The vast megacorp for which I work is very keen on AI, has spent (I suspect) vast sums of money investing in AI, and is going to great lengths to make sure that reluctant employees (i.e., yours truly) not only use it on a daily basis but learn to use it “properly.” As a result of endless roundtables, seminars and certifications, I am now significantly more up to speed with AI than I used to be. Much to my surprise, I really do use it every day and it is genuinely useful. Sometimes. I don’t see myself being replaced by a Large Language Model any time soon, but it is a terrific productivity tool.
Having been acclimated to the use of AI in my day job, I had begun to wonder whether it could be of any use in my writing life. Not to write a novel, or plot a novel, or outline a novel (no, no, no, no, NO!) but maybe something around the edges, something more administrative than creative. Could I find what Corporate America would undoubtedly call a “use case?” (Which reminds me: a new study from Cornell University seems to show that people who are impressed by corporate BS may not be very good at their jobs. 🤣) Curious, I thought I’d play around with AI in a couple of low-stakes areas.
First up, LoD, which is due out this fall, needed back cover copy. I had drafted one for the publisher, but they’d come back to tell me that it had too many syllables. Their amended version did, indeed, have the requisite number of syllables but read a little flat. I wanted to change it but was kind of stuck. Could AI help? I crafted an appropriate prompt and five seconds later it came back with an alternative version. It wasn’t that great either, but it did contain a couple of turns of phrase that inspired me to use some different turns of phrase that got the job done. I would have got there in the end, no doubt, but having been noodling the problem for around half an hour, my AI-inspired fix took about five minutes from prompt to final edit.
I have also (finally!) finished the outline of Murder in the Third. (I still don’t have a “real” title for the publisher to change 😬). What my editor usually wants, though, is a two-page synopsis to share with their colleagues. Hoping to save some time, I fed my outline into the AI and asked for a summary.
I did not save time.
To start with, the AI produced a series of summaries for non-existent novels, none of them very good. But even after I’d wrangled it into doing what I asked, the output was so grossly inaccurate it couldn’t even be described as a bad attempt at a rough draft. Large Language Models are very good with straightforward, factual prose, but the nuances and leaps of a fiction outline proved too much for it. Maybe, if I’d spent even more time on it than I already had, I could have corralled the AI into giving me something I could work with. But why? It had already taken too long. I ended up writing the synopsis from scratch, it didn’t take half as much time as I feared, and much to my surprise, I rather enjoyed it. As a productivity hack, AI as synopsis writer was an epic fail.
People in the creative industries are rightly leery of Large Language Models, not least because their original sin is the copying and synthesizing of human work product without seeking permission and without paying for it. For the creators of AI, human creativity is, quite literally, worthless—at least, until they get sued. But AI is in the world now, and we have to get used to it. For creative people, do LLMs have anything worthwhile to offer? An LLM is neither alive, nor sentient, nor intelligent. At bottom it is a probability engine generating word salad and I don’t think we should be scared of it. We do, however, need to understand what it’s good for and what it’s not good for. If my experience is anything to go by, it’s not that much use for novelists. I got the most use out of it as a glorified thesaurus. It provided a spark to my own thinking. But when I needed its help to generate something more substantial, it didn’t produce anything useful. My views will undoubtedly evolve over time, but right now I don’t think AI has a stellar future in the world of fiction.
Now, having said that, this whole exercise forced me to confront something I find extremely troubling. Bearing in mind that AI’s output suffers from the taint of stolen intellectual property, I was very careful to expose it only to my own work and no one else’s. But it turns out the LLM has no respect for that particular boundary. For instance, in the course of providing synopses of non-existent novels, the AI generated an equally non-existent character called Dorothy Vane. Now, I happen to know that Harriet Vane features in a number of Dorothy Sayers’s Golden Age detective novels. I can’t prove that the AI smushed those names together from that particular source, but it got the name from somewhere. This matters, because even if the AI is simply helping you with something minor, there’s no guarantee it isn’t stealing someone else’s intellectual property to do so. Just the other day, the New York Times cut ties with one of its book reviewers, because the book reviewer used an AI to assist him in writing one of his reviews. While no one is suggesting that the review reflected anything other than the (human) writer’s opinion, the fact of the matter is that the AI copied chunks of language from a Guardian review of the same book without informing the reviewer that it had done so. The plagiarized language ended up in the New York Times and the unsuspecting human paid the consequences.
We shouldn’t be afraid to experiment with AI, but we do need to be careful. AI is not, itself, creative but it can be a spark to our own process. What we can’t do is let it spark our process on the back of someone else’s hard work. Given AI’s original sin, is that even possible?