Hallucinating AIs and What The Words Of The Year Lists Reveal About our Modern World
Newsletter offer
Subscribe to our newsletter for exclusive editorial emails from the Byline Times Team.
If, as Samuel Johnson once wrote, “language is the dress of thought”, dictionaries around the world are encasing themselves in stainless steel and advancing towards us with an ominous clank: Artificial Intelligence (AI) – and all that this technology might one day entail – is all over 2023’s ‘Word of the Year’ lists.
In fact, AI-related words appear in the shortlists of pretty much all the dictionaries that have so far published their Word of the Year (WOTY) nominations. ‘AI’ is the winner for Collins and ‘Generative AI’ came out top in the Australian Macquarie Dictionary’s 'people’s choice', while other dictionaries such as Oxford, Cambridge, and Merriam-Webster all flag up AI-related words. What’s interesting about some of these nominations is not that AI appears – technology is (and has always been) one of the leading drivers of linguistic innovation – but the lens through which it has been viewed.
Take the Cambridge Dictionary’s decision to make ‘hallucinate’ its word of the year. That’s not a new word, you might say, and you’d be right. Well, this is ‘hallucinate’ with a very specific meaning: it’s when an AI produces false information, so responding to a ‘prompt’ (a word which is on the Oxford WOTY 2023 shortlist) such as ‘write a product description for this phone’ with a set of factually incorrect statements.
That is one of the common pitfalls of Generative AI at the moment: there is an almost infinite amount of data out there for it to train on, but can it decide what’s fact or fiction? Using ‘hallucinate’– a verb that usually has a human as its subject – seems to be ascribing almost human qualities to AIs.
That’s something that the lexicographers at the Cambridge Dictionary point out themselves in their analysis of corpus data: we are now starting to see the verb being applied to AIs as well as humans and also a shift in use from the term ‘AI’ meaning ‘the study of artificial intelligence’ to AI as a countable noun: an AI, many AIs, a world on the cusp of a dystopian future at the hands of all-powerful AIs? Is this android really dreaming of electric sheep?
The Upside Down: Saint Cuthbert – Wonder Worker
John Mitchinson reflects on why the life of Saint Cuthbert still has important things to teach us
John Mitchinson
The human lens also comes into play with the leading US dictionary, Merriam-Webster’s 2023 WOTY choice of ‘authentic’, which it admits is ‘hard to define and subject to debate’ at the best of times, tending to lead people to look it up, thus alerting the dictionary-makers to a spike in interest. How can we be our ‘authentic’ selves? What does ‘authentic’ even mean in a world where many people in the public eye – especially on social media where ‘authenticity’ is a whole look – are performing different versions of themselves all the time?
But it is the broader influence of technology and AI again that shows how appropriate ‘authentic’ is in these not-so-best-of-times, where people are increasingly asking what’s real and what’s fake online (‘deepfake’ is also on the Merriam-Webster 2023 shortlist), and it comes after its 2022 winner ‘gaslighting’, another word that reflects current fears of being misled and duped by what we see with our very eyes. There’s clearly something in the water.
The dictionaries themselves are quite transparent about how they have reached some of their decisions. Far from being the gatekeepers of vocabulary that many purists and pedants want them to be, the dictionaries source their words from actual usage. They are genuinely responding to what is being looked up and what people are saying and writing, whether that is Macquarie’s experts deciding that their WOTY choice is ‘cozzie livs’ (the distinctly Australian-sounding, but actually very British abbreviation for ‘cost of living crisis’) or Merriam-Webster noting the increased interest in the US for King Charles’s ‘coronation’ (but sadly not the ‘corrie nash’ or ‘chazzle dazzle’). And they are using technology to track these changes.
In fact, dictionaries are making use of the same kind of corpus data (massive databases of language) that Large Language Models like Chat GPT train on. But as the Cambridge Dictionary is quick to point out – never knowingly letting a WOTY announcement not be a marketing opportunity for its own dictionary brand – it still employs humans to check the output and make the final decisions with all its publications.
Don't miss a story
Meanwhile, Oxford has opted to decide its WOTY by a completely foolproof public vote from its shortlist of eight words. And when has a popular vote ever gone wrong? (Aside from last year's Oxford WOTY 'Goblin mode'; the Boaty McBoatFace debacle, and 2016, when the voting public actually voted for both Brexit and Donald Trump?) What could possibly go amiss when superfans of ‘rizz’ for ‘charisma’ might produce ‘deepfakes’ of Taylor Swift to prevent ‘Swiftie’ taking its inevitable WOTY 2023 crown?
The great dictionary maker, Samuel Johnson, probably wouldn’t have been impressed with this populist approach but then while he had a lot of interesting things to say about language, and some famously eccentric definitions in his landmark dictionary, he was never a great fan of popular opinion leading the way. And maybe ‘goblin mode’ tells us he was onto something.