Research funding is badly broken. Scientists spend a large fraction of their time writing grant proposals, which are largely speculative fiction. No one really knows for sure what they¡¯ll be researching in the next five years. Indeed, sticking to what the grant proposal said when it turns out to be a dead end, or when a much more promising direction emerges, would be a dereliction of the scientist¡¯s duty.

Moreover, the funding process is biased toward incremental research because ambitious proposals are too easy to poke holes in. Many researchers resort to the subterfuge of proposing to do what they¡¯ve already done but not yet published. The result is wasted time and money ¨C and missed scientific breakthroughs that we can only guess at. There must be a better way.

Some have proposed funding grants by lottery, but this would be even worse. Imagine if consumers chose which products to buy by lottery, or if survival of the fittest in nature was replaced by a lottery among competing creatures. Lottery schemes typically include only proposals above a given quality threshold, but this is just a combination of two bad ideas: review panels and random selection. Another popular idea is to spread funding equally among all applicants, but this is just as bad, minus the randomness.

The answer, I believe, is right before our eyes. Scientific research is now an industry, for better or worse, and it should be funded like any other industry: by results. You buy a car, not a document describing a car that may or may not be built, and the carmaker gets paid when it delivers the car. You (or your insurer or national health service) pay for a medical procedure when you receive it, not when it¡¯s prescribed. Science should be no different.

ÖйúAƬ

This is, first and foremost, the right incentive: scientists should strive for real impact, not to impress a panel of busy peers, with their agendas, biases and blind spots. Payment by results also frees up all the time scientists spend writing grant proposals, directly increasing their productivity (not to mention their quality of life).

Yet a suggestion like this is often met with horrified gasps. Science is not an industry! It would be the end of basic research! Short-termism would rule! Research results are impossible to evaluate!

ÖйúAƬ

On the contrary: properly done, funding by results would bolster basic research and favour the long term over the short, to everyone¡¯s benefit.

Here¡¯s one option: pay per publication. Funding agencies would maintain a list of approved venues, periodically updated. Selecting these is far easier and less time-intensive than selecting individual research proposals. When a paper is published, half the payment goes to the authors and half to their citations. The authors decide how the citation payment is to be apportioned: references that are major enablers or precursors of the new research receive the most. Those references, in turn, send half their revenue to their own references, and so on recursively.

This is a well-known credit assignment process in bibliometrics and the basis of Google¡¯s search ranking algorithm, using hyperlinks instead of citations. It works remarkably well there; why not try it here? A basic research breakthrough that winds up being widely influential can produce a large financial return, as it deserves, even if it¡¯s a small amount at a time. A research contribution with a long lifetime produces far more revenue than a short-lived one, incentivising long-term research.

But if research is only paid for upon publication, who makes the upfront investment? The researchers¡¯ employers, of course ¨C as in any other industry. But, with its exceptionally high risk and return, research should also be a magnet for venture capitalists, who have plenty of funds looking for opportunities and strong expertise in making similar decisions. Researchers can minimise their risk by banding together in groups, agreeing to share the resulting revenue. Pay-per-pub, as we may call this approach, will unleash a wave of innovation in research funding, increasing the collective intelligence of the system while decreasing waste.

ÖйúAƬ

The fact that pay-per-pub piggybacks on the existing peer review process is efficient and adds sorely needed motivation for reviewing to improve. It will always be imperfect, of course. All the same politics is still present, and publication is a biased and incomplete gauge of impact. But it¡¯s much better than the alternatives and need not be the whole story.

A fraction of patent revenue can similarly go to the publications cited in the patent application. Companies, governments and other organisations can be incentivised or required to produce lists of the research they found useful, with a fraction of tax revenue apportioned accordingly. While companies may have competitive reasons to hide their allocations, it can be done confidentially, with only the aggregate results published. All organisations have an interest in seeing that the research that is most useful to them is funded.

It boils down to this: every educator, biologist and technologist knows that continuous improvement requires, above all, continuous feedback. Today, that feedback is absent from research funding, with appalling results. A hundred years ago there may have been no practical way to implement it, but now there is. It¡¯s time to bring research funding into the 21st century.

Pedro Domingos is professor emeritus of computer science and engineering at the University of Washington.

ÖйúAƬ