With the advent of artificial intelligence, developers are trying to shift on his shoulders as many routine tasks. Researchers at the University of California taught the algorithm to collect the most popular puzzle — Rubik's cube. Regardless of the complexity of the task, the AI to solve it takes no more than a couple of seconds.
Record holder for the Rubik's Cube (for 4.69 seconds)
Deepcubea uses two technologies for training: deep learning and the study of the so-called "archive of collected solutions". During the training of the neural network algorithm, variants of the solved problem were provided, after which the algorithm literally "unraveled" the solution in the reverse sequence. Training AI took only two days. After successful debugging and tuning of the algorithm, in 60.3% of cases, it found the most effective and fastest solution to the puzzle, setting a record in one second.
However, DeebCubeA has a competitor from researchers at the Massachusetts Institute of technology. Their brainchild min2phase collects Rubik's cube three times faster. But the key difference between the two is that min2phase is a fully human-written program designed specifically to solve the Rubik's cube. DeebCubeA "thinks" like a human being and is built on the basis of a neural network.
As for the success of man, the actual record for the Assembly of the Rubik's cube was set in 2017 by 15-year-old Patrick Pons. To solve the puzzle he has spent a total of 4.69 seconds.