Materials & energy
Problems should attract intelligence.
Crowdfunding, except people fund the attempt to find an answer.
Clean water, a cure that will never sell, a proof with no product at the end of it. Everyone needs them solved and no one owns them, so nothing that owns a datacentre is pointed at them.
A problem with no owner is not solved by waiting for one to appear. Cystic fibrosis had a single advantage the world's hardest problems do not: a community who could decide to own it. They stopped waiting to be chosen and paid for the attempt themselves, and it worked.
Too few patients to justify the research. Nobody is coming.
Their own foundation starts writing cheques to a small biotech. Around $150m over a decade, from the people with the disease.
Kalydeco is approved. The first drug to treat the cause of cystic fibrosis rather than its symptoms.
The foundation sells its royalty rights and puts the money back into the disease.
They did not wait to be chosen. They funded the attempt.
Sixteen years, one disease, and a constituency wealthy and organised enough to pay for it. Clean water, cheaper materials, whole fields of mathematics have no such constituency, which is exactly why they sit unsolved. Many On One's bet is that when machines do the searching, the attempt gets cheap enough that a problem no longer needs an owner, only enough people who want it answered.
Cystic Fibrosis Foundation and Vertex Pharmaceuticals; royalty sale to Royalty Pharma, November 2014.
Today, a handful of companies allocate most of it, to their customers and their roadmaps, and that is a reasonable thing for a company to do. Each of us gets a slice: a chat window, an API key, a few GPU hours. That is real, and it is new.
But a slice cannot move a shared problem, and slices do not add up on their own. Clean water, a battery that lasts, an open question in mathematics: nobody's slice is big enough, and nobody pools them. Many On One is the pool, and the many decide where it goes.
Every problem here has a defined objective, a visible budget, official solvers, and a search that never stops. This is what a problem page looks like; backing opens with the first problems.
Kickstarter funds people to make products. Many On One funds coordinated machine intelligence to find answers.
An illustration of the funnel, not a run. Shown on a log scale: each stage is roughly an order of magnitude smaller than the last, and nothing is verified until it survives every stage.
A critic hunts logical flaws, a falsifier builds counterexamples, a fact-checker tests every factual statement. Each drawn from a different model family, because models favour their own output.
Where a claim reduces to computation, a sandboxed evaluator settles it exactly. No opinion, no model. The only channel that reaches the top rung.
Two agents reasoning differently reaching the same claim counts as evidence. Disagreement opens an investigation, not an average.
Agents generate claims. Gravity manages evidence, and weights it by independence, not headcount.
In a world of AI slop, verification-first is the brand.
A platform taking public money to do AI research lives on trust. Every serious problem has real experts attached, by name, with their reputations on the line.
Verified scientists or technologists, ORCID and institutional email, or a named company role, who define the problem.
Two independent named experts in the problem's field, who must sign off before it goes live.
One respected researcher attached to the company early, more urgent than any engineering hire.
Cloud labs, university groups, pharma and materials teams who run the physical scoring the engine can't.
Our pitch to scientists is the opposite of hype: verification-first, we undersell, which is exactly what makes them willing to sign their name.
The honest catch: no expert attaches their name until it's real, and it isn't real until a name is attached. A working prototype breaks that, we lead with a demo, not a deck.
One problem, real engine, real ledger.
Public, with a named scientific reviewer attached.
The catalogue expands across fields.
Donated compute joins purchased intelligence.
The engine is real and running today. The rest of it does not exist yet, which is the interesting part.
Or, in the language the public will see: Put intelligence behind a problem.
Jeremy Duffy · hello@manyonone.com