Artificial intelligence (AI) is starting to make real contributions to mathematics, transforming not only how problems are approached but also how research gets done, according to an in-depth discussion with celebrated mathematician Terence Tao. While claims that AI will soon solve the deepest problems in math remain exaggerated, recent breakthroughs have shown generative AI tools tackling previously unanswered questions — particularly in longstanding collections like the Erdős Problems posed by Hungarian mathematician Paul Erdős. Some AI-generated proofs have checked out upon review, prompting careful optimism from experts.
Tao — widely regarded as one of the world’s most brilliant living mathematicians — describes AI’s current abilities as similar to capable but not highly creative graduate-level assistance. AI models have been effective at resolving simpler or “low-hanging fruit” problems and performing tedious calculations that would otherwise consume researchers’ time. However, he emphasises that the deepest mathematical discoveries still require the context, creativity, and insight that only humans bring to the process.
One of the most interesting impacts of AI, Tao says, is the potential for hybrid human-AI collaboration. Rather than fully autonomous solutions, many of the useful advances come when mathematicians work side-by-side with AI, using it to handle computational grunt work or surface overlooked techniques. This could allow mathematicians to explore larger scales of problems and free up human effort for conceptual breakthroughs — a shift toward a new way of doing mathematics rather than replacing mathematicians outright.
Despite progress, Tao also points out limitations: current AI tools often fail to express confidence levels or reasoning clarity, making it hard to trust output without careful verification. He advocates for AI systems that support ongoing dialogue with users, rather than one-click automation, so that mathematicians can guide and interpret results meaningfully. Overall, AI is expanding the mathematical toolkit — but the core of mathematical discovery remains deeply human.
