On July 14, 2019, the 10th Annual World Cube Association (WCA) Championship in Melbourne crowned a new Rubik's Cube champion. This year's winner was Philipp Weyer of Germany who solved the 3 x 3 x 3 Rubik's Cube in 6.74 seconds.
Computer scientists and mathematicians at the University of California Irvine (UCI) have programmed an artificial intelligence(AI) called DeepCubeA which can find solve a Rubik's Cube in a second without any domain knowledge or in-game coaching from humans. The study was published in Nature Machine Intelligence on July 15, 2019.
In the study, the DeepCubeA algorithm solved 100% of all test configurations and found the shortest path to the goal - all six sides displaying a solid color - about 60% of the time. Researchers say the algorithm also works on other combinatorial games such as the sliding tile puzzle, Lights Out and Sokoban.
"This work presents an AI system that can automatically learn how to solve the iconic Rubik's Cube and other problems characterized by a vast number of possibilities and a very small number of solutions, and where random moves are unlikely to lead to a solution," says Pierre Baldi, Distinguished Professor of Computer Science at UCI.
According to Baldi, the solution to the Rubik’s Cube involves more symbolic, mathematical and abstract thinking and so a deep learning machine which can crack a puzzle like the Rubik's Cube is getting closer to becoming a system that can think, reason, plan and make decisions.
"These characteristics are shared by many other problems in robotics and other domains that require some kind of planning," added Baldi. "Imagine a robot tasked with cleaning up your kitchen: there is an astronomical number of sequences of moves, but only very few lead to a clean kitchen. And randomly moving dirty dishes around is not going to do it."
"More broadly, this work is part of a general effort to bridge machine learning AI and symbolic AI to address complex problems that humans solve through planning and reasoning," added Baldi.
In the study, researchers wanted to understand how and why the AI made its moves and how long it took to perfect its method. To do this, they started with a computer simulation of a completed puzzle and then scrambled the cube. After the code was running, DeepCubeA trained in isolation for two days, solving an increasingly difficult series of combinations, and during that time, it began to learn on its own.
Baldi says that some people can solve the Rubik's Cube in about 50 moves, but the AI in DeepCubeA can solve the cube in 20 moves with the minimum number of steps. This, says Baldi, proves the strategy is different, and that the AI's form of reasoning is different than a human’s reasoning.
