Generated by Dall.E

My idiolect – my unique way of speaking – is a bit Yorkshire, a bit Bradford, a bit South West London, a bit literary, a bit climber/surfer/general-alternative-sports-speak, a bit too much consultant and business-speak. I go out of my way not to say "just" (because I once read an interview with John Lennon who said it was a waste of a syllable, and that really stuck with me). I picked up the word "heaps" on holiday in Sydney and can't seem to shake it. My idiolect is a bit of my brother, a bit of my uncle, and a bit of my dad (worryingly more like my dad now I'm a parent). But it's all me. The language we use is a pure expression of our identity and, like our identities, is made up of our communities, our roles, and our relationships.

But now, the role of language as a marker of identity could change. Two things are driving this shift. Both of them are to do with AI.

First, the algorithmic audience that listens to us makes us change the way we speak and write online. CEOs avoid certain words during earnings calls to avoid triggering the algorithms that crawl their transcripts on behalf of analysts. Taylor Lorenz has documented the efforts of creators to get around platforms' content moderation policies by figuring out which words get their content suppressed and replacing them. 'Dead' becomes 'unalive'. 'Sex' becomes 'seggs'. Where moderation is heaviest, whole new dialects emerge.

Elsewhere, some people are using tone indicators to help people grok the meaning of their text when their content is surfaced out of context by algorithmically driven feeds. Writers add a tone tag at the end of their message. '/s' to show sarcasm. '/j' for a joke. Certainly there are a number of politicians who could benefit from attaching these to their tweets. /srs

And DoNotPay oh so nearly brought an AI lawyer into a courtroom. It would listen to the proceedings and make suggestions to the defendant on how to respond. But just like the creators outmanoeuvring content filters, I think the opposing lawyers would simply change their language to confound the listening AI, sparking a language arms race between lawyer and lawyer bot as the resulting transcripts are fed into the training data.

Already we're changing the way we speak for an unseen audience. And then there's the identity thing.

People are already being accused of being bots online. Granted, it's usually when people disagree with the views being expressed, but in an age of near perfect AI imitation, the rot of doubt over people's humanity has set in. LinkedIn have just announced features to verify you are who you say you are and you've worked where you say you've worked. Sam Altman's WorldCoin have announced World ID to "bring global proof of personhood to the internet."

Until some universally accepted system for verifying your personhood emerges, we'll be looking for markers of humanity. One such marker might be our voices rather than our words. Research by Juliana Schroeder and Nicholas Epley finds that "voice may uniquely communicate presence of mind and, ultimately, fundamental aspects of being human". And we're already seeing voice notes rise in popularity as a way of communicating between young people seeking a more intimate way to connect. (A quick aside, a few years ago I spent some time talking to a young woman whose friendship group sent each other voice notes that were thirty to sixty minutes in length. They'd listen to them on their commutes instead of podcasts. It sounded lovely.) The way that some writers are approaching audio versions of their blogs (like this example on Substack) and essays is also quite charming.

But tools like Eleven Labs do a fantastic job of reproducing your voice. So for me, it's our language, our idiolect, our style, that is the biggest proof point of our humanity. While opponents of generative AI worry that it will drown us in a sea of mediocre, generated content, my hope is that our need to stand out and to express our unique identities to the world will be the perfect foil. We've seen it happen before. Painters (and I give a huge nod of thanks to Space10 for this one) were worried when photography "abruptly automated" the production of realistic paintings. "And the next generation redefined the goal of painting altogether, sparking the whirlwind of innovation we call Modern Art."

At worst we'll see a kind of linguistic peacocking: deliberately outrageous or flamboyant language to provoke a reaction. Disregard for grammatical rules and norms as people seek out a style which cannot be simulated or emulated. Dating profiles could be one place this kind of language evolves, as people seek to stimulate connection and prove they're putting their own, authentic self on the page.

We'll see a renewed competition around neologisms, the coining of new words and phrases, a competition which I think has been hotting up for years as brands compete to be the go-to verb in a new market. When asked, ChatGPT actually suggested the phrase "virtual babelization" to sum up the impact of LLMs on human language. I had to look up 'babelization', which means "to confuse by mixing or mingling divergent or distinct languages or cultures", but I think it does nicely capture a risk posed by this linguistic competition. Then again, new terms spread so quickly on platforms like TikTok that my worry might be misplaced. And I think this will be the defining feature of language in the coming decade: its rate of change.

Subcultures develop their own language as a marker of membership. It's deliberately exclusionary. I still remember being baffled the first time someone asked me what the 'delta' was between survey results for two groups instead of the 'difference'. (Ah, consulting.) But if I can ask ChatGPT to write my email copy as if I was a member of a subgroup, I can pass as authentic without earning that membership. So, in defence, the subgroups will increase the rate at which they change the language they use. Only by investing enough time in the community will members be able to keep up. This plays out amongst teenagers already, but it will accelerate and spread beyond that demographic.

Machine learning algorithms as audience and generative AI as creator will push us to rapidly, continuously evolve our language in order to confound our machine listeners and to prove our humanity to our organic audience. But as we evolve our use of language to prove our personhood, will it no longer reveal who we are?