Thursday, December 03, 2015

Can science be made to work better?

Here is a longer version of the leader that I wrote for Nature this week.

_______________________________________________________________________

Suppose you’re seeking to develop a technique for transferring proteins from a gel to a plastic substrate for easier analysis. Useful, maybe – but will you gain much kudos for it? Will it enhance the reputation of your department? One of the sobering findings of last year’s survey of the 100 most cited papers on the Web of Science (Nature 514, 550; 2014) was how many of them reported such apparently mundane methodological research (this one was number six).

Not all prosaic work reaches such bibliometric heights, but that doesn’t deny its value. Overcoming the hurdles of nanoparticle drug delivery, for example, requires the painstaking characterization of pathways and rates of breakdown and loss in the body: work that is probably unpublishable, let alone unglamorous. One can cite comparable demands of detail for getting just about any bright idea to work in practice – but it’s the initial idea, not the hard grind, that garners the praise and citations.

An aversion to routine yet essential legwork seems at face value to be quite the opposite of the conclusions of a new study on how scientists pick their research topics. This analysis of discovery and innovation in biochemistry (A. Rzhetsky et al., Proc. Natl Acad. Sci. USA 112, 14569; 2015) finds that, in this field at least, choices of research problems are becoming more conservative and risk-averse. The results suggest that this trend over the past 30 years is quite the reverse of what is needed to make scientific discovery efficient.

But these problems – avoidance of both risk and drudge – are just opposite sides of the same coin. They reflect the fact that scientific norms, institutions and reward structures increasingly force researchers to aim at a “sweet spot” that will maximize their career prospects: work that is novel enough to be publishable but orthodox enough not to alarm or offend referees. That situation is surely driven in large degree by the importance attached to citation indices, as well as by the insistence of grant agencies that the short-term impact of the work can be defined in advance.

One might quibble with the necessarily crude measures of research strategy and knowledge generation employed in the PNAS study. But its general conclusion – that current norms discourage risk and therefore slow down scientific advance, and that the problem is worsening – ring true. It’s equally concerning that the incentives for boring but essential collection of fine-grained data to solve a specific problem are vanishing in a publish-or-perish culture.

A fashionably despairing cry of “Science is broken!” is not the way forward. The wider virtue of Rzhetsky et al.’s study is that it floats the notion of tuning practices and institutions to accelerate the process of scientific discovery. The researchers conclude, for example, that publication of experimental failures would assist this goal by avoiding wasteful repetition. Journals chasing impact factors might not welcome that, but they are no longer to sole repositories of scientific findings. Rzhetsky et al. also suggest some shifts in institutional structures that might help promote riskier but potentially more groundbreaking research – for example, spreading both risk and credit among teams or organizations, as used to be common at Bell Labs.

The danger is that efforts to streamline discovery simply become codified into another set of guidelines and procedures, creating yet more hoops that grant applicants have to jump through. If there’s one thing science needs less of, it is top-down management. A first step would be to recognize the message that research on complex systems has emphasized over the past decade or so: efficiencies are far more likely to come from the bottom up. The aim is to design systems with basic rules of engagement for participating agents that best enable an optimal state to emerge. Such principles typically confer adaptability, diversity, and robustness. There could be a wider mix of grant sources and sizes, say, less rigid disciplinary boundaries, and an acceptance that citation records are not the only measure of worth.

But perhaps more than anything, the current narrowing of objectives, opportunities and strategies in science reflects an erosion of trust. Obsessive focus on “impact” and regular scrutiny young (and not so young) researchers’ bibliometric data betray a lack of trust that would have sunk many discoveries and discoverers of the past. Bibliometrics might sometimes be hard to avoid as a first-pass filter for appointments (Nature 527, 279; 2015), but a steady stream of publications is not the only or even the best measure of potential.

Attempts to tackle these widely acknowledged problems are typically little more than a timid rearranging of deckchairs. Partly that’s because they are seen as someone else’s problem: the culprits are never the complainants, but the referees, grant agencies and tenure committees who oppress them. Yet oddly enough, these obstructive folk are, almost without exception, scientists too (or at least, once were).

It’s everyone’s problem. Given the global challenges that science now faces, inefficiencies can exact a huge price. It is time to get serious about oiling the gears.