October 16, 2014

Computers can write books now, but we probably don’t have anything to worry about


Editorial department of the future. Image via Wikipedia.

Editorial department of the future. Image via Wikipedia.

Authors United asserted a few weeks ago that “books cannot be written more cheaply”. This is because books are written by authors, and though publishers may place restrictions on their authors that cap the amount of time and money the publisher intends to invest, authors require the same resources as any human (and, often, more) to produce publishable work. Slash the production, marketing, publicity, and editorial budget to the bone and you still have the inelastic cost of the author to contend with—that is, unless Amazon plans to lower consumer costs by doing away with authors.

Out of the question, right? Well, not entirely. The idea that the human element of the creative process can be partially or completely removed has fascinated people for decades—and in honor of Ada’s Algorithm, a lovely book about the history of computers written by the very human James Essinger, I dug into the weird history of this idea.

If Amazon succeeds in fashioning a virtual house of fully digital authors, it would likely resemble the titular device in Roald Dahl’s short story “The Great Automatic Grammatizator”, written in 1954. The story is subtly terrifying (especially if you read it when you’re nine years old, as I did.) The main character, Knipe, is a disgruntled aspiring author and engineer, whose company produces a cutting edge “computing engine”. Smarting from a series of rejections by prominent magazines, Knipe comes up with the idea of programming the computer to write short stories, and after a series of beta tests, he successfully completes an algorithm that can do just that. The machine’s stories prove to be so good that Knipe programs it to write novels, and they’re also so damn good that soon Knipe convinces prominent authors to stop producing their own work and simply license their name to books produced by the machine. This is so successful that by the end of the story, the Grammatizator is producing almost all published fiction, while authors who refuse to sign over their names live in penury—including, in a final-page twist, the author of the story itself.

Chilling, but one can point to a number of examples of books written by more developed (and more boringly-named) versions of this machine—algorithmic content creation is very much real. Vonnegut made up a computer that wrote love poems in 1950—and the Brits did the same thing, but IRL. The reason why you don’t have a bunch of computer-written books on your shelf is because they’re traditionally looked on as novelties. A few years back a Russian computer wrote a Tolstoy homage in the style of Murakami, but unless McSweeney’s has hired this program on the sly, this is the first and last we’ve heard of it. And this isn’t new—over thirty years ago, a program called “Racter” allegedly composed an entire book called The Policeman’s Beard Is Half Constructed. A decade later, another programmer and his creation composed a apparently-not-bad Jacqueline Susann knockoff. Seven years later, someone managed to create the automated equivalent of a tiresome MFA student. (Wait a second…) The tone of these articles is always the same; amused, but dismissive.

They point out a crucial difference (and journalists have every right to be snarky about the idea that they can be replaced); algorithms are fairly good at making and collating contentbut not literatureThe Associated Press and Forbes uses bots to author articles; Penguin Random House doesn’t have the same option. (Though I do have a very convincing theory that James Michener was in fact a clockwork automaton.)  Which brings us to the story of Philip Parker, who created a program that’s effectively allowed him to “write” over 100,000 books—granted, they’re books that nobody would ever buy, esoteric (and expensive) market research and industry study titles like The 2007-2012 World Outlook for Wood Toilet Seats. The program is a content compiler rather than a composer – though Parker claims to be able to write poetry and fiction with it – and Parker has posed it as a crucial element in getting textbooks and other types of educational content to poor areas, all because it cuts out the author (“Authors are expensive. They want food and things like that! They want income!”).

Much like you drunkenly speculate that tobacco companies have secret fields of marijuana-ready soil just WAITING for the law to turn in their favor, I speculate that Amazon has a dedicated department (or at least a few wonks) who have pored over Parker’s work, slowly gaming out the possibility of a wholly-digital author. That’s not so hard to imagine, is it? We’ve already seen a chatterbot pass the Turing test; the dialogue in Tom Wolfe’s last couple books can’t make that claim. Smashwords founder Mark Coker speculated on this last year. However, Margaret Sarlej, an Australian scientist who programmed a fable-producing program at around the same time, has a more restrained view.

“I don’t think computational storytelling systems should necessarily aim to replace human authors and produce literary masterpieces, but rather serve as a tool for developing new ways of experiencing story. The possibilities they offer for interactivity open up a wealth of opportunities that traditional authors have probably never even considered.”

So maybe we’ll be seeing computer-written books enter into the pantheon of formats (cue the doomsayers, “the industry is dying”, etc etc). But it’s clear that the cost of developing programs that write books is prohibitive, and unless the books being written are sure to be bought – educational materials are a sure bet, and perhaps so are extremely formulaic genre exercises – there will be no incentive to do so within the publishing industry, no matter how large your market share is. And developers are expected to speak in vague but lofty terms about the future of their creations – that doesn’t mean they should be believed.

Then again, if there was a compositional algorithm that could also calculate the odds that its own output would sell, and a company large enough to either invest in its development or simply buy it outright…*shakes head, mutters, shuts down computer, goes home*



Liam O'Brien is the Sales & Marketing Manager at Melville House, and a former bookseller.