AI solves Rubik’s cube in under a second

Image: Amy T. Zielinski/Redferns via Getty Images

If civilization as we know it ends at the hands of artificial intelligence, at least we know those hands are really good at solving puzzles.

Researchers at the University of California, Irvine programmed a machine learning algorithm that can solve that most beguiling of human puzzles, the Rubik’s cube, in less than a second. 

That’s all well and good, but robots have been solving Rubik’s cubes for a while. What makes this one special is that it learned how to do it by itself, “without any specific domain knowledge or in-game coaching from humans,” according a press release from the university.

The algorithm, dubbed DeepCubeA, was the subject of a study published this week in Nature Machine Intelligence. Unlike some previous robot or AI efforts to quickly solve Rubik’s cubes, DeepCubeA had to learn from scratch. It passed the test with flying colors, solving 100 percent of the tests given to it and finding the fastest route to the end goal 60 percent of the time.

If you want to know what it looks like to see our future robot overlords solve a Rubik’s cube, this is what another research project from 2018 looked like.

To put it in a slightly different perspective, the UC Irvine researchers said their AI can solve a Rubik’s cube in about 20 moves. Humans who are good at solving the puzzle boxes, meanwhile, still take about 50 moves. 

Whether it’s a prelude to a horrible dystopia or a sign of a glorious tech future, DeepCubeA is definitely impressive at this moment. The press release noted that Rubik’s cubes can’t really be solved with random movements, so the AI had to learn and reason its way to success.

By comparison, there are plenty of humans who could stare at one of those things for hours without finding a solution. We might be doomed.