First Look

These physicists accidentally programmed AI to do their job

A team of Australian physicists wrote an artificial intelligence program to make their job easier. They didn't expect it would be able to take over the experiment.

  • close
    Australian researchers Paul Wigley (l.) and Michael Hush inadvertently designed an artificial intelligence program that could do one aspect of their job better than they could.
    Courtesy of Stuart Hay/ANU
    View Caption
  • About video ads
    View Caption

Physicists may soon program themselves out of a job by developing artificial intelligence.

A group of researchers recently developed an AI program to assist them in a complex procedure for an experiment involving finely optimized conditions. But rather than simply assist, the AI showed enough proficiency to run the experiment on its own and faster than humans or previous programs designed for the experiment.

"I didn't expect the machine could learn to do the experiment itself, from scratch, in under an hour," co-lead researcher Paul Wigley, a doctoral student at the Australian National University Research School of Physics and Engineering, said in a statement.

The physicists from the ANU, University of Adelaide, and the University of New South Wales Australian Defence Force Academy, were attempting to recreate an experiment that won the 2001 Nobel Prize – creating a Bose-Einstein condensate, a super chilled gas trapped in between laser beams.

Bose-Einstein condensates are able to reach temperatures so low that they are some of the coldest areas of the universe, in some cases less than a billionth of a degree above absolute zero, the temperature where all atoms stop moving. For comparison, the background temperature of outer space is minus 455 degrees Fahrenheit, roughly five degrees warmer than absolute zero.

The extreme cold temperature would make the Bose-Einstein condensates very effective as navigation systems, as they can make extremely precise measurements based on acute external changes. The problem is they require upkeep and optimization to keep them working. And setting them up is extremely complex.

That’s where the AI comes in.

“If we were to perform a brute force search and optimize the parameters to within a 10% accuracy of the parameters maximum-minimum bounds, the number of runs required would be 1016 ,” the researchers write in a study published Monday in the open access journal Scientific Reports.

Using the previous best online optimization algorithm, the Nelder-Mead algorithm, the Bose-Einstein condensates are able to be found much faster, in roughly 145 runs.

But with the AI program, it gets even faster: After it’s had time to learn the process, the new machine learning algorithm can find the right optimization in 10 experiments.

"A simple computer program would have taken longer than the age of the universe to run through all the combinations and work this out," Mr. Wigley said in the statement.

For the entire experiment, the scientists cooled a gas to roughly 1 microkelvin and trapped it between three lasers. They then handed over control of the experiment to the AI to cool the gas further.

The AI developed different techniques researchers and previous optimization algorithms had not thought of, including fluctuating the power in one laser up and down and compensating with another.

The scientists believes these processes could allow them to scale up the experiments and allow a larger than ever before Bose-Einstein condensate to be created.

On a practical level, the AI also has an additional advantage: convenience.

"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," co-lead researcher Dr Michael Hush from UNSW ADFA said in a statement. "It's cheaper than taking a physicist everywhere with you."

Make a Difference
Inspired? Here are some ways to make a difference on this issue.
FREE Newsletters
Get the Monitor stories you care about delivered to your inbox.

We want to hear, did we miss an angle we should have covered? Should we come back to this topic? Or just give us a rating for this story. We want to hear from you.




Save for later


Saved ( of items)

This item has been saved to read later from any device.
Access saved items through your user name at the top of the page.

View Saved Items


Failed to save

You reached the limit of 20 saved items.
Please visit following link to manage you saved items.

View Saved Items


Failed to save

You have already saved this item.

View Saved Items