Loh Appointed to Advisory Committee of $10M AIMO Prize

Challenge fund by XTX Markets aims to spur innovation in mathematical sciences

Carnegie Mellon University Mathematical Science Professor Po-Shen Loh has been using math competitions to teach creative problem-solving to students for most of his career. Now, he's advising a high-stakes competition for artificial intelligence models.

XTX Markets launched the Artificial Intelligence Mathematical Olympiad Prize (AIMO) in November 2023. The $10 million challenge fund is designed to spur the creation of a publicly shared AI model capable of winning a gold medal in the International Mathematical Olympiad (IMO), the premier math competition for high school students worldwide.

A social entrepreneur and evangelist for math, creativity and education, Loh delivers lectures around the world.

"In over 100 talks, I have interactively queried ChatGPT on math topics that are freely proposed by broad audiences," Loh said, referring to a system developed by OpenAI that allows users to generate content, answer questions and provide detailed explanations.

"Week after week, I have observed that even the current version of ChatGPT usually responds to math discourse at the level of a strong high school student, across a wide variety of topics. But it also sometimes stubbornly makes mistakes, and then sticks with those mistakes in ways that people would not," he said.

In February, XTX announced that Loh and other prominent mathematicians along with AI and machine learning specialists will serve as an advisory committee to support the development of the AIMO Prize including advising on appropriate protocols and technical aspects and designing the various competitions and prizes.

Along with Loh, the group includes Timothy Gowers and Terence Tao, both winners of the Fields Medal, one of the most prestigious awards in mathematics; Dan Roberts, an AI researcher at Sequoia Capital and MIT and a published expert in machine learning; and Geoff Smith, the former president of the IMO.

"We are delighted to welcome such esteemed members of the mathematical community to the advisory committee of the AIMO Prize," said Eoghan Flanagan, Director of the AIMO Prize. "I look forward to working with them on developing and growing the prize."

An IMO silver medalist during high school, Loh served as the national lead coach for the U.S. team for a decade. During that time, the U.S. placed first four times. In 2019, he received the U.S. Presidential Early Career Award for Scientists and Engineers, and in 2002, he received a Hertz Fellowship.

"International Math Olympiad problems are difficult because they involve many sequential layers of reasoning," Loh said. When teaching, Loh focuses on the power of mathematics, and how it provides a framework for thinking so that students focus less on following steps and more on problem-solving. In the same way that students can work collaboratively through trial and error to answer problems, AI models can do the same. "The next big breakthrough will come from solidifying the concept of proof, and this prize supplies a useful target for that research."

The first publicly shared AI model to enter an AIMO approved competition and perform at a standard equivalent to a gold medal in the IMO will receive a $5 million grand prize. There also will be a series of progress prizes, totaling up to $5 million for publicly shared AI models that achieve key milestones.

Competitions have become opportunities for researchers to shape the next frontier of AI research and advance responsible AI-driven solutions to critical societal challenges. For example, in recent years Carnegie Mellon researchers have participated in XPRIZE competitions that have incentivized the use of AI for breakthroughs in digital learning, landing and operating a robot on the moon and Defense Advanced Research Projects Agency (DARPA) Challenges intended to accelerate the use of autonomous vehicles in challenging environments. A team from Carnegie Mellon's Department of Chemistry and the University of British Columbia recently won a prize to advance Parkinson's research through using computational methods.

Gowers, a professor of combinatorics at the Collège de France, said that while machine learning methods have had spectacular success in several domains, their performance on tasks that involve multi-step reasoning still lags behind.

"Solving advanced mathematics problems has been increasingly recognized as a major frontier for AI, the passing of which has the potential to open up a new wave of important applications," Gowers said.

Tao said that solving advanced mathematical problems with the use of AI remains an incredibly complicated and multifaceted challenge.

"It will be important to experiment with multiple approaches to this goal, and to benchmark the performance of each of them," Tao said. "The AIMO Prize promises to provide at least one such set of benchmarks which will help compare different AI problem solving strategies at a technical level, in a manner that will be accessible and appealing to the broader public."

The first AIMO-approved competitions will open to participants in early 2024. There will be a presentation of progress held in Bath, England, in July 2024, as part of the 65th IMO.

For more information and updates on the AIMO Prize visit: https://aimoprize.com/.