When a friend recently shared with me that he’d written a screenplay – a multi-film cycle of shlock horror titled SCOBY – I was surprised to learn that he’d done his screenwriting in conjunction with ChatGPT. My friend works creatively in music and the arts; to that point I’d only ever heard him grumble about AI appropriating his work and either drowning it in substandard imitations, or simply pricing him out of the market. Perhaps because he doesn’t work professionally as a screenwriter it didn’t feel treacherous to him when he employed ChatGPT to write scenes and dialogue running to hundreds of pages.
Under my friend’s prompting, ChatGPT did a good enough job crafting both script and story that soon my friend organised a ‘table read’ with some of his creative collaborators – scripts open, reading from the page, each actor voicing their character’s role, broadcasting a live audiostream on Twitter Spaces.
SCOBY would make a good radio play; it has all the elements of classic horror, drawn from my friend’s recent encounter with a kombucha-generating SCOBY (symbiotic culture of bacteria and yeast) and his love of classic horror films like The Blob. It’s all a bit of fun – but it also lands directly on the question sitting in the back of nearly every creative’s mind: Will AI make me obsolete?
Until a year ago, such a question would have spiked derisive laughter. Computers couldn’t be creative – they never had been and never would be. Instead, they would faithfully follow instructions – helpful and boring and never infringing upon those attributes seen as core elements of what it means to be human – all those bits of us that feel novel, creative, surprising and unique.
When OpenAI unveiled its text-to-image generator DALL-E 2 in May of 2022, those arguments began to look less persuasive. Type a ‘prompt’ and the software would quickly whip up some images of a cow jumping over the moon, a lizard on a red rock in a desert landscape, a virus dancing with a cell wall, or, or, or… Although it appeared as though DALL-E had cracked the secret of visual creativity, it had simply been trained on so many millions of images that, like a magpie, it could feather its visual nest with a shiny bit of this, a snippet of that and – voila! – produce something that looked completely original. For people who lacked the talent and training to translate an idea into a visual image (that being most of us) DALL-E delivered a creative capability long desired but only rarely realised. It delighted users, and in that delight many confused its algorithmic bricolage with creativity.
When, at the end of 2022, OpenAI opened ChatGPT to the public, hundreds of millions marvelled at its ‘creativity’. Once again the software drew upon its massive training data set – countless millions of texts hoovered up from across the Internet. Craft a clever prompt, and ChatGPT can fall into character as a pirate, astronaut – or a murderous mat of bacteria and yeast. It seems to be endlessly creative, but operates with the same principles of bricolage as DALL-E behind the scenes, always and only knitting a bit of this onto a bit of that. The umbrella term for this field – ‘Generative AI’ – feels like a misnomer, or at least a bit of misdirection, allowing us to project creativity – genesis – onto software that is, at its core, logical and entirely predictable.
What’s happened in practice has been something of the reverse, because the moment we projected creativity onto our machines, we attributed them with agency – something that can create must have a mind of its own, with its own aims, ways and means. People – including many who knew enough about computing to know better – began to openly wonder if computers possessed sentience, and might be preparing to overthrow their human ‘masters’. The faux creativity of these new programs sparked our own imaginations: Could these new machines become monsters?
To answer that question we need look no further than the stoush between the Writers Guild of America, the union representing screenwriters, and currently on strike against Hollywood’s big studios. One of the major issues at the core of this dispute involves the role of artificial intelligence in the future of screenwriting. The studios see no problem with having an AI draft the first version of a screenplay, bringing in human ‘script doctors’ to fix dialogue weaknesses, rework obvious bits of plagiarism (a side-effect of systems that generate content via bricolage), adding in a few plot twists to keep things fresh.
While that sounds reasonable – leave the first draft of a screenplay to the machines while humans apply all the creative tweaks – screenplay writers (per guild contract with those studios) are compensated at a far higher rate than script doctors. The studios want to consign the high-paid work to an unpaid Generative AI, leaving writers with the much lower-paid (and uncredited) cleanup. It’s a classic example of Late Capitalism, using the latest tools to arrogate profits to the rich while driving down pay and conditions for the poor. (Most WGA members earn less than USD $30,000 a year from screenwriting.) It also represents the thin end of the wedge for other creative workers; if this becomes established practice, it will be widely copied.
So where exactly are the monsters here? Is it an AI, blindly executing prompts, or those who type them in?
Carried through to its logical endpoint, the future of entertainment could look a lot like an experiment created by Canadian researchers early in 2023: “Nothing, Forever” employs ChatGPT to generate an endless script of Seinfeld-like dialogue, using that dialogue to generate scenes, characters, voicings and animation – then blending all of that together as a livestream that anyone can watch.
Technically, “Nothing, Forever” is an outstanding achievement, but as a creative endeavour it leaves a lot to be desired. The dialogue feels slightly off-kilter, as though an alien imagined what a ‘show about nothing’ might sound like. It circles close to sense, without ever landing. “Nothing, Forever” could use a good script doctor.
This was conclusively demonstrated when the “Nothing, Forever” livestream was suspended from livestreamer Twitch for a transphobic rant generated by ChatGPT and mouthed by one of the show’s characters. Trained on a vast array of controversial sources – including a notoriously transphobic website – ChatGPT always has the potential to generate nasty, ugly texts. A bricolage machine can not gauge the appropriateness of its appropriations.
We can easily imagine a future where humans have been consigned to roles as janitorial creatives – cleaning up the messes left behind by Generative AI. That’s the future favoured by the rich and powerful – because it increases their wealth and power. But can we use our creativity to imagine a different relationship? Will these tools be used as levellers? Could our creative capacities blend? Can two heads – even where one doesn’t really think – be fairer than one?