What does ChatGPT mean for grammar, cyberbullying, and the future of language?

We already know the machines are going to win. Is that a good thing — grammatically speaking?

In our brave new AI world, chatbots can do things ranging from highly trivial (like giving you detailed medical diagnoses) to deadly serious (you can converse with Gritty). They use algorithms to predict which word should come next, like your phone might do when you’re typing a text message … only turbocharged.

Which begs the question: What are the grammatical implications of chatbots? For instance, if I (or more likely, you, because I mean really …) ask a chatbot an ungrammatical question (“Is our children learning?”), can it make sense of your query like a human might?

Will a bot always reply with perfect grammar? If they do, will that make us better, because we always have a well-written example to point to, or will it make us dumber, in the same way that GPS has destroyed my sense of direction? On the other hand, if machines learn bad grammar from bad inputs, will they respond with bad grammar, thus making our problems even worse?

Our AI future is terrifying.

“Our AI future is terrifying.”

“Even a few years ago, [chatbots] would generate text that was kind of ungrammatical, or definitely ungrammatical, and often off topic,” Chris Callison-Burch, an associate professor of computer and information science at the University of Pennsylvania, told me in a recent interview. “Then we got better at [keeping bots on topic] and then better at grammaticality. And now, they’re really good at a lot of stuff.”

Every fall, Callison-Burch teaches 500 students artificial intelligence, where they play with making the robots smarter. They’re getting really good at it.

Machines have “been trained off of probably on the order of a trillion words’ worth of texts from the internet,” said Callison-Burch. “So that includes both grammatical and ungrammatical.” The machines learn language in a way that’s “100% descriptive,” he said, and then use probability to predict what’s likely to come next. So while it’s unlikely to spit out mistakes that are otherwise rare for humans to make, those mistakes that are more prevalent are more likely to become ingrained in the bot’s vernacular. If enough people use a made-up word — say, turnt or stonks or stimmy — machines could pick it up.

That means machines could accelerate language changes already underway. Naturally spreading neologisms — new words or phrases — might end up in a chatbot transcript before they end up in a dictionary. Cool, right?

» READ MORE: We asked ChatGPT to tell us a story about Gritty. It delivered in seconds.

Descriptivism is dandy, but reinforcing the ways people speak can be perilous. Callison-Burch raised the cautionary tale of Tay, the Microsoft chatbot unveiled back in the prehistoric internet of 2016. Tay was built to mimic what others said to it. So less than 24 hours after its debut, racists and misogynists peppered Tay with enough trash to turn it into a hate machine that began tweeting out things like, “Hitler was right i hate the jews.”

Despite some humans’ tendency to take new technology and use it for bullying or worse, chatbots show promise for language — especially for non-native speakers looking to improve how they speak. Callison-Burch mentioned Grammarly, a writing assistant that — though its language suggestions are often generic — can prove useful for some who struggle with grammar. “If you compare [chatbots] against what Microsoft Word used to suggest to you, they’ve come a long way,” he said. RIP, Clippy.

To make the technology better, though, the scientists working on this are doing what all of us who study language do: trying to understand how words work.

“There’s no other species that really communicates to the same degree we do,” said Callison-Burch. “Animals have their own forms of communication, but we have this thing that’s so rich, and imbues us with knowledge and culture and all sorts of traits that are amazing and uniquely human.”

Maybe, if the machines can understand that, we’ll all win.

The Grammarian, otherwise known as Jeffrey Barg, looks at how language, grammar, and punctuation shape our world, and appears biweekly. Send comments, questions, and interlingua to jeff@theangrygrammarian.com.