Artificial Intelligence Learns To Run Nobel-Winning Physics Experiment


Artificial intelligence systems are getting better at performing jobs that were once thought to be exclusive only to humans.

Researchers have created sophisticated AI systems that can write political speeches and even accurately predict who will win the Kentucky Derby.

Now, scientists are placing themselves at risk of losing their own job by using AI to run a complex science experiment.

A pair of researchers has taught an AI to replicate a physics experiment that won for three physicists the 2001 Nobel Prize. The experiment involves trapping the Bose-Einstein condensate (BEC), a dilute gas of bosons cooled to extremely cold temperatures in lasers.

Bose-Einstein condensates are extremely sensitive to fluctuations in energy, which makes them useful in experiments and measurements.

Sensors that are powered by these trapped gases can potentially detect tiny changes in the Earth's magnetic field or gravity, so researchers work on ways to tap them as sensors for mineral exploration and navigational systems.

Paul Wigley, from the Australian National University, and Michael Hush, from the University of New South Wales, developed an AI system that they trained to manage gas-trapping laser device.

They cooled gas to 1 microkelvin then handed over the control of the laser system to the AI, which cooled the gas to nanokelvin by applying different techniques to manipulate lasers the physicists said humans would not have likely guessed.

"Through repeated machine-controlled scientific experimentation and observations our 'learner' discovers an optimal evaporation ramp for BEC production," the researchers reported in the journal Scientific Reports on May 16.

The researchers acknowledged the potentials of using artificial intelligence in making experiments bigger and better. The AI, which learned to conduct the experiment from scratch in less than an hour, may be capable of figuring out complex ways humans have not thought of to make measurements more accurate and get experiments colder.

The AI system's ability to create condensates faster and in greater quantities under more conditions could also make it very useful in field measurements.

"It's cheaper than taking a physicist everywhere with you," Hush said.

"You could make a working device to measure gravity that you could take in the back of a car, and the artificial intelligence would recalibrate and fix itself no matter what."

The researchers said that they plan to use the AI to build larger condensates faster than before.

ⓒ 2018 All rights reserved. Do not reproduce without permission.
Real Time Analytics