There is no doubt that the coming squeeze on global public finances will test governments’ commitment to science and research like never before. No doubt it will also tempt them to tie whatever funding they do release ever more closely to particular industries and challenges.

Indeed, according to Steve Fuller, Auguste Comte professor of social epistemology at the University of Warwick. “the time is ripe for a radical, multinational rethink of why taxpayers should be funding research at all”.

He suggests that the best results are promised by outcome-driven programmes, such as those pursued by the US Defense Advanced Research Projects Agency (and its forthcoming UK version) because devolving research agendas to scientists has “a poor record of success”.

The facts in the UK are otherwise, however. Such a devolution, commonly referred to (however inaccurately) as the Haldane Principle, is designed precisely to let the best work thrive without the political interference that might be used to fund pork-barrelling and pet projects of little scientific (or, indeed, economic) value.

����Vlog

This independence of research funding agencies has seen the UK punch far above its weight given the relatively low proportion of GDP the country spends on R&D (even the government’s pre-pandemic pledge to double public spending over the next five years is part of an ambition just to get overall spending up to the OECD average).

As well as the discoveries that investigator-driven projects make, they also provide the postdoctorally trained person power that is the bedrock of innovation. I have no problems with research designed to solve problems (and I have started two technology companies that are still trading), nor of regional funding structures. Indeed, taxes from the Länder in Germany are the backbone of much of the more applied funding of the Helmholtz Institutes, major sources of innovation, jobs and wealth creation in that country. However, such funding streams should be additional to those for discovery research (often meaninglessly referred to as “blue skies” research), not substitutes for it.

����Vlog

Fuller points out the possibility that “applied” research, such as Pasteur’s, can lead to breakthroughs in fundamental understanding. I agree with him that the distinction between basic and applied research is artificial – a famous aphorism of Nobel laureate Sydney Brenner is to the effect that there are only two kinds of research: applied and not yet applied – but he draws the wrong conclusion in assuming that it is the best or only route.

Governments of whatever stripe have, for decades, desired quick wins. But these are close to impossible for scientists to deliver. What “impact” science does deliver usually originates a decade or two earlier in discoveries by smart people who could not possibly have dreamed of the applications that emerged. The entire bioeconomy is based on discovery research that enabled DNA cloning, the synthesis and sequencing of DNA and proteins, recombinant protein production, the massive boosting of agricultural productivity, and so on. Such research has now enabled the emergence of synthetic biology. 

This is why a from 2010 by Haskel and Wallis showed that “for maximum market sector productivity impact government innovation policy should focus on direct spending on research councils”. In another Brenner aphorism, “science advances by new techniques, new developments and new ideas, probably in that order”. Take deep learning (in the form of large, artificial neural nets, ANNs), commonly bandied about nowadays as “AI”. This has major origins in work in the 1940s by Donald Hebb (Hebbian learning), leading in the 1980s to what were then the most frequent kinds of ANNs: nonlinear multilayer perceptrons.

When the larger versions of these turned out to be hard or impossible to train, the field mostly stood still for almost two decades. But it was kept alive chiefly by three “believers” (Geoffrey Hinton, Yoshua Bengio and Yann Le Cun), who continued to receive limited funding. Only with the advent of new architectures, new algorithms, much larger training sets and massively increased computer power did the modern deep learning revolution take off. The results – including written and spoken natural language processing, image understanding, driverless cars and world-leading computational Go masters – are already spectacular, but only countries that retained and developed expertise in e-science were immediately able to reap the economic benefits – which will be massive.

����Vlog

Curiously, e-science, agritech, autonomous robotic systems and synthetic biology are four of the “eight great technologies” championed by the UK’s former minister for universities and science, Lord Willetts. These secured some £600 million in extra funding in 2013, despite the cuts, delivered entirely via the research councils. It is fine for politicians to recognise the importance of major areas (and provide the extra funding for them). Indeed, even research councils spend about half their funds in “grand challenge” areas, but the selection of what to fund therein must be left to experts.

The structure of the UK system is not broken. All that has been lacking is the wherewithal and will to fund it properly over extended periods. It may be that at a certain level of funding, quality begins to drop, but the present success rate for UK funders is barely 20 per cent, when at least twice that fraction of applications are of international quality.

In other words, we are firmly in the linear part of the curve relating economic benefit to science funding: more input leads to more output. It is good to see that recognised in the government’s recent funding pledges, but it must hold firm on them.

Governments that wish to reap the economic benefits of research should learn from history and ignore advice to upend systems that have long been shown to work extremely well.

����Vlog

Douglas Kell is professor of systems biology at the University of Liverpool and a former chief executive of the UK’s Biotechnology and Biological Sciences Research Council.