ilicon Valley runs on failure. Its unofficial motto, after all, is “Fail fast, fail often,” and it is the region that gave birth to FailCon, where stories of entrepreneurial failure are badges of honor.
That’s more than just cute marketing. As the New Yorker’s James Surowiecki has written, “In the delusions of entrepreneurs are the seeds of technological progress.” Failures, in other words, are not only acceptable, but beneficial.
Scientists would do well to take that message to heart. For proof, look no further than a recent study that suggests researchers, with the help of computer algorithms, can find nuggets of scientific insight in the failed experiments that gather dust in forgotten lab notebooks.
In a new article in Nature, a team of chemists at Haverford College and Purdue University described how they used a computer to outperform scientists at predicting ways to make crystals. Their algorithm considered not only previously successful experiments but those that had failed — what they call “dark reactions” because the data never become public. Not only did the software work, it worked better than chemists with years of experience in the field at coming up with likely molecules.
The Haverford/Purdue group has created a website, the Dark Reactions Project, to encourage other chemists to share their own failures. “The planning and development of such tools is essential if we are to eventually make full use of our ‘failed’ experiments,” Richard Cooper, a crystallographer at the University of Oxford, in England, tells Nature in a story accompanying the paper.
Failures, Stuart Firestein writes in his book “Failure: Why Science is So Successful,” “don’t just lead to a discovery by providing a correction … they lead to a fundamental change in the way we think about future experiments as well.” Firestein, a neuroscientist at Columbia University, continues: “We don’t think of confirming failures, but that’s what we often do. Then we selectively remember the success, since it’s such a relief, and the failure goes unsung.”
And this happens, he notes, throughout one’s career. Successful grant applications, not surprisingly, breed successful grant applications — which, Firestein says, have a tendency to look a lot like each other. “You naturally highlight the successes and propose experiments that will continue this successful line of work with its high likelihood of producing results,” he writes. Meanwhile, those experiments that haven’t yet panned out (but might) gather dust. In the end, he laments, “the lab becomes a kind of machine, a hopper — money in, papers out.”
Part of that, of course, is because science doesn’t appear to have much appetite for anything but success. Journals overwhelmingly publish studies that report positive findings and they ignore negative results, as even a member of the staff of Elsevier, the world’s largest publisher, acknowledges. The bias is rampant and widespread across disciplines. In a 2014 study of social science, for example, researchers at Stanford University found that only 20 percent of null studies — often referred to, somewhat imprecisely, as “negative studies” — were published. Indeed, two-thirds never made it into manuscript form. Meanwhile, 60 percent of studies that turned up positive results were published, according to the researchers.
But publishing more failed results may advance science, as the “dark reactions” study shows. Researchers wouldn’t waste time doing experiments that had already proved to be unsuccessful. Big data approaches could turn up insights into the pattern of failures, and give researchers a better sense of what might work the next time around. It would humanize the practice of science, and perhaps help policymakers and the public understand what their tax dollars support.
So how can we encourage scientists to be more willing to fail, and to tell the world about it? For starters, as long as publishing is so important to academic advancement, let’s make sure null studies see the light of day. Increased sharing of data will help, too, because all information — not just affirmative findings — becomes more valuable. Finally, as Firestein writes, science needs a cultural shift. Failure will become more acceptable in research “when we cease, or at least reduce, our devotion to facts and collections of them, when we decide that science education is not a memorization marathon, when we — scientists and nonscientists — recognize that science is not a body of infallible work, of immutable laws and facts.”
Maybe other branches of science should create their own Dark Reactions Project — and perhaps it’s even time for a FailCon for science. Let’s help failure to come into the light.