The Senate's detailed report on the excesses and ineffectuality of the CIA's "enhanced interrogation techniques" has dominated the news cycle. Rightfully so; the document was a gut-churning reminder of American brutality just past—a slippage towards institutionalized barbarism that most of us are still unsure how to process. Maybe a good novel would help; if only humans could write that fast.
Machines can. By coincidence, I trust, this same miserable week was the one I came upon the NYU communication scholar Ross Goodwin's Fiction Generator, by way of Alexis Madrigal's indispensable "5 Intriguing Things."
It's an algorithmic novel builder, essentially; feed it keywords, characters, and your dramatic preferences and it spits out a ready-made novel. I asked if Goodwin would harness his robot to generate a story for our speculative fiction project, Terraform—he suggested we "write" a novel about the torture report.
So he did. By taking the complete text of the 500-page executive summary of the CIA torture report, factoring in tens of thousands of names from the US Census and Social Security records, and unleashing them on his program, he created a 728-page "novel" about one of the darkest chapters in recent American history.
Here's a sample:
Lacey was on a liquid diet quite appropriate because Lacey was recovering from abdominal surgery at the time…
The CIA continued to use Fransisco's enhanced interrogation techniques against Abu Zubaydah was also subjected to the United States, so far.. ..He did vomit a couple of times during the water board with some beans and rice. It's was 10 hours since Lacey ate so this was surprising and disturbing. Mary plan to only feed Ensure for a while now. Im head[ing] back for another water board with some beans and rice.
"I inserted character names (in the place of personal pronouns, 'CIA,' and 'detainee'—based on word frequency analysis) and changed verbs to past tense using the algorithm from my fiction generator," Goodwin explains in an email. "I then took the altered segments and generated new text using an Ngram language model."
The result is a stew of the banal and the horrific. It hardly resembles anything you'd consider literary; it is, as Goodwin notes, "about two steps above complete gibberish." But when I read the lines aloud to a table of friends, it elicited protests, smirks, raised eyebrows, pained sighs, and, importantly, a perplexed empathy; a genuinely new way to try to make sense of the horror.
Maybe it was enough to see American-sounding names and everyday nouns alongside the stark, clinical descriptions of torture—Mary, waterboarding, beans and rice—to pull our jumble of impressions of enhanced interrogation outside the walls of Guantanamo. Maybe it actually forced us to read the torture report, or to consider, perversely, how, why, and if the source material made sense before the "fictional" version did. (It's the same reason that this workhorse press printed the report out in the form of a book, perhaps, though it renders an entirely different impact.)
It was useful, to a point, and certainly a provocative glimpse of the kind of tool that may be telling us all sorts of automated stories before long. That was Goodwin's aim in building the Generator, after all.
"At this point, I'm more interested in augmenting human creativity than granting machines emotive capabilities," he says. "Beyond creating a readable novel with assistance from a computer, it's about granting people the process of generating their own personalized novels." In this case, a version of the CIA report.
The blurring of news and fiction is always inevitable, and in this case, productively disturbing.
"Ultimately, there's a fine line between making a fiction generator and a plagiarism machine," Goodwin says. "I try to stay firmly on the fiction generator side of that line."
Even so, it seems to me his machine just helped spit out something closer to the truth.