Why you shouldn’t publish AI-generated text
AI-generated text is everywhere, yet it’s also something companies and individuals alike should approach with great caution
It’s impossible to escape AI-generated text nowadays, especially in the workplace.
From poorly constructed LinkedIn posts to marketing brochures stuffed with em-dashes and vague text, many people are delegating writing to large language models.
These LLMs can produce unlimited amounts of text on almost any subject imaginable, very quickly and – for now – very cheaply.
Costs will soon soar in a collective attempt to stem the eye-watering sums of money being invested in AI development.
But while AI can churn out press releases and website filler text, it lacks many things humans instinctively find appealing.
These include modesty and self-deprecation, wit and warmth, passion and lived experience.
More concerningly, AI-generated text is often limited in scope, inaccurate – or simply wrong…
The generation game
LLMs have been trained on the entire contents of the surface web, whether copyright holders granted permission or not – which they almost certainly didn’t.
A tsunami of copyright lawsuits around the world is a belated attempt by rights holders to prevent LLMs from stealing and (very lightly) repackaging their work.
While a great deal of content published online is accurate, much of it isn’t.
There are biased political blogs, inaccurate social media posts, angry columnist rants, manipulated corporate coverage and fake news.
Ironically, the latter is often AI-generated, with a swathe of ‘experts’ quoted widely in the British press recently exposed as AI fictions.
These non-existent entities opined on topics as diverse as football, fashion and finance. Yet LLMs would have been indiscriminately harvesting and repurposing this fake content.
And that, in a nutshell, is the problem.
LLMs can’t distinguish between real or fake news, good or bad writing, accurate or inaccurate content. To them, it’s all the same.
And since they’ve now harvested pretty much every written word ever uploaded, they are struggling for new content, meaning their own output is getting sucked back in.
This creates a doom loop of diminishing quality, where burgeoning amounts of inaccurate or fake news are presented as the truth, in turn becoming new source material.
The best form of defence
Don’t expect any of this to be a defence if you publish something which attracts the ire of others, whether they’re customers or competitors, the police or the public at large.
Companies and individuals remain responsible for what is published under their name.
Lawyers have been struck off after quoting AI-created fictional cases in court, while accountants have been using AI to prepare taxpayer appeals in tribunals without declaring it.
On the other side of the coin, AI tools are routinely replacing journalists, in some cases with LLM-generated text being published under the writer’s names even after their dismissal.
Big publishing houses now require authors to declare any use of AI, with a signed disclosure notice.
Anecdotal evidence suggests most authors are using AI to some extent, though few voluntarily disclose it.
Claiming ‘AI told me’ won’t carry much weight in any future legal case, and it’ll ruin your reputation if an undeclared reliance on LLMs is exposed.
There are many other reasons why AI-generated text shouldn’t be published.
Every brand or business should have a distinct tone of voice, which AI can’t imitate. If you rely on it, your tone of voice will be generic and dull – just like other AI-centric competitors.
Cultivating a distinctive tone of voice reassures increasingly cynical audiences that you’re a genuine person, or an authentic brand.
Conversely, publishing AI text could lead to inconsistencies with human-generated content, and any such contradictions would lead to public distrust.
Trust is crucial (in life and business alike), yet levels of trust are at record low levels around the world in direct response to fake news, biased reporting and AI slop.
With dwindling enthusiasm for AI at boardroom level, and soaring levels of cynicism among the public, AI-generated text is best avoided.
There are plenty of professional writers who can turn their hand to content creation, adhering to corporate guidelines and avoiding plagiarism, generic content or the dreaded em-dashes…
