Do Not Go Gentle: Why We Must Not Cede Writing to the Machines

In a pivotal scene in Good Will Hunting, Sean Maguire, the therapist played by Robin Williams, confronts Matt Damon’s arrogant but troubled Will, who had recently mocked his painting of a boat. Sean had been hurt by Will’s comments until he reminded himself that his patient is just a kid. Will, he goes on to explain, can expound on many areas of life—on art, relationships, war, grief—but has experienced very little of it. “If I ask you about love, you’d probably quote me a sonnet,” he says.

(ChatGPT: A Writer’s Best Friend…for Now.)

In 2023, writers now find ourselves facing a formidable new competitor with a similar air of bravado. It has experienced nothing at all, but can also quote us a sonnet—and then write its own in a matter of seconds. This might not result in a masterpiece, but there’s no denying that changes are afoot and that the drumbeats are only getting louder. 

There’s the AI-written novel that was reviewed in The New York Times Book Review. There’s the science fiction magazine that was forced to suspend short story submissions after being flooded with AI-generated work. There’s the Hollywood writers strike, fueled in part by studios refusing to rule out using AI to write their screenplays (the concurrent actor’s strike revolves around similar concerns). Meanwhile, creative writing instructors like myself are struggling to flag work that may not have been written by human beings let alone our students.

In the midst of all this, working writers have tried to reassure ourselves that we are not about to be replaced. The computer-generated writing is pedestrian and predictable, we insist. The Times gave the aptly-titled Death of an Author a pretty negative review, concluding that, “A.I. will never pose a threat to the real thing — to writing with convictions, honest doubts, riddling wit, a personal vision of the world, rawness and originality.” Meanwhile, the science fiction editor, Neil Clarke, was able to catch the AI submissions partially because they were “bad in spectacular ways.” 

However, any writer taking too much solace in AI’s significant shortcomings is ignoring the fact that this technology is in its infancy. As has been widely reported, while Chat GPT3 scored in the bottom 10 percent on major standardized tests, its successor, Chat GPT4, scores in the top 10 percent. If AI can’t yet produce high quality prose, might it just be a matter of time before its work equals and eventually surpasses our own?

(Will AI Steal My Job? A Writer’s Exploration.)

I don’t think so. To start with the obvious, no matter how far this technology advances, AI will always be missing two key ingredients that have always driven great literature: experience and imagination. Granted not all writing requires these elements to be effective, but work that aims for true power and transcendence does. How could Chat GPT or Google’s Bard have anything worthwhile to say about love or pain or joy when they’ve never felt any of these emotions—and wouldn’t be able to express anything original about them even if they had?

The best AI can ever hope to do is ape what humans have already written about such subjects, which brings me to my next point, one which often gets ignored in conversations about AI-generated prose. Since these models are all trained on pre-existing text, if we human authors were to eventually put down our pens and allow AI to take over, that means even further down the line, AI’s own source material would eventually consist primarily of AI-generated content. 

Thus, as AI tools continued to train on AI-generated writing, like vampires drinking their own blood, their prose would inevitably become successive degrees removed from actual human experience. Sure, readers might settle for this “literature” in the absence of anything better, but make no mistake: It would have all the authenticity, resonance, and relevance of a fifth generation copy. It would be lousy.

Still, even if we concede that literature should be left to human authors, can’t more common, everyday writing tasks be left to AI for convenience’s sake? For instance, why break our pencils trying to come up with the perfect “thank you” note or apology letter when ChatGPT can so easily draft one for us? Yes, we might become poorer writers in the long term if we depend too much on AI, but we’ve surely become weaker at math by letting calculators do our long division for us, and most people would agree the tradeoff has been well worth it.

However, we should be wary of comparing writing, a vehicle for human communication and expression, to mathematical calculations. Using ChatGPT to draft a press release or legal document is prudent; such documents are meant to feel anonymous and to simply convey information. However, using AI to craft personal or even business correspondence is another matter, as doing so severely dilutes our connections with other people. In such cases, the technology isn’t merely helping you communicate; it’s communicating for you.

In fact, it’s questionable whether an apology outsourced to ChatGPT even counts as an apology. When I posed this question to my brother Jeff, a philosopher at UC Irvine who’s written extensively on speech acts, he thought it doesn’t. “The work that apologies do depends essentially on them being an expression of the inner mental life of the author,” he said. If you’re still not convinced, just imagine how you would feel if a friend wronged you and then sent you an earnest apology that turned out to be the work of a crisis management firm. Would you be inclined to forgive them?

Writing is one of the chief ways we extend our impact in this world beyond our immediate surroundings. At its best, it allows us to reach across time and space and exert our influence far into the future, to engage and inspire readers who may not even exist yet. At a minimum, it’s an essential means of self-expression and communication that serves as a bridge between ourselves and many of the people in our lives. It is how we record our history and, frequently, how we imagine our future.

Now, the machines that have populated some of these imagined futures are upon us. As they continue to advance, they will inevitably get smarter and they will often seem like more than machines. That should not be an excuse to make ourselves any less human.