Language Does More: Forget Job Security - LLMs Are After Your Soul

How often do you use ChatGPT? The latest figures I’ve found suggest that half of adults use Large Language Models (LLMs) like ChatGPT several times a week, with a third using them several times a day. They save people lots of time on tasks they’d rather not do.

By now, everyone’s familiar with the basic workings of LLMs, or at least everyone has a spiel they can deliver so it sounds like they are. May I recommend this one. It’s in the spirit of similar explainers on era-defining technologies of eras past, including Deep Blue, Watson, and (if you’ve got time) Go. What’s so remarkable about developments like those, of course, is not that machines got really good at something – it’s that machines became more able to get really good at anything. Now they can do what only humans were able to do. It’s a story of scientific progress.

As a linguist, I find this astonishing for language. Previous technologies could learn how to win at games, which have rules and scores, and it’s not surprising to me that computers could figure that out. But my sect of linguistics taught me that language was too complex for that. Language starts with sounds that don’t mean anything on their own – we’re not like animals that make one sound when they’re scared and another when they’re happy. Then language combines those sounds into words – there can be a word for almost anything, even things that we’ve never experienced ourselves, even things that don’t exist! And then we all go through our days and string words together into new utterances. We can express messages of fantastic specificity on any topic, like “Most Christians celebrate Easter on the first Sunday after the first full moon after the Spring Equinox, but there was no name for that day before the holiday was canonized.”

So language is big. We can capture almost anything in the world – whatever, whenever, wherever, however, whyever. I’m constantly amazed that children acquire language at all. So were Chomsky and even Plato.

But LLMs seem to have cracked it. Companies like OpenAI train LLMs on hundreds of billions of words’ worth of language. They basically fed their models the internet. (No matter how much time you think kids spend online these days, it’s not that much.) And it’s hard not to say that they’ve mastered language – I’ve never once seen a ChatGPT response that needed copyediting. So Sam Altman just sidesteps Noam Chomsky and bowls Alan Turing over.

But this story of scientific progress may turn out to be Gothic literature – less Higgs-Boson, more Frankenstein. When I first saw what these chatbots can do, I felt like I was witnessing something perverse, not just amazing. That couldn’t be real! The zeitgeist shifted almost immediately from one of scientific novelty to one of economic threat. We became convinced that they were coming for our jobs – again. It sure is a hard worry to shake, if your job is writing social media copy or legal memos, to watch ChatGPT spit out a B- version of your work in seconds.

To me, the thing we should be afraid of with LLMs is not that they’ll take our jobs, but that they’ll take our souls. That’s hyperbole, obviously, but I think I mean it. While it’s certainly a story of scientific progress, it might become a story of human regress.

Take a detour with me to the grocery store, and consider the humble self checkout machine. If you have 10 items or less (no I don’t mean “fewer”) you can use a self checkout machine. No need to bother anybody else. No need to make conversation. How hard can it be? Well, suffice it to say, they have not worked nearly as well as planned. They drive people nuts. Ever since I started noticing this, I’ve tried to notice how my experiences in old fashioned grocery checkout lanes go. They’re fine. I usually have to answer two questions – if I found everything okay and if I have a rewards card – and speak a third time to say “thank you” at the end. Easy peasy. Still, I cannot prevail upon most people I know that this is the superior option.

Compounding this is a theory I developed years ago: that how we talk to malfunctioning gadgetry is who we really are deep down. It’s a variant of Dave Barry’s so-called waiter rule: “If someone is nice to you but rude to the waiter, they are not a nice person.” Machines are meant to serve us, and we learn a lot about ourselves by how we interact with servers.

And so I’m concerned that we’re playing with a Rube Goldberg machine designed to sap our humanity. People spend unfathomable sums of money on energy intensive machines that perform a mediocre version of what humans can do, all so we don’t have to interact with each other, leaving us bitter when we do have to interact with each other. We insist on using the miraculous get-away-from-me machines only to hate how much better it could have been. One very possible future is that we’ll spend even more time alone, this time communicating directly with a machine. How small our world would become.

As a linguist, I find this astonishing for language. The whole reason language exists is to bring people together. You can use language alone and benefit from it alone, but that’s just a curb cut effect – it would never have developed without a social need for it. Most of us could stand to think a little more about quality socializing. Now I think of that every time I ask a chatbot a question. It’s like my friends who text me asking where this or that word comes from – they could just google it, but it sure is nice to communicate.

That’s how the columnist David Brooks can sleep at night. He and others have argued that there are certain jobs we’ll never let robots take: the ones where we need another person. I know that I’d be be open to AI assisting with processing a medical test – but I’d never let one tell me the results. There are some things that machines just aren’t for.

At the end of the day, this all reminds me of the play Inherit the Wind, a dramatization of the Scopes Monkey Trial. There’s a monologue by the hero that captures the bittersweet balance of all scientific advancements: Progress has never been a bargain. You have to pay for it. Sometimes I think there's a man who sits behind a counter and says, "Alright, you can have a telephone, but you lose privacy and the charm of distance." "Madam, you may vote, but at a price: you lose the right to retreat behind the powder-puff or your petticoat." "Mr., you may conquer the air, but the birds will lose their wonder and the clouds will smell of gasoline."

I sure am glad for the playwright who used language – our big, large language – to model this epiphany for me. I don’t think LLMs can do that. If you do, well, let’s talk sometime.