A New Algorithm Can Solve A Rubix Cube Faster Than A Person
Have you ever spent hours trying to solve a Rubix Cube? It’s nothing to be ashamed of. We’ve all been there. The Rubix Cube is no easy puzzle. Us humans have spent hours trying to crack the code. Meanwhile, scientists have developed an artificial intelligence program, by the name of DeepCubeA, that can solve this puzzle faster than any of us.
The Rubix Cube is a wildly popular game, leaving everyone puzzled since it was first made. We’ve all spent a significant amount of time trying to solve this cube. The Rubix Cube was made to be a puzzle, only solvable through a particular algorithm.
Contrary to convention, randomly switching sides of the cube won’t get you anywhere. Over time, some people have figured out how to crack this puzzle. While these people believe they’ve learned the technique or secret behind solving the cube, they are using a complex algorithm.
On average, people who know the right algorithm to solve this puzzle take approximately 50 moves to do so. Not to mention, it takes them hundreds of tries to get it right the first time. In fact, it’s entirely possible that only a minute number of people actually cracked the code, and the others just learned from copying them.
Scientists at the University of California, Irvine (UCI) have now created a deep reinforcement learning algorithm that can solve the Rubix Cube much faster than any human! Artificial intelligence in the past has been able to beat human players at some logic games, such as chess. However, no AI had been able to solve the Rubix Cube until the DeepCubeA.
Compared to our 50 moves, DeepCubeA only uses 20 moves! In fact, when given different versions of an unsolved Rubix Cube, this algorithm could solve the cube 100% of the time. What’s even more impressive is that approximately 60% of the time, it found the shortest path to solving the puzzle.
The algorithm’s ability to solve the cube is impressive. However, what’s even more astounding is how this program learned to solve the cube. Once the scientists made the code for DeepCubeA, they left the algorithm in isolation for two days, where it kept solving increasingly difficult versions of the cube.
Essentially, this piece of AI learned how to solve the Rubix cube all on its own. This is a remarkable achievement since the Rubix cube is no easy puzzle to solve. There are billions of possible paths, but only one solution. In addition, DeepCubeA can also solve other games that involve combinations, like the Rubix Cube, such as Lights Out and Sokoban.
The creation of DeepCubeA has many implications for the future of artificial intelligence. Up till now, artificial intelligence had been able to solve simple games involving a bit of critical thinking and strategy. However, the Rubix Cube requires much more abstract thinking and the use of complex mathematical models. DeepCubeA’s ability to solve this puzzle means that scientists are getting closer to creating deep learning machines that can think, reason, and make decisions just like humans do.
Written by The Original PC Doctor on 24/10/2019.