There’s an interesting review on prediction errors and temporal difference learning theory in the latest Trends in Cognitive Sciences. (Really, it’s fascinating stuff.) But I don’t want to talk today about the content of the article. Instead, I want to discuss its form.
The vast, vast majority of science articles follow the same basic pattern: abstract, introduction, methods, results, discussion. (Update: As bsci points out in the comments, review articles obey a slightly less strict pattern, but they’re still pretty predictable.) There are no stories, no narrative, no amusing anecdotes. (Occasionally, there’s a pretentious epigraph – papers on LTP and memory always quote Proust – but that’s about it.) Rather, there’s just line after line of jargon leaden prose in the passive tense. If you’re lucky, you might get some passive-aggressive criticism of a competing hypothesis.
This is why the prediction error article got my attention. The authors, Yael Niv and Geoffrey Schoenbaum, didn’t obey the typical journal format. Instead, their paper consists of ten fundamental questions (i.e., “What is a prediction error and what is it good for?”) followed by a series of straightforward answers. The prose is elegant and clear, sprinkled with helpful metaphors. (Their opening metaphor about fancy Bordeaux wine was particularly helpful.) The images are relevant Dilbert cartoons, which say more than most bar graphs:
Why can’t more science papers break the mold, even a little? I’m not saying scientists should start writing like Robbe-Grillet, but why not experiment a little with the writing process? Niv and Schoenbaum confess that the form of their article happened by accident:
This manuscript grew out of a set of e-mail questions Geoff sent Yael after they met at a reward circuits meeting held at Lake Arrowhead in California in 2006. The answers to the questions were so useful that they were often printed and passed around the laboratory. This generated further discussion and ideas and, ultimately, more questions. Because these dialogues were extraordinarily useful to the authors, they thought they also might be helpful to others.
It’s too bad scientists can only come up with new ways of writing science papers when they think they’re not actually writing a science paper. I understand, of course, the need for standardization and clarity, but I still think it’s time to update and liberate the format of most peer-review articles. Consider the results section. Instead of wasting space on PCR primers or the details of fMRI experiments, why not put all that stuff online, in a supplemental file? What I’d find much more useful is a youtube video of the experimental process, a video demonstration of how you sorted the fruit flies, or measured the mice in the water maze, or tested subjects in the brain scanner.
And why is there a prohibition against good science stories in science journals? Why not take a few sentences and describe where the idea for the experiment came from, or some of the experimental failures along the way, or the most surprising results. I’m not asking for a novel: all I want is an anecdote or two, a few honest insights into the messy reality of the scientific process.