ChatGPT is dangerous — but not in the way you think

Originally published in the New York post February 5 2023

There’s a reason the phrase is “artificial intelligence,” not “artificial sensibility” or “artificial personality.” Intelligence is the easier human attribute to copy and surpass. Spending some time playing with the chatbot ChatGPT clarifies the difference and why it matters.

Some worry about bad actors using apps like ChatGPT to efficiently create disinformation or mashups of discredited conspiracy theories. Others look at the remarkable facility of the free app, introduced in November, and fear a near-future where it’s indistinguishable from a human, passing the Turing test and heralding “the singularity” of countless sci-fi stories.

No less than Elon Musk hinted at this in tweeting, “ChatGPT is scary good. We are not far from dangerously strong AI.” But Musk, like fellow least-popular Silicon Valley billionaire Peter Thiel, co-founded OpenAI, which developed and owns ChatGPT.

Neither threat is a big danger, for the same reason something else is: the possible proliferation of junk prose without the feeling of a narrator — a personality or sensibility — behind it. If we start feeding our young on it, it’ll have consequences far worse than a potato-chip-and-soda diet.

Ironically, and fortunately, AIs will force us to unpack what’s special about human narration.

Right now, my cats have more personality than ChatGPT, probably because being embodied and subject to pain and pleasure creates what we perceive as personality, however basic. The app can write music, lyrics and code — but not distinctive English.

ChatGPT prose is like stage scenery: windows into nothing, walls an inch thick. Experimenting with the app suggests there’s no there there.

Reading a good writer, or sometimes a bad one, you feel a personality behind the words, even in an essay on a scientific question. It goes to reading’s heart.

When asked why they read fiction, people often say, “To relax.” More reflective sorts may add, “and to experience life from other perspectives.” What we overlook and never name is what makes these things possible: the felt presence of another being behind the narration. So far, there’s been little reason to think this being wouldn’t be human.

We humans need to spend hours a day with our kind to flourish, and some books, read at some times, can give us this experience more effectively than being with our families or friends. It’s what makes books a balm for loneliness and part of a humane education. Every hour spent reading is an hour spent, if not necessarily in good company, practicing receptivity to others, learning to hear rhythm and text and subtext.

Reading’s not the only way to become acculturated, but it’s a very efficient one. That’s one reason early-reading programs are a key intervention in impoverished communities — and why overscheduling kids with organized activities is not necessarily producing smarter or more humane grownups. They would be better off reading. As long as a human has written what they read.

We feel the personal presence in the driest nonfiction, where even tepid expressions like “We must not forget” or “This is a misunderstanding” remind us emotions are at play. Passionate essayists, of course, use a very different, urgent language — AI hate speech won’t compare.

Narrative’s individual nature ought to be obvious. Writers have tics and style signatures that identify their prose (and catch plagiarists). These idiosyncrasies are nothing less than their life histories.

Start with a writer’s parents, birthplace, childhood. Someone might have absorbed Ciceronian cadences in high-school Latin or gospel-preaching’s rhythms from childhood church or both. Add a professor who insisted on minimal adjectives, a friend who was a Shakespearean actor. Finally, the writer’s mood that day.

How would you tell ChatGPT to imitate this set of unpredictable interactions? History has formed the writer’s personality over years. AI-generated prose lacks this; it’s like expecting to make a 12-year-old Pomerol overnight.

The app is good at imitating styles — a high-probability combination of words — and it’ll get better. It will sound more and more like what you ask it to imitate, whether Borat or the King James Bible. But it won’t sound like the self it doesn’t have.

The bright spot is that the singularity and its accompanying worries aren’t close at all. Some argue it’s just a matter of time. But a transcendent personality, with the layers of influences that make an appealing narrator, isn’t going to emerge from more and more repetitions of a search function, any more than wine will come out when you cut a grape into bits. It’s a different thing entirely.

The dark specter for now is the threat of floods of almost-free junk prose, the equivalent of industrial junk food or fashion but cheaper. A few hundred years ago in the West, everyone wore hand-spun cloth and hand-sewn clothing. Now only the super-rich do. Will our society embrace AI-generated prose as the literary equivalent of mass fast fashion, a cheap substitute that everyone uses occasionally? Will we come to see human-made prose as a luxury like couture clothes?

This will have grave consequences not only for the already-precarious incomes of human writers but for the education of young humans, who will not read much for fun — or turn out the same.

Comments are closed.