Terence Tao Warns: AI Is Depleting Mathematics' Open Problems

Fields Medal winner Terence Tao recently posted a thought-provoking comment on the social platform Mathstodon, pointing to the potentially profound impact artificial intelligence could have on the mathematical research ecosystem. He warned that AI's powerful problem-solving capabilities are consuming mathematics' precious open problems in a "non-renewable" way — and may even undermine an open science tradition that has persisted for centuries.

Open Problems Are Being "Non-Renewably Mined"

Tao argues that high-quality, fruitful open problems are inherently a scarce resource. These problems often take years or even decades of accumulated effort before they can even be properly posed. They serve not only as milestones of mathematical progress, but also as vehicles for training young researchers and incubating new methodologies.

"I recently wrote that these good, fruitful open problems are now being mined in a non-renewable way, leading to a potential scenario where such problems become scarce."

The analogy is sharp. Traditionally, a good mathematical problem ferments slowly within the academic community: researchers develop partial results around it, build new tools, train students, and ultimately drive the maturation of entire subfields. AI's involvement threatens to drastically compress this natural growth process.

Even a Rumor Can Trigger an AI Feeding Frenzy

Tao has observed a troubling new phenomenon: even the mere rumor that someone is working on a problem is enough to trigger a massive AI-driven assault, "flattening" the problem before the original research project has had time to fully develop.

"We are already seeing cases where even a rumor that someone is working on a problem can trigger a large amount of AI-driven effort to flatten it before the original research project has had time to develop its full potential."

The word "flatten" is telling. It describes not just the act of solving a problem, but the instantaneous erasure of all the research depth that problem contained — fertile ground that might have given rise to a whole series of derivative works is turned into a wasteland by a single AI blitz. For research programs built around a central problem, this means years of accumulated value could evaporate overnight.

Could This End the Tradition of Open Science?

What concerns Tao most is how this dynamic distorts the incentives driving researcher behavior. When sharing a promising research direction means immediately having it snatched away by an AI onslaught, the rational choice may become — to stop sharing any valuable research ideas with the academic community at all.

"The incentives now seem to be pointing toward no longer sharing any promising research directions with the broader community, which would reverse centuries of open science tradition and cause serious long-term damage to the field."

The foundation of open science is the free flow of information: researchers share unfinished ideas through preprints, conferences, and public discussions, sparking collisions and collaboration. If AI turns "sharing" into an act of self-harm, researchers will be driven toward secrecy and closure — the very opposite of the open spirit upon which the scientific community thrives.

A Structural Risk That Has Gone Largely Undiscussed

Interestingly, Tao himself is no opponent of AI — he has publicly experimented with large language models and formal proof tools to assist mathematical research on multiple occasions, and has expressed an optimistic view of AI's potential in mathematics. That is precisely what makes his warning all the more worth heeding: this is not a reflexive resistance to new technology, but a calm observation from a deeply engaged practitioner about ecosystem-level side effects.

The efficiency gains AI delivers to individual researchers are obvious. But AI may simultaneously be eroding the collaborative incentive structure of the entire academic community. When the speed of competition is infinitely amplified by AI, the "first to publish wins" pressure squeezes out the deliberateness and depth that academic research depends on.

Implications for the Broader Research World

Although Tao's remarks focus on mathematics, the problem's relevance extends far beyond. Any discipline driven by open problems and the sharing of research directions — from theoretical physics to computer science — could face similar difficulties.

This raises a set of questions worth pondering across the entire research community:

• How do we harness AI for efficiency while protecting the "ecological diversity" of research problems?
• Do academic incentive mechanisms (such as attribution and priority recognition) need to adapt to AI's new pace?
• Does the framework of open science need to be redesigned to prevent the trap of "publish and get harvested"?

Tao's remarks offer no answers, but they precisely identify an emerging structural tension: the more capable AI becomes, the more urgently we need to revisit the foundational rules of scientific collaboration. In chasing AI's problem-solving prowess, how do we protect the "soil" in which great problems are nurtured? That may be the defining challenge mathematics — and fundamental science as a whole — will have to confront in the years ahead.