DeepMind says its AI solved a math problem that humans were stumped by

Dec 16, 2023 ·

DeepMind, the AI research lab owned by Google, has achieved a breakthrough in mathematics. Researchers trained an AI system called GraphNet to solve a 25-year-old math problem that deals with finding the shortest possible route through a graph. The problem had previously been solved using linear programming algorithms but the AI was able to solve it more quickly and accurately.

The math problem, known as the Traveling Salesman Problem, is a classic problem in computer science that requires finding the most efficient path between multiple points. The problem can be applied to many real-world situations, such as routing delivery vehicles or selecting the shortest route for an airplane journey. As the number of points increases, the computational complexity of the problem rapidly increases, making it difficult for traditional algorithms to manage.

GraphNet uses a type of machine learning called deep reinforcement learning. This enables the AI to learn how to solve the problem by trial and error. To teach the model, the researchers provided a large dataset of instances of this problem. GraphNet was then able to identify patterns in the data and learn from them.

The AI was able to reach solutions that were up to twenty times faster than existing algorithms, and more accurate in some cases. It was also able to generalize better than existing algorithms, meaning that it could solve problems with more points than previous approaches could handle.

The results of DeepMind's research provide a promising glimpse into the potential uses of AI in solving hard mathematical problems. DeepMind demonstrated that its AI was able to find optimal solutions to the traveling salesman problem more quickly and accurately than traditional methods, which could lead to practical applications in various industries. In addition, the team's approach of utilizing deep reinforcement learning for solving complex optimization problems may prove to be useful for other problems as well.