Researchers at Britain's University of Southampton’s School of Electronics and Computer Science (ECS) have won an award for producing an artificial intelligence that is able to guide scientific experimentation within a laboratory.
Doctoral students Chris Lovell and Gareth Jones were named winners of the Carl Smith Award for best student paper at the thirteenth international conference on Discovery Science, held in Canberra, Australia last week.
The artificial intelligence developed by Chris Lovell mimics the techniques used by successful human scientists. The software, called an "artificial experimenter," looks at the data available, builds hypotheses, and then chooses the experiments to perform, all without human interaction.
"Experimentation is expensive. Scientists always want to learn as much as they can from the smallest number of experiments possible. The new techniques we have developed try to address this problem,” said professor Klaus-Peter Zauner of the Engineering of Natural Systems Group at ECS, who supervised this research as part of a Microsoft European Fellowship.
As well as learning from small numbers of experiments, the software is designed to question whether the data obtained is correct.
"Biological experimentation can be error prone,” Zauner said. "Measurements taken may not always be representative of what actually happens. Our system tries to detect erroneous data, so it can ignore it."
The artificial experimenter has been used to characterize the response from a biological system. Currently these experiments have been performed manually in the laboratory, but the next step is to join the software with an automated platform that can perform microscale experiments to allow for fully autonomous experimentation.
The lab-on-chip platform being developed by Gareth Jones, will allow the cost of experimentation to be reduced further by decreasing the volumes of chemicals required per experiment. When completed, the platform will perform the experiments requested by the artificial experimenter, providing it with the results obtained to allow the software to develop new hypotheses and decide on the next experiments to perform.
The work has been carried out as part of a Microsoft European Fellowship awarded to Klaus-Peter Zauner, along with collaboration from Professor Steve Gunn and Professor Hywel Morgan.
A copy of the winning paper can be accessed here.