AI Mathematicians on the Rise: Experts Debate Future Roles

Research from Microsoft identifies “mathematician” as one of the 40 jobs most likely to be impacted by generative artificial intelligence (gen AI). This revelation surprised many, especially during discussions at the recent Heidelberg Laureate Forum (HLF), which connects emerging researchers with leading figures in mathematics and computer science. Experts at the event expressed a range of sentiments as they explored the implications of AI in their field.

One of the notable voices at HLF was Yang-Hui He, a mathematical physicist at the London Institute for Mathematical Sciences. Despite his enthusiasm for AI, reflected in his authorship of a textbook on the subject, He has voiced concerns about the capabilities of gen AI. “It is doing mathematics so well without any understanding of mathematics,” He remarked during a panel discussion, revealing a mix of admiration and apprehension. He pondered, “Where is our place?” as AI continues to advance.

Many speakers at HLF highlighted the tireless nature of AI as a significant advantage over human mathematicians. Theoretical computer scientist Sanjeev Arora noted that AI’s performance stems from its relentless capacity to analyze and solve problems. As he succinctly put it, AI does not require breaks, unlike humans, who have historically been described as “machines for turning coffee into theorems.”

Arora further emphasized the potential of AI in mathematics through reinforcement learning. This process involves the AI model receiving a set of questions, generating multiple solutions, and refining its approach based on human feedback. Given the formalized nature of mathematics, Arora anticipates a future where AI systems could function as “proof assistants,” verifying the correctness of solutions without human intervention. He is currently developing one such assistant, named Lean, with colleagues at Princeton University.

The potential applications of AI do not end with solving problems. Arora suggested that AI could also be tasked with generating mathematical questions, writing papers, and even participating in peer reviews. His perspective resonates with others in the field. Javier Gómez-Serrano, a mathematician at Brown University, commented on AI’s ability to deliver results in a matter of days that would previously have taken human mathematicians months. “Progress has been quite quick,” he stated, acknowledging the rapid advancements in the field.

On the other hand, Maia Fraser from the University of Ottawa raised concerns about the pace of these changes. She emphasized that while the evolution of AI presents remarkable opportunities, it also risks proceeding without adequate consideration for human needs. “While we can’t un-invent AI, we do have agency over what we want,” Fraser pointed out. This brings to light a critical question: Do we want a future where AI mathematicians operate entirely independently of human input?

In response to the potential dislocation of human mathematicians, He acknowledged the benefits AI could bring. “I really want to see a proof of the Riemann hypothesis,” he quipped, eliciting laughter from the audience. He further mused that if this means human mathematicians become “priests to oracles,” then that might be an acceptable trade-off in the quest for knowledge.

As the dialogue around AI’s role in mathematics continues, the implications for both the discipline and its practitioners are profound. The prospect of superintelligent AI mathematicians is no longer a distant possibility; it is an evolving reality that demands thoughtful engagement from the mathematical community.