Source articles: Why P vs NP Is Equally Hard for AI and Computers · The Strange Case of Navier–Stokes · The Math Heist: AI Must Be Kingmaker, Not King
Pip: Welcome to hemantpandeyhub, where this week the questions are: who owns a theorem, and can a machine even find one in the first place.
Mara: Hemant Pandey has been writing about AI’s role in mathematical discovery — the credit, attribution, and governance problems it creates — and separately about why computational complexity might resist AI’s raw speed advantage entirely.
Pip: So: credit where credit is due, and also, good luck finding the answer at all.
Mara: Let’s start with what happens when AI enters the mathematical ecosystem.
AI, Credit, and the Mathematical Ecosystem
Pip: The core tension here is that mathematics runs on recognition — priority, citation, reputation — and AI is disrupting that economy faster than the rules can keep up.
Mara: The post frames it precisely: “What happens when machines become capable of producing mathematical results faster than the human system can assign credit for them?” That’s the question the Mathathon controversy and the Navier–Stokes dispute both surface.
Pip: And the Navier–Stokes episode is striking — ten thousand AI agents, eighty-eight hours, a claimed Millennium Prize solution — before anyone had agreed on what attribution even looks like in that situation.
Mara: The Buckmaster and Alpöge dispute illustrates the structural problem. A researcher develops an idea, discusses it with an AI system, and the company controlling that infrastructure may not be the same as the person controlling the mathematics.
Pip: The shovel-doesn’t-own-the-gold analogy is doing a lot of work here, and honestly, it earns it.
Mara: It does. The companion piece, “The Strange Case of Navier–Stokes,” pushes back on the other side — arguing that a solved problem opens doors rather than closing them, and that the freedom to struggle is worth preserving but the necessity of struggling is not.
Mara: Both posts land on the same practical demand: stronger provenance. Which AI system, for what tasks, whose ideas, what was independently verified, what conversations were protected.
Pip: Terence Tao’s warning threads through the first post — that pure mathematics involves curiosity, wandering, and apparently unproductive exploration, and an AI optimizing toward a designated answer might eliminate the detours through which humans discover the next problem.
Mara: The proposed remedy isn’t slowing AI down. It’s making AI a kingmaker — enormous power to explore, search, and conjecture — while keeping the mathematical ecosystem’s human incentive structures intact.
Pip: From credit problems to a harder question: whether the search itself can be avoided.
When Faster Search Isn’t Enough
Pip: P vs NP sits at an unusual intersection — a problem where AI’s core advantages, speed and scale, may not touch the actual obstacle.
Mara: The post draws the distinction cleanly: “The arithmetic is easy; selecting the right elements is hard.” Faster computation searches the haystack faster. It does not make the hay disappear.
Mara: The math makes this concrete — reducing an exponential search space by 99.9% sounds spectacular, but zero point zero zero one times two-to-the-n is still theta of two-to-the-n.
Pip: So the question isn’t whether AI can search brilliantly. It’s whether intelligence — human or artificial — can discover a structural rule that makes the search unnecessary in the first place.
Mara: The proposed matrix-transformation approach to TSP is aimed exactly there: finding relationships between distance functions that eliminate non-optimal configurations before they have to be searched, using only polynomially many safe transformations.
Pip: Don’t search the haystack. Repel the hay. That’s the whole thesis in four words.
Mara: The post is careful to label this a working research hypothesis, not a claimed proof. The proof obligations are explicit: polynomial transformations, polynomial work per step, safe elimination, preserved optimum, polynomial reconstruction.
Mara: And the broader point holds regardless of whether this particular approach succeeds — P vs NP may require a conceptual leap that changes the representation of the problem, not just more computation aimed at the existing one.
Pip: Two posts on credit, two on complexity — and the same underlying question: what is intelligence actually doing when it does mathematics?
Mara: Whether the problem is attribution or exponential search, the answer seems to be: build better mechanisms, not slower machines. That’s probably the right place to keep pulling.
Continue the discussion
This episode draws together three HPB articles on AI, mathematical discovery, attribution, and computational complexity. The P vs NP matrix approach discussed in the episode is presented in the source article as a working research hypothesis, not a proof.