When using AI goes wrong, horribly wrong

Over the last week, one story has caused quite the stir in the content world. The New York Times dropped a freelance writer – Alex Preston – after it transpired that he had used AI to write a book review, a review that bore striking similarities to a Guardian review of the same book published months earlier. 

No doubt he believed the AI model was producing answers that merely summarised points in a new way; it surely wasn’t intentional. But even the thought of using AI for a review that is meant to be offering your personal response/reaction/opinion raises multiple alarm bells. Preston wasn’t just any regular run-of-the-mill reviewer – he’s a six-time author with a whole backlog of published reviews.

Preston’s reliance on AI didn’t just expose a lack of critical thinking, it was a case of blatant, if unintentional, plagiarism. It replicates what schoolkids used to do, copying another person’s work and whacking a thesaurus out to change a word here or there. At least it gives PR firms a clear and pertinent example to exemplify what’s already been communicated since the advent of AI: opinion pieces require your opinion. Obviously. 

A reverberation of existing opinions

The catalogue of media and social media activity in response to the news has, of course, rightly, picked up on the dangers of AI reliance or overuse, its proclivity to spout answers that resemble a regurgitated set of opinions that already exist and have been published online. Or, in this case, to pretty much directly quote other people’s work without attribution. Again, it’s that all-too-familiar, yet all-too-ignored, concept of how AI’s used and what for. 

A lot of people utilise AI to help them write daily communications, to spark ideas, and so on. In the tech sphere, if you’re writing about a technical topic, AI can process vast swathes of data and provide concise bullet points of information that you wouldn’t necessarily have discovered otherwise (although even this process is prone to hallucinations and misinformation – always check your facts!).

What AI doesn’t do is add new angles into the conversation. It doesn’t have an emotional and visceral reaction to topics of debate, or observe events in the real world, and it of course can’t give someone’s personal response. There’s a difference between writing a straightforward FAQs blog and an opinion piece. There’s also a natural rhythm to someone’s writing voice that can be diminished with AI. Imagine if columnists at national publications were using AI in the opinion section?

When writing, you often research others’ views and opinions on a topic, some you agree with, some you disagree with, to inform and then insert your own viewpoint into the discussion. Total AI dependency removes that nuance and judgement. 

Meeting and setting PR expectations

There’s another question simmering under the surface of all this: why did a writer in his position feel the need to use AI for a review at all? That raises a whole stream of discussion points around workplace and societal expectations, of taking shortcuts and not checking information, of critical thinking skills flowing down the drain. But the answer is probably far simpler in the tech/business world.

Not every tech leader is a writer, or has the time to write. That’s not the problem. But they do, or should, have strong thoughts, insights and stories about their role and sector – what they need is a writer who can then write them up. There’s also a direct business benefit of articles too. AI models crave new opinions, not other AI -generated ones. If companies want to appear in AI searches, then they need human-led content. 

More than anything, the NYT story is an illustration of trust being eroded. With a review, there is a dialogue between the person and/or their art being reviewed, the author of the review and the reader reading the review. This relationship is maintained on the basis that both parties on either side of the author believe that it is said author’s opinion on it. 

Tech leaders should see writing opinion pieces in a similar way. Readers trust that they are hearing their personal view on the topic or trend in discussion. What’s the point in using AI to form their opinion, when readers could simply search for that answer themselves?