Adam’s daily routine is fairly unremarkable. Like thousands of postdocs in labs across the world, he formulates hypotheses, designs experiments, runs tests, and — every once in a while — makes a discovery in his field. Only, Adam isn’t a postdoctoral researcher. He’s a robot.
More precisely, Adam is an automated scientist programmed by a team of researchers at Aberystwyth University and the University of Cambridge to carry out each step of the scientific process — from generating hypotheses to making conclusions — without any direct help from humans.
He spends his days and nights working in a laboratory, studying the yeast, Saccharomyces cerevisiae. Yeast is a model organism for living systems, and understanding the ways in which it functions provides insight into how human cells work. The robot investigates what are called orphan enzymes, which play a role in the life processes of the yeast and are encoded by genes not yet known to scientists. Adam seeks to identify these encoding genes.
“Adam is doing respectable scientific research,” says Ross King, one of the researchers who designed the robot. Using a form of artificial intelligence, Adam first examines a model of the life processes of the yeast and determines which enzymes are orphans. He then compares these orphans to similar enzymes in other organisms. And based on this comparison, he formulates an original hypothesis about which genes might encode for the orphans.
“This robot is very exciting,” says Bart Selman, a computer scientist and artificial intelligence expert at Cornell University. “It demonstrates active machine learning, where a robot actually decides what data to collect and what type of experiment to run.”
Adam carries out experiments in a completely automated laboratory. He grows strains of yeast with and without the genes he thinks encode for orphan enzymes. “If Adam finds a significant difference between these yeasts, then he considers his hypothesis confirmed,” says King. If, however, his hypothesis proves false, Adam — like any good scientist — starts again from scratch.
The task of uncovering which genes encode for which enzymes in baker’s yeast isn’t the most exciting work. But that’s partly the point. “Automation relieves people from having to do the more onerous parts of scientific research,” says King. “It takes away a lot of the benchwork.” Robots like Adam could potentially free scientists to pursue larger ideas.
Yet Adam offers scientists more than just relief from the daily drudgery of laboratory work. He provides them with a new way to understand and share their research. Each step of Adam’s experiments is recorded in a formalized logical language that can be carefully examined and easily replicated.
The researchers were able to reuse Adam’s experimental data in order to investigate other phenomena. “You would expect that when you remove enzymes from yeast, it would become less efficient because it evolved to have those enzymes for a reason,” King says. “But we found that in many cases the opposite was true.”
A freezer, a plate washer, incubators, air filters, other laboratory equipment, and a sophisticated AI system make up Adam the robot scientist.
Because of the meticulous documentation of Adam’s experimental procedures, the researchers were able to make this discovery without running any new tests. In fact, says King, many different discoveries can be made using the same sets of experimental data, provided that it is recorded with a precise logical structure.
Programming Adam to come up with scientifically meaningful hypotheses turned out to be surprisingly easy. Much of what makes the process difficult for human scientists is the amount of information they must consider. But for a robot driven by computer software, it’s just a question of processing power. “My personal belief is that robots will eventually be better than humans at formulating hypotheses,” King says.
“I was never very good in the lab,” King remembers. “I was always spilling and dropping things. And humans shouldn’t be doing the tasks that are ultimately done much better by machines.” Tasks that, if automated scientists like Adam continue to be developed, might sometimes include the scientific process itself.
Looking forward, Selman sees “other labs in other domains attempting to build these robots.”
But Adam is not designed to replace scientists. On the contrary, as the researchers are careful to point out, the idea is to develop a way of enabling teams of robot and human scientists to work together.
The Automation of Science
Science April 3, 2009
Originally published April 2, 2009