Science Needs Failure — And the Systems That Fund It Rarely Admit That

The experiments that don’t work are still doing work

Ask most people to picture science and they will imagine a breakthrough: a discovery announced, a prize awarded, a headline declaring that something long suspected has finally been proved. What that picture leaves out is the overwhelming majority of scientific labour — the runs that produce nothing, the hypotheses that collapse under scrutiny, the instruments that refuse to cooperate, the years spent on a promising avenue that turns out to be a dead end.

A new argument circulating from Princeton University Press puts that omission at the centre of the conversation: science needs failure. Not as an unfortunate by-product to be minimised, but as a structural necessity — the mechanism by which a field learns what is not true.

Why failure is not waste

The logic is simple enough that it is easy to overlook. Knowledge advances by elimination as much as by confirmation. Every well-designed experiment that returns a negative result narrows the space of possible explanations. A hypothesis that has been tested and rejected is a piece of information, and often an expensive one to obtain. Treating it as a wasted effort is like treating a map’s blank spaces as a failure of cartography.

There is also a subtler point about risk. Research that cannot fail is usually research that was never especially uncertain to begin with. If a project is guaranteed to produce a publishable, tidy result, the questions it asks are probably safe ones. Genuinely ambitious science — the kind that reshapes a field rather than incrementally extending it — carries a high probability of going nowhere. A research ecosystem with a low failure rate is not an efficient one; it is a timid one.

The incentives point the other way

The difficulty is that almost every institutional signal in modern research rewards the appearance of success. Grant applications are strongest when the outcome looks assured. Journals have historically preferred positive, striking findings to null results. Careers are built on publication counts, citations and the narrative of steady progress. Graduate students learn early that a failed project is something to be quietly reframed rather than reported.

The consequences are well documented in the broader debate over research reliability. When negative results go unpublished, other laboratories repeat the same dead ends without knowing it. When only the most eye-catching findings reach print, the published record drifts away from the underlying reality. Efforts such as registered reports — where a study is reviewed and accepted on the strength of its design, before the data arrive — exist precisely to break that link between outcome and reward.

Making room for the dead ends

Taking failure seriously means more than telling scientists it is acceptable. It means funding bodies willing to back projects that may return nothing, journals and repositories willing to host null results, hiring committees willing to read a CV for the quality of the questions asked rather than the neatness of the answers, and institutions patient enough to let a difficult problem take a decade.

It also means being honest with the public. Science communicated purely as a sequence of triumphs sets an expectation that research cannot meet, and leaves audiences bewildered when findings are revised or retracted. Revision is not a malfunction of the scientific method; it is the method. A discipline that could not be wrong would have no claim on being right.

The uncomfortable conclusion is that a healthier scientific culture would look, from the outside, rather less successful — and would almost certainly discover more. Read More