Engineering

Your next creative partner could be a bot

A tech-savvy science-fiction author shows how AI might inspire art rather than replace it

person and spirit
Image Credit: Matt Huynh

When machines are able to do anything humans can — be it farming, factory-working, or stock-trading — they’ll do it in a fraction of the time, experts reason, eventually rendering the human workforce redundant. Even creative fields aren’t safe: artificial intelligence has already begun making its mark in journalism, fine art, and music. That’s enough to unsettle any artist. How could something so fundamentally subjective, so human, as art be mechanized?

Robin Sloan isn’t worried. In fact, the Oakland-based novelist and self-described “tech-adjacent media inventor” is so unbothered by the idea, he’s already invited robots into his workplace.

 

Robin Sloan

 

“Playing with [the program] actually has that feeling [of having a writing partner], except that your partner is truly, deeply, strange,” says Sloan. “They have very specific strengths — not succumbing to clichés, creating stunningly beautiful turns of phrase — and very specific weaknesses.”

Earlier this year, Sloan, a former Twitter employee, pieced together a program powered by basic applied AI to co-write short stories with him. His machine — or “AI sidekick,” as he describes it — combines a recurrent neural network (or RNN, a basic machine learning structure) with a massive database (about 150mb, or 78.6 million words) of old science-fiction magazines to spit out autocomplete suggestions as he types out a story in the machine’s text editor. (Think of it as a more complex version of that widget from a few years back that mimicked your Facebook statuses.)

“Playing with [the program] actually has that feeling [of having a writing partner], except that your partner is truly, deeply, strange,” says Sloan. “They have very specific strengths — not succumbing to clichés, creating stunningly beautiful turns of phrase — and very specific weaknesses.”

To be fair to his writing partner, Sloan’s source text is science-fiction, so nonsense words like “goathemaker” are more common than they would be if his machine was drawing from, say, the National Geographic archives. His machine is also learning letter by letter (as opposed to word-by-word models like Google’s search suggestions), which allows it to make up its own words.

 

 

 

Apart from that, however, the weaknesses to which he refers — like the program’s inherent inability to understand larger literary context and give related suggestions, for example — are temporary. Give it a couple of decades, and robo-co-authors of the future could evolve from parlor-trick gadget to essential writing tool, once innovators finally create general artificial intelligence, or at least develop machines that can understand context rather than just rules.

“There’s a difference between the [current] concept and the promise of a tool like this,” Sloan explains. “With the actual capabilities of these neural networks right now, it’s not useful for anything. It’s not sending business emails, or writing e-commerce copy. It’s a weird, creative input, a way to get out of a rut, a way to inject some cool, interesting language into your writing.”

His own robo-partner, he says, is currently being used to produce the dialogue for a single character in one of his short stories, in order to give their speech a purely alien aesthetic.

 

“For the hardcore researchers making these amazing strides in this field, [creative application] isn’t really something they consider, and probably for good reason,” he says. “They can’t publish a paper that says, ‘Personally, I think this is freaking awesome.’ They have to be able to use objective measures, and those are just not creatively interesting.”

 

When applied AI — that is, the current state of the art, in which an intelligent machine is built to do a specific thing rather than built with the ability to learn anything — becomes more complicated, however, or gives way entirely to general intelligence, a robo-scribe could easily create something more akin to actual human language, rather than regurgitating quirky approximations of it.

Still, that doesn’t bother Sloan. First, he explains, there’s very little incentive for AI scientists to invest much time in developing creative widgets like this.

“For the hardcore researchers making these amazing strides in this field, [creative application] isn’t really something they consider, and probably for good reason,” he says. “They can’t publish a paper that says, ‘Personally, I think this is freaking awesome.’ They have to be able to use objective measures, and those are just not creatively interesting.”

But more importantly, when people think about AI impacting the workforce, they usually imagine a program doing all the work in our place, rather than the more likely reality: that we’ll still be creating, and AI will simply facilitate and transform it.

“The question becomes: What, specifically, does the world of writing and literature need? It always needs interesting tensions,” and a cyborg assistant can offer that, he says. (Imagine having one to help out when you’ve got a word on the tip of your tongue.) “It doesn’t necessarily have to be, ‘The AI is writing.’ It could make all sorts of different transformations [to our work] that are just not available yet.”

AI, in other words, is no more a threat to creative fields than the printing press or word processor: initially seen as blasphemous or cheating, then eventually becoming part and parcel of the creative landscape — or even transforming it altogether.

“Writing changes. It’s not some mystical, age-old art,” he says. “[It’s] totally a function of technology and context. It’s hard to know the ways in which technology changes how we write. The effects can be really subtle. It changes what we write, too.”

 

 

Here’s the full text of the story that Sloan wrote for Slack with the help of his bot:

The first robot I ever owned was a refined brain that could speak nine languages and recommend definite changes to my life. Basically, it was a coach. I needed one. At that time, I was listless, directionless. I spent my days breakfasting, staring up at the dome, admiring the stars of the maze.

The robot changed all that. It suggested, first, that I get a job, so I did. I became a professional reader, writing summaries for the maze-lords of popular novels. Then, the robot told me to buy some better clothes. I did that, too; I looked better in dark tones. Finally, and somewhat gingerly, the robot asked: “Do you think perhaps you should not be able to control your own private problem?”

I gasped. For all our intimacy, I thought I had kept that a secret from the robot. I said, “I can’t control it. You know that.”

The robot said, “Seth. I’ll do the job for you.”

Which made sense, after all. The robot reached into my brain — my own robot brain — and made the tiniest changes. Do you know what they were?

Neither do I anymore. I go to my job and read my novels. I hardly ever look up at the dome anymore.

 

Devon Maloney could use an AI assistant. Her dog’s suggestions are crap.

 

Slack is the collaboration hub, where the right people are always in the loop and key information is always at their fingertips. Teamwork in Slack happens in channels — searchable conversations that keep work organized and teams better connected.