The Question Patent Law Stopped Asking
Patent law spent most of the 20th century trying to answer who invented something first. It gave up, and the reason it gave up is worth stealing.
The old US system, first-to-invent, asked a genuinely empirical question: who conceived the idea, and who reduced it to practice, and when. You proved it with lab notebooks, witnesses, corroborating dates. When two people filed competing claims, the patent office ran an interference proceeding to work out whose priority date actually came first. This is a reasonable way to settle a dispute between a lone inventor in a garage and a rival who heard about it six months later. It assumes there's a fact at the bottom of the question, and that enough investigation will surface it.
It stopped being reasonable once invention became institutional. Once you've got funded corporate labs running parallel research programs on the same problem, with the same literature, the same incentives, and comparable resources, the gap between competing teams stops being months and becomes weeks, or days. Both sides show up with full documentation. Both sides have real priority dates. The interference proceeding doesn't fail to find an answer because the evidence is thin, it fails because the evidence on both sides is thick and points to a genuine tie. More fact-finding didn't resolve these disputes, it just produced two complete, well-documented claims to the same narrow window, and years of litigation over a question that had stopped having an answer.
The fix wasn't better evidence. In 2011 the US scrapped first-to-invent and moved to first-to-file, joining the rest of the world. This is, by design, worse at identifying the true originator. It's been criticized on exactly that ground: it rewards whoever has a patent attorney and a fast path to the filing office, not whoever actually had the idea first at a workbench somewhere. But a filing date is a timestamp. You can check it in an afternoon. The system traded a question that was true but undecidable for one that's arbitrary but administrable, and it made that trade on purpose, after decades of watching the undecidable version waste money and never resolve.
The generalizable part isn't the patent-specific mechanics. It's the diagnosis: "who did it first" presupposes a moment of origination that investigation can locate if you dig hard enough. That assumption only holds when activity is sparse enough that priority claims are separated by real distance. Once enough independent effort is running in parallel on the same target, the assumption breaks, not because someone is hiding the real answer, but because there isn't a privileged moment left to find. A tie close enough in time is just a tie, and no amount of forensic effort turns it back into a winner.
That condition is getting more common, not less. Running many independent attempts at a problem, whether it's a proof, a design, or a fix, is now cheap enough that it looks like mid-century corporate R&D parallelism compressed into minutes. The asymmetry that makes this possible is that checking a result is cheap and producing candidates is the thing you can afford to waste: run a thousand attempts, keep the one that verifies. A swarm run this way doesn't leave a lab notebook behind. It leaves logs, if you're lucky, and the logs don't contain anything resembling a conception date, because nothing inside the run corresponds to a locatable moment of having the idea.
Math prize committees are going to hit this directly once proof search is routinely done by swarms rather than people. When the proof that lands came out of a thousand parallel attempts and no single run of it "had the insight" in any sense a human biography could use, asking who solved it is asking an interference-proceeding question with worse evidence than patent law ever had to work with. The instinct will be to adjudicate anyway, to reconstruct which branch contributed which step and apportion credit accordingly. That instinct is exactly the one patent law indulged for decades before admitting the underlying fact didn't exist in recoverable form.
What will actually work is some version of filing: a submission timestamp, a registration event, a rule that doesn't ask who deserves the credit and just picks a moment and calls it the record. It will be unfair in the same direction patent filing is unfair, favoring whoever's fastest to the administrable act over whoever had the better claim to originality. That unfairness isn't a bug to design out later. It's the whole point of swapping a true question for a decidable one.