Discovery & AI

Doing Research With Critics and Experts

Also called: Scientific collaboration

Personal interest
Philosophical question
Working interpretation

Real progress in science needs critics, experts in the field and people willing to strengthen weak arguments. I want to build working partnerships around evidence rather than around promoting myself.

Who could disagree most usefully with the result I like best?

Why it attracts me

Every author has blind spots about their own work, and I am no exception. The most useful person to meet is often the one who can disagree most usefully with the result I like best.

The idea

Collaboration in science is not only dividing up tasks. It includes roles that exist to slow things down: reviewers, specialists who know the earlier literature, and people who try to break the argument. In 2009 the mathematician Tim Gowers asked whether many people, each adding small ideas in an open online discussion, could solve a hard problem together. The first of those Polymath projects reached its main goal, a new proof of a known theorem, within weeks.

An example

In my Agentic Maths and Agentic Solvers side quests, the automated work deliberately stops at a human gate. Once a result is ready for expert review, more machine polishing is judged less valuable than a specialist's judgment, a check of the literature and peer review. The Agentic Solvers package is frozen while it waits, so that any further change comes from an error report, newly found earlier work or human feedback.

Where it connects

Collaboration links to mentorship (Teaching Judgment, Not Just Answers): teaching others how to reason spreads capability further than solving one problem myself. It also depends on honesty about what is new (Checking Whether Someone Got There First) and on results that others can rerun (Results That Hold Up When Rerun).

Questions I'm still exploring

How do I find experts willing to look at work from someone outside their field?
What makes criticism feel useful rather than threatening?
How should AI contributions be described when work is shared with collaborators?

Sources and further reading

Tim Gowers, "Is massively collaborative mathematics possible?", Gowers's Weblog (27 January 2009)
Timothy Gowers and Michael Nielsen, "Massively collaborative mathematics", Nature 461 (2009): 879–881 — A short account of what the first Polymath project achieved and how it worked.

Working interpretation: drafted from my notes and interests for review. It is not a direct quotation, and I may still change it.

In short

Scientific collaboration is people with different skills working on the same question, including people whose job is to find what is wrong. Critics catch errors the author cannot see. Specialists know the earlier work and the traps in a field. Collaborators can turn a weak argument into a strong one. Good collaboration is organized around the evidence, so that disagreement improves the result instead of becoming personal.

Why this matters to me

I can push a project a long way on my own, especially with AI tools, but I cannot be my own best critic. Real progress needs people who know a field better than I do and who are willing to disagree. I want to build partnerships around evidence rather than around promoting myself, and I would rather hear the strongest objection early than late.

Connections

Teaching Judgment, Not Just AnswersImpact

Working with others and mentoring them both spread the ability to do research beyond what one person can do alone. Mentoring and writing research together both multiply what people can do, because skill and judgment spread between them.

AI Systems That Discover, Not Just SummarizePart of

Even a highly automated discovery process ends with people: critics and specialists who test a result and judge whether it matters.