March 11, 2015

The educational limitations of watching someone write


It was hard to think of a good image for this post, so here we are. (Via

It was hard to think of a good image for this post, so here we are. (Via

You’re going to judge this article on the caliber of the writing presented to you—all of it, beginning to end. You’ll likely form an opinion incrementally, word by word, sentence by sentence, paragraph by paragraph, until you reach your conclusion at the same time I do. Even if you were to analyze a piece of writing in discrete sections, the next one follows the last immediately, the natural momentum of reading compelling you to continue.

It’s an imperfect system of evaluation; every system of evaluation fundamentally is. But if you could see my writing formed a keystroke at a time, pauses and revisions included, what would you learn from it? What would I learn from it?

Draftback, an extension to Google Chrome, takes literally no time to download. Click “Add to Chrome” and it’s part of your browser so quickly a download bar doesn’t even have time to appear; the application has already been installed. It can now track the keystroke inputs of any Google Doc to which you have editing rights—even those written before you had the app—capable of playing back the journey of how a finished text formed on a once-blank page.

This is a data-minded approach to a discipline still in the nascent stages of collaborating with the full potential of information technology. Yes, the Internet itself was a pretty revolutionary boon for writing dissemination (you wouldn’t be reading this without it, obviously), but using data to inform how writing is made is still trickier. We’ve seen a programmer-turned-author publish all of his novel’s drafts online, but those are still iterations of a whole product. Draftback provides the capability to watch revisions as they occur, not just look at the before and after.

According to Chadwick Maltin over at FiveThirtyEight, James Somers, the programmer behind Draftback, sees his invention as potentially revolutionary to the teaching of writing.

Somers started all this because he thinks the way we teach writing is broken. “We know how to make a violinist better. We know how to make a pitcher better. We do not know how to make a writer better,” Somers told me. In other disciplines, the teaching happens as the student performs. A music instructor may adjust a student’s finger placement, or a pitching coach may tweak a lefty’s mechanics. But there’s no good way to look over a writer’s shoulder as she’s writing; if anything, that’ll prevent good writing.

It’s utopian thinking that Draftback, in its current form, is the missing link that will make a writer productively tweakable; that, granting that exceptions will exist, the data can reveal a formulaically successful writing process. And true to utopian thinking, it comes with inherent flaws.

The first of them is a simple, common issue that most every endeavor into data analysis faces: limited sample size.

While your Google Drive might already house the manifesto to your new religion or your Empire fan fiction (or both in one), your writing is not Somers’ greatest interest—that is, unless you’re A.O. Scott, in which case please tell me more about your new Empire-based faith.

Somers wants to use Draftback to peek over somebody’s shoulder — ideally somebody really good. His personal goal is to get A.O. Scott, the film critic for The New York Times, to write a review or essay in Draftback. “He’s a beautiful prose stylist (diction, cadence, etc.), his writing is accessible and unpretentious but world-class, and he seems to always put his finger on the essence of whatever it is he’s talking about.” Somers is curious about whether all that comes naturally to Scott.

But if not Scott, he’ll take any sensational writer. “I am not going to let a tiger go by the tail until somebody really, really great writes something important and then people can break it down. Because that’s going to be an artifact that’s valuable to every single high school teacher, high school student, college teacher, every literate person in the world.”

First of all, not any sensational writer would do. Scott would work perfectly not just because of his talent, but his brevity. Playing back a review is much more consumable than a chapter, for example, let alone an entire book. It’s a wholly unwieldy tool for long-form; not a problem, just important to recognize.

The real problem here lies in “a review or essay,” singular.

Yes, getting a master’s work would be huge for Draftback. But I dispute the idea that whatever piece of text an A.O.-esque scribe offers would certainly be representative of his or her general writing process. Between concrete factors like the specific material of the piece in question and abstract ones like the writer’s mood while working that day, one essay couldn’t prove any of the techniques that made it are the stuff of greatness.

Somers said he’s interested in allowing writers to annotate Draftback to explain the choices it shows in action. A.O. Scott’s thoughts on A.O. Scott would be valuable, but what would really help Somers would be more A.O. Scott articles.

If we’re looking at writing on data’s terms, we need to strive for statistical significance. The first major essay will be a major step towards credibility for Draftback, but it can’t be Somers’ literary artifact on reputation alone. It will take analysis of a body of work to determine the trends of a given writer.

Draftback doesn’t have the capability to compare and contrast writing processes from different documents—and asking the app to synthesize and convey that information in a coherent way would be a gargantuan endeavor compared to the simple utilization of a Google Doc’s saved keystrokes.

Short of computers unlocking the keys to great writing and spitting out the answers back at us, Draftback needs scholarly attention, close examination to determine the vocabulary of real-time edits, to see what consistently works and what doesn’t as writing gets gradually revised, before it can be of educational use to every literate person in the world.

Then there’s another roadblock: cognitive bias resulting from writing for Draftback, commissioned by Somers.

It’s the basic idea that writing with the knowledge in mind that people will use what you produce to study your process will have an affect on the way in which you write. Maybe you write more passively, more floridly, more precisely, more broadly. Regardless, there’s deviation from the mean.

Or maybe you can successfully write for Draftback without cognitive bias, which would make you an outlier. Perhaps that could say something, how people who maintain their process without regard for who would be observing their writing write. But that’s not the same as how masters write.

The simple workaround here would be to find writers who have already been doing their writing on Google Docs and examining their back catalogs; that could potentially solve the sample size problem at the same time. Then again, Google Docs is a relative newcomer in terms of prominent word processors. Assuming the demographic using a more novel online technology skews younger, Somers’ class of established greats is less likely to have extensive Google Doc offerings already written.

To get enough Google Docs and to study them enough to come to any substantial conclusions would take years. And even then, it’s not at all guaranteed Draftback will yield any illuminating discoveries. Draftback is a tool with the potential to teach us unprecedented things about how we write, but at this point in time, that’s less a destiny and more a hypothesis. Time may ultimately prove it false.


Josh Cohen is a contributing editor for MobyLives.