Researchers Develop AI System that Thinks Millions of Steps Ahead

Researchers at the California Institute of Technology have developed an artificially intelligent system that can think millions of steps ahead to solve complex math problems that have stumped mathematicians for decades. Led by mathematician Sergei Gukov, the team created a new machine-learning algorithm that tackled the Andrews–Curtis conjecture, a challenging math puzzle that involves finding long sequences of steps to solve.

The AI program, developed by Ali Shehper and his colleagues, was able to disprove potential counterexamples to the conjecture, providing new insights and bolstering confidence in the validity of the original problem. By using a reinforcement learning approach, the AI was trained to generate long sequences of unexpected moves, termed “super moves,” that surpassed human capabilities in problem-solving.

Gukov compared the math problems to solving a Rubik’s Cube, where the AI had to test out various moves to find the right path. The researchers believe that their methods could one day contribute to intelligent forecasting, such as predicting financial crashes, by teaching the AI to “learn to learn” and think outside the box.

Despite the seemingly abstract nature of their work, the researchers have made significant advancements in a decades-old area of math and have prioritized approaches that do not require large amounts of computing power. Their achievement adds to a growing body of research aimed at optimizing machine-learning algorithms to solve complex problems and benefit humanity.