AI invades ‘word of the year’ lists at Oxford, Cambridge, and Merriam-Webster

Few would disagree that 2023 was, in the world of technology at least, dominated by artificial intelligence. The dictionaries have taken note in their “word of the year” lists, and notably all the AI-related words they highlight are, in fact, existing words that have been appropriated and regurgitated with new meanings. A little on the nose, isn’t it?

Cambridge’s word is “hallucinate,” which is of course the habit of generative AI models like ChatGPT to invent anything from dates to entire people rather than admit it doesn’t know. The problem is that these systems don’t know what they don’t know, because they don’t know anything at all.

As complex word prediction models, all that matters is that they produce a sentence that resembles their training data. If you ask it for famous 18th-century German surgeons and it doesn’t have any exact matches, it will simply hallucinate something close, like Arman Verdigger of the Einschloss Research Hospital in Tulingen. See, I can do it too! All that matters is that it sounds plausible. Unfortunately, these hallucinations are so confidently stated that countless of them have been accepted without question as real.

Hallucinations can be put to good use, though: generative imagery and audio is entirely and deliberately “hallucinated” in that it is a mishmash of the model’s training data but not an exact recreation of any of it (though it can get mighty close). This too has its dangers, as AI-generated art and photos of varying quality proliferate in numerous contexts.

The acceptance of the word despite its original limitation to human perception “underscores our readiness to ascribe human-like attributes to AI,” said Cambridge AI ethicist Henry Shevlin. “As this decade progresses, I expect our psychological vocabulary will be further extended to encompass the strange abilities of the new intelligences we’re creating.”

Merriam-Webster grabbed the other end of the stick with the selection of “authentic” as their word of the year. “With the rise of artificial intelligence—and its impact on deepfake videos, actors’ contracts, academic honesty, and a vast number of other topics—the line between ‘real’ and ‘fake’ has become increasingly blurred.”

While “authentic” didn’t get a brand new definition, it did get a new and important connotation. For years we have worried about whether or not something we or others are doing is authentic. Authenticity is a paradox modern consumerism: it can’t be bought or sold, and as such it is perhaps the most valuable and marketable quality in the world.

Before, we had to worry whether a trend or item represented the authentic interests and choices of a person or group. Now we have to wonder whether, like the Pope’s fabulous Balenciaga puffer, a thing is real in the first place.

“Deepfake” also made the longlist at M-W, graduating (whether mercifully or unfortunately) from a niche tech for revenge porn to a general purpose term for generative AI. Its antecedents may not be respectable, but we can’t choose what enters the zeitgest.

Case in point, Oxford’s word of the year — which it would be much better for this article had it been AI-related, but unfortunately the AI term is relegated to runner-up. “Prompt,” a versatile and underused word, has gained another definition with its now well-known meaning relating to the human side of generative AI.

When you tell an AI system to put together a list of article ideas based on the current weather, you are providing the “prompt,” and indeed the word quickly became a verb, and one “prompts” a system now.

Of course these are perfectly appropriate extensions of prompt’s existing definitions. We have prompted a response for centuries. And as a noun, the use of “prompt” was originally reversed in computer interfaces: the command line prompt was itself prompting the human for a response. So here we have an interesting reversal. Who is prompting whom — or what? Whether this has empowered or diluted the word is a matter of taste.

If you were wondering what Oxford’s actual word of the year is, it’s “rizz,” a playful shorthand for “charisma” and something that AI arguably lacks entirely, like Tom Holland.

It was inevitable that AI terminology would infiltrate the lexicon, though I’m a little sad that the cooler terms like “latent space” have yet to enter general use. The technology is moving fast enough, however, that it is perhaps better to stick to the well established, as indicated by the judgment exercised by my peers, as I would like to think them, in the lexicographic world. We await further words of the year, however, as bolder dictionary content teams consider whether vectors and embeddings deserve a boost as well.