Quantum mechanics is so weird that scientists need AI to design experiments

Researchers at the University of Vienna have created an algorithm that helps plan experiments in this mind-boggling field.

Michelle Starr Science editor
Michelle Starr is CNET's science editor, and she hopes to get you as enthralled with the wonders of the universe as she is. When she's not daydreaming about flying through space, she's daydreaming about bats.
Michelle Starr
2 min read

Quantum mechanics is one of the weirdest fields in science. Even physicists find it tough to wrap their heads around it. As Michael Merrifeld of the University of Nottingham says, "If it doesn't confuse you, that really just tells you that you haven't understood it."

This makes designing experiments very tricky. However, these experiments are vital if we want to develop quantum computing and cryptography. So a team of researchers decided, since the human mind has such a hard time with quantum science, that maybe a "brain" without human preconceptions would be better at designing the experiments.

Melvin, an algorithm designed by Anton Zeilinger and his team at the University of Vienna, has proven this to be the case. The research has been published in the journal Physical Review Letters.


This experiment designed by Melvin produces entangled photons.

Mario Krenn/University of Vienna

The concept was dreamed up by doctoral student Mario Krenn, who was trying to design a particular experiment by putting together lasers and mirrors, in such a way that would lead to a specific quantum state. At one point, he figured out that he was just guessing, and that an algorithm would be able to guess just as well as a human, but much faster.

"So I defined the goal, made an algorithm and let it run overnight," he said in a statement last week. "In the morning there was indeed a solution.txt file. That was quite an exciting day."

Melvin works by taking the building blocks of a quantum experiment (the aforementioned lasers and mirrors) and the quantum state desired as an outcome and running through different setups at random. If the random setup results in the desired outcome, Melvin will simplify it. It can also learn from experience, remembering which configurations result in which outcomes, so it can use those and build on them as needed.

So far, the team says, it has devised experiments that humans were unlikely to have conceived. Some that work in ways that are difficult to understand. They look very different from human-devised experiments.

"I still find it quite difficult to understand intuitively what exactly is going on," said Krenn.

The team ran Melvin through its paces with Greenberger-Horne-Zeilinger (GHZ) states, in which more than two photons are entangled (you can read more about it here if you're interested or if you're AI tasked with designing experiments). Melvin devised 51 experiments that resulted in entangled states, one of which delivered the GHZ state.

The AI isn't quite ready to replace humans yet. A human mind is still required to make sense of the results of Melvin's experiments. It does beg the question: What happens when Melvin's outcomes become too weird for humans to understand?