“Robots that write fiction? You couldn’t make it up.” The Guardian headline plays on a trope that, at its core, is rather insulting to fiction writers, who have a pretty good track record of making up wild things -- concepts far more imaginative than stories composed by AI.
Putting that aside, James Bridle’s breezy blog in The Guardian looks ahead eagerly to a time when fiction will no longer be the exclusive realm of living authors. “Robot writers could become co-authors of our most complex subjects, helping to write the narratives of climate change and political upheaval,” he breathlessly speculates, without specifying why robots would excel at writing about these topics or how they would collaborate with humans, as he implies. Still, could they?
Well, perhaps, the way that “soothing mothers” might “give up their babies, plot bank robberies and become threatening bank robbers” -- just one scenario generated by the Metaphor Magnet, a fiction-writing robot, for an earlier Guardian piece. It’s not impossible, but there’s no particular evidence that it will happen that way.
Lest we get too eager, there are still significant hurdles for machines to clear before the day they make novelists superfluous.
Take the infusion of personal sense memories or emotions into a text. “[Fiction’s] magic derives as much from the writer’s own lived experience -- emotional, sensory or otherwise – as from their creativity,” wrote Hephzibah Anderson for the BBC earlier this year. Partly because truly great fiction involves reconceptualizing common experiences in a way that allows us to perceive them in a new or startling way, this poses a particular challenge for artificial intelligence, which can log data about human experience and learn to mimic it, but can’t actually have the same sensations and emotions.
What about jokes, irony, and figurative speech? Researchers have tackled the challenge of making robots produce humor and metaphor. In The Guardian, Tom Meltzer pointed to the previously referenced Metaphor Magnet, a program that produces witty ironies, though the examples he gives suggest “witty” is a highly subjective quality. “When the sandals that are worn by humble monks are worn by the smuggest hipsters” might contain a somewhat unexpected juxtaposition, but it still seems to lack the humorous essence needed to elicit a chuckle.
Metaphor Magnet's Twitter bot spits out various such aphorisms, most of which are funny because they are completely vacuous (“When it comes to the followers they command, some princes can be far from caring and can be downright brutal”) or completely nonsensical (“Coachman. noun. A fraud who would rather operate unbelievably fake coaches than unbelievably fake scams”).
Of a short story generated by the Metaphor Magnet, Nicholas Lezard wrote: “Had these been presented to me as an early experimental work by, say, Ben Marcus, I may have been fooled” -- in large part, he admits, because certain conceptual authors might purposely try to mimic the stiff alien tone of a robot -- “had it not been for the general sense of ineptitude.” Basically, he could almost have believed it was a bad story by an author trying to sound like a robot, if it weren’t just so dreadful.
The What-If Machine tosses out intriguing hypotheticals, of the sort that might spark a full work of fiction; if a writer wants to loosen up by checking out some totally crazy computer-generated scenarios, WHIM is a pretty solid bet to get the creative juices flowing.
That said, these scenarios don’t really make up stories, and most sound stilted, trite or absurd. “What if there was a camel who woke up in a porch, and had become a dog that was able to see a colour?” “What if there was an old pen, who couldn't write anymore, which he used to do for pleasure, so decided instead to eat a cookie?” “What if blessed immortals were to give up their immortality, be cast out of heaven and become cursed demons?” (Hey, I think a robot just wrote Paradise Lost!)
The Daily Dot listed a few that they thought “could certainly lead to an interesting story," including “What if there was a little lawyer who forgot how to destroy your enemy?” a scenario that sounds like it was dreamed up by someone barely tethered to reality. There’s intrigue in the oddity of the concepts, but the draw is bound up in knowing a person thought of the idea and had a reason to develop it into a story. Otherwise, this sort of hypothetical is just amusing word salad.
Scheherezade, an algorithm that, according to New Scientist, crowdsources strings of actions that could possibly follow each other, then creates stories centered around these actions, satisfies the logical demands of fiction a bit better, but at the cost of interest. Here’s an excerpt from a story quoted by New Scientist:
Sally got scared when John approached because he looked suspicious. John pulled out a handgun that was concealed in his jacket pocket. John wore a stern stare as he pointed the gun at Sally. Sally was very scared and screamed out of fear for her life. In a rough, coarse voice, John demanded the money.
The bot tossed in applicable emotions, actions, and adjectives to a bank robbery sequence, but because the bot can’t really understand what is and isn’t relevant outside of rather specific parameters, the window dressing added is so uniformly relevant it’s relentlessly dull. There’s no reason to read this story rather than any other story about a bank robbery in existence.
“At its best, Scheherazade writes fairly convincing vignettes,” wrote Meltzer of the program. Even its creator, computer scientist Mark Riedl, told him, “For Scheherazade, a successful story is one in which people will read the story and recognize the activity and not find too many obvious errors.” That’s not exactly a successful piece of longform journalism, let alone a successful piece of fiction. Down the line, such a program could conceivably replicate bottom-bucket, rigidly genre-dependent pulp that already demands little by way of polished prose or narrative imagination from its writers, but why do we need robots to make bad fiction?
The deeper problem with all of these attempts to make a real robot literary fiction author won’t be solved, not by any of these algorithms, at least. Maybe we can be fooled by a particularly well-tooled program, in time. But is that art? Literary fiction relies on a bond of trust between the author and its readership -- we trust that the author has chosen words to convey considered meaning to us, and the author trusts that we’ll read generously, looking to understand that meaning.
If the words have been chosen by algorithm to resemble the words a human might choose, that bond no longer applies. Instead of the highest form of human expression, a finely crafted vessel for a person’s thoughts and experiences, a literary novel composed by a robot would comprise nothing but a shell. The shell might glitter appealingly, but there would be nothing inside.