Every once in a while you find an article that seems just for you. John Seabrook writes in The New Yorker about the group OpenAI and their A.I. called GPT-2, which is so advanced that they are keeping it under lock and key. Not only is the whole thing about the bleeding edge of writing writers, it’s also a wonderful example of a digital article. It also appears in print, of course, but the online presentation is so subtle yet inventive, with its interactions to reveal AI-written text and a mini-game of “spot the AI” within a paragraph, that I cannot imagine the print version competing with the web one.
I do wish Seabrook had talked to Robin Sloan about his experiments in writing fiction using the same A.I. It might have encouraged him to explore the idea of A.I. as a potential partner or generator of raw material to be re-shaped.
On occasion, the two women went to lunch and she came home offended by some pettiness. And he would say, “Why do this to yourself?” He wanted to keep her from being hurt. He also wanted his wife and her friend to drift apart so that he never had to sit through another dinner party with the friend and her husband. But after a few months the rift would inevitably heal and the friendship return to good standing. He couldn’t blame her. They went back a long way and you get only so many old friends.
He leaped four hours ahead of himself. He ruminated on the evening in future retrospect and recalled every gesture, every word. He walked back to the kitchen and stood with a new drink in front of the fridge, out of the way. “I can’t do it,” he said.
Did you catch that? A new drink. Ferris could have had another paragraph or two there, with beautiful and clever language explaining that our narrator had started drinking two hours ago, was on his third, and liked to pair his dry reds with cutting loquaciousness. Continue reading “How not to say something”→
Robin Sloan, novelist, media inventor, olive oil entrepreneur:
Imagine a sentence. “I went looking for adventure.”
Imagine another one. “I never returned.”
Now imagine a sentence gradient between them—not a story, but a smooth interpolation of meaning. This is a weird thing to ask for! I’d never even bothered to imagine an interpolation between sentences before encountering the idea in a recent academic paper. But as soon as I did, I found it captivating, both for the thing itself—a sentence… gradient?—and for the larger artifact it suggested: a dense cloud of sentences, all related; a space you might navigate and explore.
My project called sentencespace, now public on GitHub, serves up an API that provides two things.
Sentence gradients: smooth interpolations between two input sentences.
Sentence neighborhoods: clouds of alternative sentences closely related to an input sentence.
Sentence neighborhoods are simpler than gradients. Given an input sentence, what if we imagine ourselves standing at its location in sentence space, peering around, jotting down some of the other sentences we see nearby?
In a couple previous posts, I wondered about what it means for humans’ role in the creative process when computers begin generating texts. Will we be promoted to editors and curators, or be out of a job?
Ross Goodwin provides a different metaphor:
I would have been more nervous about sharing the machine’s poetic output in front of so many people, but the poetry had already passed what was, in my opinion, a more genuine test of its integrity: a small reading at a library in Brooklyn alongside traditional poets.
Earlier in February, I was invited to share some work at the Leonard Library in Williamsburg. The theme of the evening’s event was love and romance, so I generated several poems [1,2] from images I considered romantic. My reading was met with overwhelming approval from the other poets at the event, one of whom said that the poem I had generated from the iconic Times Square V-J Day kiss photograph by Alfred Eisenstaedt “messed [him] up” as it seemed to contain a plausible description of a flashback from the man’s perspective.
I had been worried because, as I once heard Allison Parrish say, so much commentary about computational creative writing focuses on computers replacing humans—but as anyone who has worked with computers and language knows, that perspective (which Allison summarized as “Now they’re even taking the poet’s job!”) is highly uninformed.
When we teach computers to write, the computers don’t replace us any more than pianos replace pianists—in a certain way, they become our pens, and we become more than writers. We become writers of writers.
“Writers of writers.”
Anybody interested in machine-generated text should read Goodwin’s “Adventures in Narrated Reality”: Part 1, Part 2.
I am thinking of a writing machine that would bring to the page all those things that we are accustomed to consider as the most jealously guarded attributes of our psychological life, of our daily experience, our unpredictable changes of mood and inner elations, despairs and moments of illumination. What are these if not so many linguistic “fields,” for which we might well succeed in establishing the vocabulary, grammar, syntax, and properties of permutation?
But one consequence is common in Goldsmith’s and Calvino’s visions: the death of the traditional author. Calvino again:
Literature as I knew it was a constant series of attempts to make one word stay put after another by following certain definite rules; or, more often, rules that were neither definite nor definable, but that might be extracted from a series of examples, or rules made up for the occasion—that is to say, derived from the rules followed by other writers. […] A writing machine that has been fed an instruction appropriate to the case could also devise an exact and unmistakable “personality” of an author, or else it could be adjusted in such a way as to evolve or change “personality” with each work it composes. Writers, as they have always been up to now, are already writing machines; or at least they are when things are going well.
Calvino is content with his thought experiment: he says that “it would not be worth the trouble of constructing such a complicated machine” as he describes. But one wonders whether artificial intelligence might progress to such a place. We are already trying to write under the influence of data. People decry that as forcing the writer into a well-traveled rut, but add a little AI and writing might become more adventurous:
…given that developments in cybernetics lean toward machines capable of learning, of changing their own programs, of developing their own sensibilities and their own needs, nothing prevents us from foreseeing a literature machine that at a certain point feels unsatisfied with its own traditionalism and starts to propose new ways of writing, turning its own codes completely upside down.
Calvino perhaps understates the difficultly in getting machines to ‘develop their own sensibilities’ (he was writing in 1967) but still—imagine what might be added to Goldsmith’s vision of fashioning new art from existing literary material if that material began to have even a rudimentary mind of its own, if it danced with us a little bit, like sculpting with living clay.
I’ve been reading Uncreative Writing by writer and UbuWeb founder Kenneth Goldsmith, who thinks that writing is ripe for a revolution. He thinks of language not solely as semantic content but also as raw material — material that can be transformed by computers, or written from scratch by computers, sometimes even meant to be read only by other computers. In effect, writers get abstracted (promoted?) a level, from generating language to managing its creation and manipulation.
It has me wondering what this would mean for stories. Goldsmith’s narrative examples involve the appropriation of existing narrative material. What would it look like to not make up a story but to manage a machine generating a story?
The only example I’ve found is A Ship Adrift. Perfectly, the credits say that “a ship adrift is a thing “by” james bridle.” Bridle explains on his blog that he began by creating a system to read information from a weather station:
A Ship Adrift takes the data from that weather station and applies it to an imaginary airship piloted by a lost, mad AI autopilot. The ship is drifting because the pilot is mad or the pilot is mad because the ship is drifting; it doesn’t really matter.
If the wind whips eastwards across the roof of the Southbank centre at 5mph, then the Ship Adrift floats five miles to the East. […]
As the Ship drifts, it looks around itself. It doesn’t know where it is, but it is listening. It’s listening out for tweets and foursquare check-ins and posts on dating sites and geotagged Wikipedia articles and it is remembering them and it is trying to make something out of them. It is trying to understand.
The ship is lost, and I don’t know where it’s going. I don’t know what it’s going to learn, but I want to work with it to tell some stories. I want to build a system for cooperating with software and chance.
The result is, to my eye, gibberish. But it’s gibberish on a timeline, written by hundreds of people and amalgamated through a partnership of man and machine. And if you look at the text in the context of the wandering little dot representing the ship, it’s even a little poignant. Perhaps the beginning of something.