Adam, the robot, at work in Aberystwyth University, UK. Picture by Jen Rowland

New Delhi, April 2: Scientists have unveiled artificial intelligence systems capable of making scientific discoveries, a leap forward in the quest to build machines that perform mental tasks long considered the exclusive preserve of humans.

A computer program developed by Cornell University scientists in the US has deduced Newton’s laws of motion and laws of energy conservation by observing swinging pendulums without prior knowledge of physics.

And a robot named Adam, built by scientists at the University of Cambridge and Aberystwyth University in the UK, has generated scientific hypotheses about yeast genes, tested them through experiments, and discovered new scientific knowledge.

Conventional robots perform myriad tasks — from assembling cars to assisting in surgery and vacuumcleaning carpets — in a slavelike fashion, following orders they are programmed to carry out. Scientists believe the new work, to be described in the journal Science on Friday, may herald an era where robots are assigned greater responsibilities in scientific research.

“The unique feature of Adam is its ability to generate new scientific ideas, think of experiments to test these ideas, actually do the experiments in the lab, and analyse the results,” Ross King, a computational biologist at Aberystwyth, told The Telegraph. “This is the first time a system has been combined with laboratory automation to discover knowledge new to science.”

Adam’s first task was to find hitherto unidentified genes on the genome of the yeast cell.

The computer program Cornell researchers Hod Lipson and Michael Schmidt designed was able to deduce within hours the energy and motion laws that physicists had taken centuries to formulate. “It built a model of natural laws from scratch without prior knowledge,” Schmidt said. “While this advance will not replace humans, it could greatly change how scientists investigate new phenomenon.”

The scientists say human imagination and intellect are still indispensable. Humans would need to set up experiments, assign observation tasks for the machines, and analyse and interpret the results.

“A computer will yield mathematical equations or laws... humans would still have to understand what those laws mean and put them in perspective,” Schmidt said. “We need humans to say, ‘Aha! This is energy conservation’.”

The scientists armed their systems with basic knowledge to analyse their observations. Adam had knowledge about yeast metabolism, software to generate hypotheses about candidate genes and to plan and design experiments, laboratory automation software to physically execute the experiment, and software to connect the observations to hypotheses. Cornell’s program used mathematics to correlate its observations of the pendulums and generate equations describing the behaviour of the motion.

Schmidt said such a program could be ideally used in areas such as biology or cosmology where observational data is abundant but there are large gaps in theory.

Human scientists for instance, Newton himself have typically combined previous knowledge, observations and their own imagination to deduce laws of nature. The artificial intelligence systems combine mere observations and computer programs to figure out equations or laws that fit the observations.