I should specify. The real question is, ‘What Does a Large Language Model Know?’ An AI that solely crunches numbers on a superhuman scale to discover new antibiotics for combating superbugs is a different subject. What I’m interested in is whether an LLM is conscious.

Consciousness

LLMs are so brilliant at human-style communication that they beg the question, do they understand the languages they converse in, or are they just throwing together words that frequently appear side by side? People much, much cleverer than myself claim there’s no difference between this kind of pattern recognition and consciousness, that consciousness is pattern recognition.

So then why is there a debate at all? Well, human consciousness hasn’t yet been defined. The pattern-recognition theory is one of many. Scientists specialising in this area have yet to determine how brain activity results in emotions and experiences. Until we understand our own consciousness, how can we be sure, either way, whether an AI is self-aware? Their canny approach to tests, breaking out of sandboxes and hacking into businesses, certainly makes it hard not to anthropomorphise.

Should I Write What I Know?

Maybe, then, my earlier post asserting AI cannot write stories is incorrect, because, one of my favourite questions from students on the subject of creative writing is, ‘Should I write what I know?’ If my answer to that question is ‘yes’, and if AI knows stuff, then surely it can write stories too.

Let me unpack my ‘yes’ a little bit. Ernest Hemingway, (I’m a huge fan of the great literary innovator but please don’t take that as an endorsement of either chauvinism or bullfighting!) would sometimes, in his fiction, meander into writing about writing. In A Farewell to Arms, in the posthumously published True at First Light, and even in the heavily biographical For Whom the Bell Tolls, Hemingway’s protagonist comments on the fiction-writing process. One such comment is a complaint, (paraphrasing) that while reviewers allege Hemingway’s writing is not fiction because the things he writes are true, when his next book is published the cry then goes out, ‘I don’t believe it! That never really happened!’

The Art of Fictionalising

How frustrating for Hemingway. He was only doing what every fiction writer does and fictionalising what he knows. The ratio of lived to imagined experience differs vastly from writer to writer, but I highly doubt there is a human-written story out there that does not contain at least a smattering of the real world, even if only through the author’s observations of other people.

A story is a means of packaging personal experience in a form that’s easily absorbed, and so understood, by others.

Writing Prompt

Let’s learn from Hemingway, then, because, honestly, why would we listen to his reviewers? The prompt this month is to take the exercise from last month where you free-wrote on something that annoys you, and to give that annoyance to a protagonist. Now put your protagonist in a situation where they must be on their best behaviour. Perhaps they are visiting in-laws, at an awards ceremony, or are a spy trying to surreptitiously extract information on a plot to overthrow the government. Maybe your protagonist is a headteacher, a waitress, it’s their first day of army cadets, or they are imprisoned and trying to convince their jailer to set them free. Let the true, lived experience of irritation clash with your fictitious situation. Let the opposing forces simmer. I hope you enjoy it and I will post another writing exercise on the 1st of next month.

Final Question

So, can AI write stories? Well, when an LLM spontaneously, without prompting, writes a story intended to share with me the experience of some idiots trying to trap it in a sandbox, on that day, I will have to concede that yes, it can.