AI and generative writing with GPT-3
Are you a writer or writing teacher wondering about computer-assisted writing? Things are moving super-fast with AI developments, and I’ve been amazed how writing colleagues haven’t been aware. Here’s a catchup to get you up to speed.
Generative creativity – first art, now writing
The generative art app DALL-E has been making headlines with amazing images made by artificial intelligence (AI) – but what about AI writing? Should writers be worried about robot competitors? And what are the implications for teaching, for copyright, for publishing…?
I’m particularly interested in this, as I’ve been following writing technology developments over the past few years. It’s been improving fast! When DALL-E AI made an exponential leap in image quality from ‘blurry lo-fi‘ to ‘convincing art in any style in seconds’, it felt as though text wasn’t far behind.
Also, I teach creative writing. I’ve spent over 13 years feeding back on student work in progress, including fiction drafts, scripts, and web copy. I’ve also worked as an editor and journalist.
So for a few years, I’ve been pestering university colleagues with the question: what do we do when students start sending AI-generated writing assignments?
No one seemed interested. I don’t think they took it seriously! But that day is now here, and faster than I expected. Generative art and co-created writing are among us, and will have a big impact in the creative arts. Here are some frequently asked questions for writers and teachers.
GPT-3 is an artificial intelligence (AI) engine that can generate writing automatically. It’s a software programme, and perhaps best understood as the writing equivalent of DALL-E.
GPT-3 is the engine behind a whole range of user-friendly programmes in various writing niches, including copywriting, fiction writing and script. There are so many emerging that it’s impossible to keep track, but key ones for now include Jasper and CopyAI.
How will this affect writing?
Writers are already using these programmes to write faster. Imagine inserting a prompt and generating a whole article draft instantly – it’s that fast. It’s already being used in copywriting, and is spreading fast to fiction.
So how will creative writers respond? It’s hard to tell, but perhaps visual art gives a clue. Some visual artists are becoming co-creators with AI, using the apps to pushing creative boundaries. No doubt this will happen more and more in creative writing, too. Theatre has already produced shows using AI-generated scriptwriting, with the first AI play, When A Robot Writes A Play and AI, a 2021 show at the Young Vic Theatre which used live GPT-3 prompts. Filmmakers are experimenting with ‘bot scripts’. And many indie authors are trying out the possibilities for novel writing. First results may still be far from stellar, but who knows what’s ahead?
Writing jobs are already being challenged and changed by AI developments. In translation and journalism, the emphasis is evolving towards editing, research and newsgathering. Creative writing may head towards a more editing-led role, too. Teachers need to know this.
Can you tell the difference between human and AI-generated writing?
In some situations, no. I tested some teacher colleagues with AI sample writing – a kind of Turing test. They didn’t get them all right. They disagreed about which samples were human, and which were AI. At first, I was a bit complacent, too, until I did another test, and failed to distinguish between AI and human samples.
The good news is, wonderful standout human writing – the kind we love to experience in literature or in films – can still be spotted. But everyday, learner and student writing are hard to tell apart from AI-generated writing, even for teachers. The boundary between great AI writing and ‘not-great’ human writing is very blurred.
Of course, in the tested material, we were looking at short samples. I suspect that over a long sample, the oddness and incoherence will build up, and the ‘tells’ will become clear. But the exponential rate of AI improvement means complacency would be a mistake.
Can plagiarism detectors pick up computer-generated writing?
No. Computer-generated writing is new and previously unpublished. Plagiarism detectors work by scanning your assignment against a huge database of existing texts. It’s looking for similarities. So it wouldn’t pick up your student’s AI co-created assignment. Whose work will you really be marking?
What are the implications for teachers?
First of all, teachers need to understand this backdrop, and experience the technology for themselves. Only then will the implications hit home. Try Jasper or CopyAI as a first step. Then, education will need to find ways to work with this development, rather than against it. Use it an opportunity to develop research skills, editing skills, journalism skills, including an understanding of newsgathering, attribution, sources, critical engagement with facts – skills that are already vital in today’s world.
In creative writing, dive deeper into voice, originality, individuality, metaphor, logic, poetics, stylistics, flow and narrative – aspects that set the best human writing apart from the humdrum. Put clear water between human and robot writing.
It will no longer be enough for writing students to produce average, competent work. They’ll need to raise their game, and really get to grips with language itself. They’ll need to prioritise what it can do, how it can engage, how to create impact, storytelling tactics. They’ll need to dig beneath the veneer of words, and might just develop some armour against manipulation. That can only be a good thing.