A small team has been working for years on a difficult question. It publishes a breakthrough. Shortly afterwards, an organisation with computing resources of an entirely different order announces a result on the neighbouring, more ambitious question. The first team wonders whether the second accelerated after seeing its work. The second denies it. No one can settle the matter. This configuration will repeat itself.
This morning we reported an announcement accompanied by a dispute of this kind, without taking sides and without any means of doing so. Let us take the time to consider the general question.
What priority actually protects
Science has long operated on a simple rule: credit goes to the first to publish. This rule is not a convention of vanity; it serves three functions.
It rewards risk-taking. Working for years on a question that may yield nothing presupposes a counterpart if one succeeds.
It encourages publication. Without priority, it would be rational to keep one's results to oneself. The rule pushes towards sharing.
It distributes resources. Posts, funding, doctoral students: everything follows publications.
This rule presupposes something that is never stated: that competitors have comparable resources. The race is fair if everyone runs on legs.
What an extreme asymmetry changes
When one actor can launch ten thousand explorations in parallel and another has an office and a few doctoral students, the race no longer means the same thing.
This does not mean the former is cheating. Using the means at one's disposal is legitimate, and refusing a tool because it is powerful would be absurd.
But it produces a perverse effect: publishing a breakthrough becomes dangerous. If announcing a partial result signals to a better-resourced actor which direction is promising, the rational strategy becomes to stay silent until one has everything.
Yet the publication of intermediate steps is what moves a field forward. A rule that discourages sharing attacks the very functioning of research.
This problem was not born with AI. The race to sequence the human genome, in the 1990s, pitted a public consortium against a private company with machines and capital. There were numerous accusations of using public data to accelerate a private effort. The resolution is interesting: it was not legal but normative. A rule of immediate data publication was introduced, precisely so that an advantage in resources would not turn into capture. Science has already faced this question; it answered it with norms, not with courts.
What can reasonably be asked
Three principles, and they are more modest than a theory of ownership of questions.
A question belongs to no one. This is the foundation, and it must not move. An open problem is a common good, and no one can forbid anyone from working on it. Claiming the opposite would close off research.
Context is declared. What can be asked is transparency about what informed a piece of work: which publications were read, when the effort was launched, what steered the direction. This is not an accusation, it is good practice.
Contribution is described. As we wrote yesterday in our article on the authorship of a discovery: stating what each party contributed is more useful than a debate about paternity.
The real risk, which is not injustice
We must name what is at stake beyond individuals.
If working on a difficult question becomes risky for a small team — because a better-resourced actor can take over the direction and finish in a few days — then small teams will stop tackling the big questions.
They will retreat to less visible subjects, less likely to interest an industrial laboratory. The field would stratify: the big questions to the big resources, the rest to the others.
That would be a loss, and not only a moral one. The history of mathematics suggests that breakthroughs often come from lone individuals with an unexpected angle, not from computing power. This is what we noted in our article on the capacity of models to have ideas: these systems exploit remarkably and invent poorly.
What to take away
The question in the title has a firm answer: a research question belongs to no one, and that must remain the case.
What must therefore be discussed is not ownership but the norms of conduct in a competition that has become highly asymmetrical. Declare what one has read, say when one started, describe who did what. These are light requirements, and they would suffice to avoid most disputes.
They do, however, presuppose something that science has built slowly and that is quickly lost: the idea that the manner of obtaining a result matters as much as the result. A discipline in which only the finish line counts ends up no longer knowing whom to believe.