Credit: Sergei Gukov (This graphic included AI-generated art)
Caltech
Introduction
In recent years, mathematicians around the world have begun to use artificial intelligence in order to solve extremely difficult math problems. These math problems require thousands to millions of steps in order to solve them but at Caltech University, they are working on a program that can solve the problems quickly. The team is led by Caltech’s Sergei Gukov, the John D. MacArthur Professor of Theoretical Physics and Mathematics. He and his team hope to be able to use the program to solve million step problems. Gukov compares the math problems to an extremely complex rubik’s cube.
Why not other AI tools?
Mathematicians can not just use AI programs such as ChatGPT and Gemini because those tools are not very unique and are likely to give simple, expected solutions. Whereas with the AI program the researchers at Caltech are developing, it will focus on finding rare, unexpected solutions to the math problems. Additionally, these new artificial intelligence programs like the program at Caltech will find outliers or solutions that are very different from the others. “If you ask ChatGPT to write a letter, it will come up with something typical. It’s unlikely to come up with anything unique and highly original. It’s a good parrot,” Gukov says. “Our program is good at coming up with outliers.”
How are they creating it?
The mathematicians and researchers at Caltech are not using the approach of just feeding the tool information which could take a very long time. Instead, they started off by giving the program simple problems but gradually working its way up to more difficult problems. This technique is called machine learning. Machine learning is the process of computers learning to recognize patterns and make decisions based on data, without specifically being programmed for every task. It uses algorithms to improve its performance over time as it processes more data and experiences.
What impact does it have?
The team has been finding ways in order to make the tool use little computer power. The team said, “We try to demonstrate good performance on small-scale computers, easily accessible to a small academic collaboration, so that any of our colleagues around the globe can easily reproduce these results.” This means that the team wants to make sure almost anyone can use the program when it is completed so that it can help with many advanced problems. The AI program also demonstrates the potential of AI in discovering unexpected, outlier solutions, pushing the boundaries of “normal” problem-solving strategies.
By:Aarush Vajha