Terence Tao Foresees AI-Driven Revolution in Mathematical Research at AI for Science Kickoff 2026

Terence Tao Foresees AI-Driven Revolution in Mathematical Research at AI for Science Kickoff 2026

On February 10, 2026, renowned mathematician Terence Tao delivered a landmark address at the AI for Science Kickoff 2026, co-organized by the Institute for Pure and Applied Mathematics (IPAM) at UCLA and the SAIR Foundation. Speaking before a packed audience of researchers, computer scientists, and policymakers, Tao surveyed the rapid evolution of machine-assisted mathematics and offered a nuanced vision for its future role in the discipline.

"The tools we now have—formal proof assistants like Lean and Coq, large language models trained on mathematical corpora, and collaborative platforms that integrate real-time verification—are no longer experimental," Tao stated. "They are becoming integral to how we do research. The question is no longer if AI will change mathematics, but how we will adapt our culture, training, and publishing norms to embrace it."

Tao highlighted several concrete developments that have emerged in the past three years. Formal proof assistants, once confined to niche applications, are now being used by graduate students to verify complex lemmas in number theory and combinatorics. Meanwhile, large language models such as those fine-tuned on arXiv and MathOverflow have demonstrated an uncanny ability to suggest plausible conjectures, identify overlooked connections between fields, and even generate draft proofs that human mathematicians can refine.

One notable example Tao cited was a 2025 collaboration between a team at DeepMind and UCLA’s IPAM, in which an AI system proposed a novel bound in additive combinatorics that had eluded researchers for over a decade. The conjecture was subsequently proven by Tao and colleagues using a hybrid approach: the AI generated the hypothesis, while human insight provided the structural framework for the proof. "This wasn’t automation—it was augmentation," Tao emphasized. "The machine didn’t replace the mathematician; it expanded the scope of what the mathematician could even imagine."

However, Tao also sounded a cautionary note. He warned against the growing trend of "black-box theorem generation," where AI outputs are accepted without rigorous human scrutiny. "We risk creating a generation of mathematicians who trust the machine more than their own reasoning," he said. "Mathematics isn’t just about correctness—it’s about understanding. If we lose the journey of discovery, we lose the soul of the field."

Tao further discussed the rise of "collaborative proof ecosystems," where mathematicians, AI systems, and automated verification tools operate in a feedback loop. Platforms like LeanDojo and MathHub now allow researchers to annotate proofs with machine-generated comments, flagging potential gaps or suggesting alternative approaches. These systems are beginning to influence how papers are peer-reviewed, with some journals now requiring AI-augmented verification logs alongside traditional submissions.

Looking ahead, Tao predicted that within a decade, the majority of mathematical research will involve some form of machine assistance. He called for universities to integrate computational proof literacy into graduate curricula and urged funding agencies to prioritize grants for human-AI collaborative projects. "The future of mathematics isn’t human versus machine," he concluded. "It’s human with machine. And our responsibility is to ensure that partnership deepens, rather than dilutes, the beauty and rigor of our discipline."

The event, hosted at UCLA’s campus and streamed globally, marked the official launch of SAIR’s $50 million AI for Science initiative, which aims to fund interdisciplinary research at the intersection of artificial intelligence and fundamental science. Over 120 research teams have already applied for grants, with mathematics and theoretical physics receiving the highest number of proposals.