Researchers have taught AI to solve a Rubik's Cube in just 1.2 seconds
Science is a powerful thing. It’s taken mankind past the earth’s atmosphere and onto the surface of the moon, and it has also brought us numerous advances in the fields of medicine, food, and energy. However, as impressive as those many feats are, it’s often the less-important advances that make people stop and marvel at how far tech has come.
For example, you may already know that engineers have designed robots that can solve Rubik’s Cube puzzles faster than any human. In March of 2018, Ben Katz and Jared Di Carlo constructed a device that completed a Rubik’s Cube in just .38 seconds.
Kats and Di Carlo’s Rubik’s Cube-solving machine.
While the latest major advancement in Rubik’s Cube-solving tech hasn’t led to a faster solve time than Katz and Di Carlo’s, it is arguably more interesting for other reasons. a new research paper describes “DeepCubeA,” a “deep reinforcement learning” system that uses artificial intelligence to solve Rubik’s Cubes.
The researchers behind DeepCubeA, through plenty of practice and training, managed to teach their AI to solve a Cube in a mere 1.2 seconds. As we said, that isn’t quite as speedy as previous world records, but it’s certainly faster than most humans could complete the puzzle. On average, it takes the best Rubik’s Cube pros around 50 moves to finish, whereas DeepCubeA finished in a mere 28 moves (after being fed roughly 10 billion different puzzle combinations).
The researchers behind DeepCubeA, through plenty of practice and training, managed to teach their AI to solve a Cube in a mere 1.2 seconds.
That’s where the key difference lies between Katz’s machine and DeepCubeA: the former performs as many moves as it needs, but it simply does them quickly. The latter uses real learning and experience to improve its speeds, which reduces both puzzle solve time and the number of moves needed for completion.
We sadly do not have any videos to offer that demonstrate the feat, but you can read the full DeepCubeA research paper for yourself over on Nature, or watch Katz and Di Carlo’s impressive machine in action above.
Image credit: Shutterstock