The Future of Math: Why AI Progress is a Mirror, Not a Replacement

The rapid advancement of AI in mathematics provides a clear preview of how these tools will change the broader economy. While the current focus is on solving discrete problems like those in the International Math Olympiad, the real shift is moving from theorem proving to conceptual curation. This transition shows that the bottleneck for both AI and human experts is not raw computing power, but the ability to create and verify new, useful abstractions. For the reader, understanding this shift offers a distinct advantage: rather than competing with AI on brute force execution, the highest value will go to those who act as curators, navigating the infinite landscape of AI-generated insights to identify what is actually worth building.

The Hidden Cost of Solved Problems

Most people view AI’s ability to solve complex math problems as a simple indicator of capability. However, the dynamics are deceptive. As Grant Sanderson notes, the dirty secret of competitive math is that it is often grindable, meaning problems can be solved through massive parallel rollouts. The immediate benefit is a high score on a benchmark, but the hidden cost is a reliance on environments where success is easily containerized.

When we look ahead, we see a divergence. In fields like coding or math, where progress is deterministic and grindable, AI will dominate. In fields requiring theory of mind, or the ability to project oneself into the future needs of a reader or user, AI remains fundamentally limited. The competitive advantage for humans lies not in the proof, which is the output, but in the definition, which is the intent.

How good mathematicians prove theorems, great mathematicians come up with conjectures, and the greatest mathematicians come up with definitions.

-- Grant Sanderson

The 100-Year Verification Loop

A common fear is that AI will solve problems like the Riemann Hypothesis but leave human understanding stagnant. Sanderson points out that the history of mathematics is full of unsolved expository problems, which are theorems that are proven but not understood.

The system responds to this by creating a theorem economy that currently acts as a parasite on the definition economy. Historically, the person who defined the concept also proved the theorem. AI threatens to decouple these. If AI automates the proof but fails to provide the conceptual mountain building required for human intuition, we face a future of alien mathematics that is technically correct but practically useless for human progress. The lasting advantage goes to the curator who can distill these complex, machine-generated chains of logic into human-parsable narratives.

There is a difference between proof and explanation.

-- Grant Sanderson

Why Immediate Pain Creates Lasting Moats

Sanderson highlights that the most productive learning occurs when a human artifact, such as a book, video, or lecture, organizes concepts in a way that builds motivation. AI struggles here because it lacks the social coaching element of teaching.

Most teams and students optimize for the immediate problem, ignoring the downstream effect of how knowledge is structured. In a world of AI abundance, the proof is a commodity, while the curation of what is worth knowing is scarce. The systems thinking here is clear: as AI lowers the cost of generating answers, the value of asking the right questions and curating the right paths compounds exponentially.

The social role that mathematicians serve actually does not change that much. You still have a sense of as a public we sort of feel like there is value to basic science and we are trusting in the judgment of mathematicians to determine where their time is best spent.

-- Grant Sanderson

Key Action Items

• Shift from Execution to Curation (Immediate): Stop viewing your value as the doer of tasks. Start building a personal repository of high-quality, human-curated resources like books, papers, and lectures that you can use to ground AI outputs.
• Audit Your Grindability (Next Quarter): Evaluate your current work. If your tasks are highly deterministic and containerizable, you are in a high-risk zone for automation. If they require theory of mind or complex stakeholder management, lean into those areas.
• Develop Theory of Mind Skills (12-18 Months): Focus on roles that require deep relational understanding, such as teaching, mentoring, and high-level strategy. These are the most stable roles because they are inherently social rather than just informational.
• Adopt the Curator Mindset (Ongoing): When using AI to learn, stop asking it to explain everything. Use it to identify the best human-written sources. Treat the AI as a search engine for expertise, not the source of truth itself.
• Seek Out Unsolvable Problems (6-12 Months): Invest time in learning how to define new problems and conceptual frameworks. As AI becomes a master of solving existing problems, the premium on defining new domains of inquiry will skyrocket.