Can AI Help Us With Our Unknown Unknowns?

When tackling with cutting edge science, it’s not always easy to know what it is that we, as humans, lack understanding of.  It’s what Donald Rumsfeld would call the ‘unknown unknowns’.

To try and overcome this, a team from the University of Vienna have developed an algorithm that can propose new quantum physics experiments itself, with the hope being that it suggests ideas that humans would never have thought of themselves.

Automated Experiments

The researchers, after struggling to generate so called Greenberger-Horne-Zeilinger (GHZ) states themselves, began formulating an algorithm that might do a better job.

They developed an application, called MELVIN, that takes the common building blocks of quantum experiments and then arranges these to propose unique configurations.  It then plays around with these and once it finds one that meets the goals set of it, MELVIN attempts to simplify the design and report back to the researchers.

“I started the program in the evening and by the next morning, after a few hundred thousand different trials, it found one correct solution,” the developers say. “You can imagine that was a pretty exciting day.”

The program has already proved itself capable of devising interesting approaches in a variety of fields.  For instance, the researchers used MELVIN to devise novel ways of switching the properties of entangled particles, which could be useful in areas such as quantum cryptography.

Smarter Research

MELVIN begins its work by randomly shuffling components, but it uses machine learning to modify its approach based upon the experience it gains.

“That means if it found one good solution, it stores the good solution and can use it for follow-up experiments. This improves its speed significantly, by more than one order of magnitude,” the team say.

Now, suffice to say the use of MELVIN is likely to raise as many questions as it provides answers, and as such it will remain very much a work in progress for the time being.

It does nonetheless provide an interesting approach to the division of novel research methods that allow us to take research in unexpected directions.  As such, it seems certainly worth persevering with and seeing where it can take us next.