The rise of robot writers
It’s early morning in March, 2018, Brooklyn. The ruddy hue of the brownstone-clad buildings are a dead giveaway.
A black Cadillac is parked on a sidestreet; passers-by regard it with suspicion. A conspicuous surveillance camera is mounted crudely on the trunk. Wires loop through the passenger door and feed into a laptop on the backseat, which, in turn, is hooked to a ragtag receipt printer.
As the driver pulls away from the curb, the printer chugs into action, spitting out words onto the first scroll of paper. It was nine-seventeen in the morning, and the house was heavy.
This, believe it or not, is the opening line of a novel, though not one penned by a human. They’re the words of a bot. An AI machine designed and built by Ross Goodwin, a ‘writer of writers’ and who now works for Google as a creative technologist.
Goodwin’s aim was simple: to create a computer that could write a novel in the style of Kerouac’s On The Road, based on the data it pulled from a camera, a GPS, a microphone and his laptop’s internal clock. Over the next four days, he chauffeured this DIY contraption across the US, filling the backseat and eventually the floor of his rented Cadillac with manuscript pages.
Confused? You’re not the only one. It is pretty baffling to think artificial intelligence might be capable of writing a novel – especially a breed whose aesthetics are more R2-D2 than Ex Machina.
But artificial intelligence has gone above and beyond what was predicted of it in the 1940s. It's extracurricular skills alone are enough to make us green-eyed with envy: it can create visual art, write poetry, compose songs, write social media posts, tell jokes and carry civilised conversations (for the most part).
Automated storytelling in itself isn’t a new concept. In fact, the technology behind it is already used by the likes of Forbes, The Washington Post and ad agencies to create commercial content. And if you think these are nothing more than formulaic sentences, think again. According to a trial run by JP Morgan, AI ads performed better, with a higher clickthrough rate than those written by humans.
That’s probably why AI received $1bn in investment in the UK alone last year, and why companies have built their own in-house AI solutions, like Bertie and Heliograf. They’re the new breadwinners.
AI uses natural language generation (NLG) to produce written content. Humans feed it data and context, and the computer generates a narrative based on these using its prescribed conditional logic. This premise also applies, in a basic sense, to AI literature. No typical chatbot is able to produce lyrical content on its own. To do so would require self-awareness, human depth and the ability to tap into human emotions from a pool of experience.
Goodwin’s computer didn’t have self-awareness, but it did have an LSTM neural network, which was trained with decades-worth of books and poetry (amounting to roughly 60 million words) representing the same styles which he wanted to replicate. At this point, and especially if you’re a writer like myself, you might want to know what’s stopping you from spreading all the AI hate propaganda you can get your mitts on. You’re out of a job, right?
Well, no. Although Jean Boîte Éditions published 1 The Road, people didn’t swarm to their nearest bookstores in eager anticipation of its release. The response was lukewarm, to say the least. I was curious to see what a bot experiences on an all-American road trip. The answer is, not much.
It was seven minutes to ten o’clock in the morning, and it was the only good thing that had happened.
Thirty-six minutes in and it’s already unimpressed. But what’s the good thing? I have no idea. It continues:
The time was three minutes to ten o’clock in the morning, and the conversation was finished while the same interview was over.
What conversation? Where’s the dialogue, bot? In places, the prose sounds more like something Alexa would sprout if asked to describe a Dali painting:
The time was ten forty-seven in the morning, and the picnic showed a past that already had hair from the side of the track somewhere in the middle of the room.
In fairness to Goodwin, the lack of literary splendour isn’t what’s important. Neither is the computer’s singular ability to process conversations which the mic picked up, or its decoding photographs into prose. It’s how the computer chose to express these as a whole, using the car as a (literal) vehicle to creation.
While philosophers might say that a computer’s choice, no matter how arbitrary, points to some semblance of conscience, it boils down to how you define choice. Yes, AI can make decisions, but they’re limited by the conditional logic a human has built for it. We’re only hit with an existential crisis if and when an AI can learn to outgrow this conditional logic. My thoughts can’t help but turn to the countless times I’ve asked Alexa a question, only to be met with stony silence. (It’s not personal, my boyfriend assures me. It’s just a glitch.)
In the good old days, long before Alexa, back when AI wasn’t even really a buzzword in business, a computer wrote a novel by the name of True Love, later published in St. Petersburg. Researchers working on the project claimed the book, 254 pages in length, was a variation of Tolstoy’s Anna Karenina. It’s hard to believe, unless Tolstoy had originally been planning to set his epic on a desert island where the characters have no memory of who they are or their relationship to one another.
The AI was fed roughly 17 works of modern Russian literature and a book by Haruki Murakami (whose style the AI mimics). Everything it knew, it knew from books. True Love begins in true Russian style:
“There's nothing else here but the bloody sea and the bloody rocks… And it is in such a drab place that I am going to kill you," the woman muttered.
For someone who has no memory, she’s pretty aggressive. The novel continues:
Kitty couldn’t fall asleep for a long time. Her nerves were strained as two tight strings, and even a glass of hot wine, that Vronsky made her drink, did not help her. Lying in bed she kept going over and over that monstrous scene at the meadow.
Not really sure where Vronsky managed to get hot wine on an island, or the bed for that matter, but I’ll go along with it. Suspension of belief is a human trait, after all.
The setting sun was painting pink the underbellies of the clouds hanging low above the grey sea. White caps could be seen here and there, but it was obvious that the storm he had been expecting all day was not going to happen.
Sure, the syntax is clumsy, the descriptions a bit hit-and-miss, but it’s far more human-like than the Kerouac approximation. How would it fare under the Turing test?
In 2016, the judges of Japan’s Nikkei Hoshi Shinichi Literary Award were dead-set on carrying out their own version of the Turing test, and so they welcomed both humans and machines to apply with a piece of writing. To even the playing field, all submissions were kept anonymous.
The majority of AI submissions bombed the first trial. There was only one that made it through to the next round -- The Day a Computer Writes a Novel. I think you can probably guess who the main character is. Like Goodwin’s bot, it’s unimpressed by its surroundings.
Gotta find something fun. At this rate, if my inability to attain a sense of fulfillment were to continue, it felt like I could shut down at any moment. Enter Miss Yoko, human party pooper.
If Miss Yoko would just leave for once then maybe I could sing a song or something, but right now I couldn't even do that. Unable to move, unable to make a peep… it became necessary to entertain myself in such a position.
Well, why don’t I try to write a novel? The moment the thought came to mind, I opened up a new file, and entered the first byte.
I entered 6 additional bytes.
0, 1, 1
I couldn’t stop myself.
0, 1, 1, 0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233, 377, 610, 987, 1597, 2584, 4181, 6765, 10946…
This goes on for a while.
Still, it did fool the judges. At first. The piece didn’t make it through the second round, though personally, I think the bot’s last words might have dealt the final blow.
The day a computer wrote a novel. The computer, placing priority on the pursuit of its own joy, stopped working for humans.
I take comfort in the figures, because the team behind this AI program admitted to doing 80% of the overall work, having defined everything from plot to characters to outcome. Only then did the AI show up, spat out a novel, and took all the glory.
There’s no doubt that AI is developing at a breakneck pace; in one world survey, experts concluded that by 2040, machines will be capable of equalling us in understanding.
A scary idea for some. Others, like mathematician and author Marcus Sautoy, take on a more optimistic approach. He sees AI as a tool to further human creativity, rather than a threat.
Apps like Notebook.ai and Granthika help writers unload the grunt work of novel writing, by keeping track of a novel’s storylines, plot points, facts and chronological details in one place.
Other AI software like Magic Realism Bot churns out bite-sized story ideas every four hours in a bizarre homage to great authors of the genre, making for some interesting writing prompts.
AI is learning fast, sure, but we shouldn’t shy away from it; if anything, we should keep ahead of the curve by seeing where it takes us, using it to our advantage. Or, if that fails, we could take a leaf out of this robot’s book:
You know what: I would have been very scared of what’s to come had I not been so drunk.