AI in Mathematics: Tool or Thinker?

Modern AI systems are very good at one thing that matters to mathematicians: searching vast spaces of possibilities and surfacing patterns a person might miss. They can suggest a promising direction, generate examples, or point out a similarity between two problems that looked unrelated.

In the episode, Professor Murty frames AI in mathematics primarily as this kind of tool, a powerful assistant for exploration and search, rather than as a replacement for the mathematician. Used this way, it widens the field of view. It helps you look in more places, faster.

The catch is that these systems produce answers that sound confident whether or not they are correct. An AI can generate a chain of reasoning that reads like a proof but contains a subtle, fatal error, or simply invent a step that is not true. It is optimised to produce plausible text, not guaranteed truth.

This is why a fluent, well-formatted mathematical answer from an AI is not, by itself, evidence that the answer is right. Plausibility and correctness are different things, and in mathematics only correctness counts.

Mathematics has an unusually strict standard of truth. A claim is not accepted because an authority asserts it, or because it usually works, but because there is a proof: a chain of logical steps, each of which can be checked, leading from agreed assumptions to the conclusion.

That standard applies to AI output exactly as it applies to a human's. Whatever an AI suggests still has to be verified, step by step, before it counts as mathematics. The verification is not a formality; it is the thing that makes it knowledge.

There is a difference between having the right answer and understanding why it is right. An AI, or a lucky guess, might land on a true statement. But mathematics is not just a collection of true statements; it is the web of reasons that connect them.

Understanding is what lets you adapt an idea to a new problem, spot when it breaks, and build on it. A student who memorises that a result is true has less than a student who can explain why. The same is true of a machine that outputs an answer without the reasoning that secures it.

Deciding which questions are worth asking, recognising when a line of reasoning is elegant or suspect, and knowing when a proof is truly finished are acts of judgment. So far these remain human. AI can accelerate the search, but a person still chooses the destination and certifies the arrival.

That is the balance the episode points to: treat AI as a genuinely useful instrument, and keep the responsibility for truth where it has always been, with the mathematician who verifies the work.