take weeks. Mitchell has worked on the project since 1986, when Lindsey asked for help in "smartening" his prototypes. But Mitchell decided to go further. After analyzing a year's worth of Lindsey's logbooks, he found that many of the decisions the chemist had used to plan his next experiments were based on his experience in spotting profitable areas of inquiry and promising trends - two skills that AI-driven "expert systems" had already shown some success in duplicating. He suggested using advanced AI methods to push robots to the point of making many of those judgment calls. The goal is to link up robots and AI into "intelligent laboratories." The AI-aided robots will conduct and monitor experiments, interpret results, refine hypotheses, and plan experiments - "in short, most of the scientific method," Mitchell says. A further advantage of computer-controlled experimenting is that robots keep perfect records of their actions and measurements. "If there is a machine watching all the to-ing and fro-ing of the experiment, nothing is lost. You can benefit from accidental experiments because 'somebody' was there to observe it," says Carnegie Mellon robot researcher Melvin Siegal. Lindsey and Mitchell use a West German commercial robot marketed in the U.S. by the Techno Co. Attached to the robot's gantry-mounted "arm" is a mechanical syringe, which can move over a tray of 98 vials and a rack of reagents. During the reaction, the robot can pick up any sample and analyze it with an adjacent thin-layer chromatograph or an absorption spectrometer. Although the analytical instruments used in the Carnegie Mellon trial cannot handle reactions involving precipitates, the scientists are working to include that capability. Ten thousand lines of computer code control the process, but none of it so far fits within the classic definition of AI. The routines to find the best recipe for a particular compound, for example, use a traditional "hill climbing" algorithm, rather than exotic expert-system techniques. The future, however, clearly belongs to AI. "We're trying to automate the process of forming theories," says Mitchell. And that will take more analysis on how scientists think, as well as powerful computer code to simulate it. The recent advances in laboratory robotics raise intriguing questions about the relationship between scientists and automation. Why, many wonder, are research labs - presumably a bastion of technophilia - still years behind industry in the acceptance of robotics? Two important reasons are cost and the nature of scientific inquiry. "First of all, robots are expensive," explains Takeo Kanade, codirector of the Carnegie Mellon's Robotics Institute. Only recently has it been possible to outfit a useful robotic workstation for less than $50,000, and a duplicate of Lindsey and Mitchell's machine would cost more than $120,000. Although prices are falling as electronics get cheaper, specialized laboratory machines are still a niche market, and are likely to remain more expensive than their industrial brethren. Then there's what Kanade calls "the psychological hurdle." Spotting scientific phenomena and trends is a critical part of the scientific process and, as such, the rightful domain of the scientist, the argument goes. Handing that task to a robot turns the researcher into a mere programmer. But the widespread use of computation has gradually made the move to computerized automation more palatable, Kanade says. Meanwhile, slick new "user interfaces," with pull-down menus, graphics, and rudimentary intelligence, have made robot programming relatively painless. Will robots ever be able to plan and execute experiments unaided? Mitchell is optimistic, but his collaborator has reservations. "I know how much [of a role] serendipity plays in discovery," says Lindsey. "For the time being, I'm happy to have robotics remain a productivity tool."

Slaves No More, `Smart' Robots Invade The Lab

Interested in reading more?

Become a Member of