An uninspired piece at the New York Times on machine-generated books by Pagan Kennedy. It should come as no surprise that an author who makes money writing for a living would disapprove of the idea. Actually the books are just quickly mashed-up wikipedia articles, so there is very little machine in these machine-generated products.
Of course, there is a long history to the fear (and fascination) of books that write themselves. Whether it was eighteenth-century automatons that were able to write letters, Victorian creations of autotelegraphy, or modernist dreams of unconscious creativity, automatic writing has a long past. Today’s band of writers at work on the relationship between algorithm and creativity is just the latest incarnation. Kenneth Goldsmith’s Uncreative Writing might be thought of as the movement’s handbook.
My personal vision is to figure out how to “write” novels algorithmically generated from the Times best-seller list. This way you could have something like the statistical aggregate of readers’ median taste. There was a delightful experiment in this direction in the field of painting by Vitaly Komar and Alexander Melamid. They produced a painting, so they claimed, that was going to be the statistically most preferable painting based on thousands of museum goers’ replies.
As far as Kennedy’s lamentations over the end of human writers is concerned, I’m never sure where the fear comes from — why worry about machine-generated text? Authors are always worrying about someone or something about to take away their livelihood (if you can call it that). But I’m also never sure why things have to boil down to such stark either/or’s — machine v. human. As I’m writing in my new book, having to model human practices in a computational environment is an excellent method of learning what it means to be human. Thinking through how to generate text “like a person” is a way of thinking through “mindedness” in a deeply analytical way.
The real significance of machine-generation will likely have very little to do with the actual generation of the texts. Sure, it would be great if I could algorithmically write novels — if each sold for $0.99, and if it took a few hours to “write” each one, and if people actually enjoyed them, well, what a great way to make money (sort of). But that’s a lot of if’s.
No, I think the significance will be in the way knowledge about machine-generated texts will help us create better machine reading tools. And these will be indispensable as we move into the future of more and more text. Having tools that reliably sort through more writing than we can by hand will be of value to many audiences of various stripes — doctors trying to survey medical literature, stock-brokers trying to understand the relationship between social communication and shifts in valuation, corporations trying to pick our brains, and last, but not least, humanists who wish to know more about the bast trove of our textual heritage.
And yet the most tantalizing thought of them all: computers buying novels written by computers. What a market that would be!