Here’s another post about AI-generated stories I wrote on Substack first before sharing it here.
Not so long ago, the indie writing and reading world erupted in uproar when reading one of the winning short stories from the Commonwealth Foundation competition.
The Serpent in the Grove, by Jamir Nazir, published in Granta magazine, raised all sorts of alarms. In fact, the reaction has been so violent that even Granta editors felt the need to explain that they have no input in the judging of the stories. The second part of Granta’s preface commentary refers to the speculations that some of the stories may have been AI-generated to some extent. As they state, until there is clear and irrefutable evidence, the suspected story will remain on their website, as per their agreement with the Commonwealth Foundation.
Now, this is an interesting statement to say the least because Granta doesn’t usually comment. The statement remains diplomatic and carefully avoids taking sides. But it speaks volumes precisely because it allows Granta to wash its metaphorical hands of taking any position: “It may be AI, or it may not be. We don’t know, and we’ve got nothing to do with it.”
Ultimately, Granta has no role to play in this. The magazine provides only a platform to showcase the short stories that made the final cut. It’s like blaming the washing line for the bad fashion sense of the clothes hung on it. It has no bearing on anything.
Where Do The AI Allegations Come From?
Some structural elements are frequently seen in AI-generated texts.
- It isn’t that or that, but this
- Quiet qualification
- Frequent enumerations passing as depth/wordcount
- Frequent em dashes (or are they en dashes?) incorrectly used
- Descriptions instead of narration
- Excessive use of disconnected and comparative imagery
- Short, disconnected paragraphs
- Popular AI metaphors that carry on throughout (bees, hum, belly, serpent, rum speaking, etc.)
- Frequently occurring AI phrasing
To name but a few elements.
The thing about AI-generated texts is that you need to have come across AI texts in the past to recognise the signs.
Others who may not be able to name potential AI markers clearly describe the short story as long social media content. It reads just like a LinkedIn post and makes you feel just as angry as reading one of those. Perhaps the subconscious spots the signs so clearly that all is left is how the text itself makes you feel. Angry for wasting your time.
Honestly, who wouldn’t feel angry? You only need to spend a few minutes on LinkedIn to be bombarded by the opinions of the latest influential thought leaders.
Interestingly enough, there is some common phrasing and text structure between LinkedIn posts and Jamir Nazir. It’s no surprise, as most of these posts are AI-generated to the point where users are complaining about it.
Thought leaders, while they are using AI to do the thinking for them, do exist. They simply happen to be leading prompts only, rather than thoughts.
But when it comes to Nazir, questions remain currently unanswered. Is the man real? The latest hypothesis is that his digital existence could be entirely AI-generated.
Are AI Suspicions Correct?
There is nothing more concerning than measuring the veracity of AI suspicions.
On the one hand, there’s this sense of fear that the machine is coming too close to mimicking human beings. Yet, we can all agree that close to human writing is the last possible description we would give of an AI-generated text. Why? Because even if you can’t pinpoint the obvious markers, you can’t fake how it makes you feel. Many readers don’t know why, but they can still tell when the emotional narrative feels artificial. It’s almost as if it lacked understanding of both narration and emotion…
However, at the same time, this brings a much more difficult question about the AI witch hunt. Where do we stop, or more exactly, where do we start suspecting a text is AI? This goes far beyond the matter of The Serpent in the Grove. It’s an everyday problem.
Should any text using em dashes be suspected of being written by AI? Is any poor use of metaphor an AI red flag? Does bad writing constitute evidence of AI use? The logical answer should be “no”, but unfortunately, the logical answer also requires effort. One has to read the text and think about it. Why bother when there’s an easy way out? We could simply use an AI detector to find out.
AI detectors are nothing but one AI tool looking for the presence of another AI tool. How accurate are they? It varies, depending on the detector you choose to use. One AI detector has returned 100% AI results on an extract from Mary Shelley’s Frankenstein. Not all detectors, but one did, and that is precisely where things get tricky. Detectors are not even producing the same results, which would imply they are not using the same detection factors.
Additionally, what is the consequence of running a creative text through an AI detection tool? Doesn’t this further train the LLM behind the tool with a new text, potentially violating the precious copyright and reproduction rules?
Finally, some AI markers are also part of human writing. Ultimately, LLMs have been trained on human writings, some from classics and other readily accessible books, and most of the rest found on platforms such as Wattpad. LLMs have learned to identify patterns and statistically determine which writing patterns are more likely to appear next. To copy one of the preferred ChatGPT phrasing: It’s not creating writing; it’s high pattern recognition and imitation, devoid of any intelligence. So, unfortunately, recognising a few AI red flags in a text may not be sufficient to declare it AI-written.
That being said, The Serpent in the Grove has a lot more than just a few markers. It consists almost entirely of markers, arranged at random in the most statistical approach, and yet the least human way. But who knows? Perhaps, behind all this. Nazir is a real writer who has been booking all his writing retreats at the ChatGPT hotel.
The Actual Damage of AI
The AI witch hunt doesn’t seem as aggressive as we may think at first. With hundreds of thousands of writers, indie and agented ones, the number that has been AI-flagged remains fairly low. Did some escape the vigilance of readers? It’s hard to believe. There is no harsher critic than a reader.
What’s alarming, however, is the dual damage of AI, both to our cognitive processes and to the environment. Turning to AI for writing a text, whether a short story or a LinkedIn post, can’t be without consequences on your own ability to think, to create, and to write. It atrophies your cognitive processing, maybe even faster than dementia. Who knows? Perhaps the brain health risk of the future is not Alzheimer’s, but AI.
As if permanently leaving our brains understimulated for the sake of GPTing our way to zero cognitive ability wasn’t bad enough, it’s also using a humongous amount of resources.
Data centres are the foundation for AI services and technology. The more we use AI, the more data centres we need. By 2027, global AI demand may consume more water than Denmark does in one year. This is without mentioning the GHG emissions, rare elements and mineral consumption, and electronic waste. ChatGPT itself needs 10 times as much electricity to answer a question as a single Google search. Imagine how much more it requires to write a book.
And in the end, the real question left to ask is whether there will still be a planet for us to live on and read the next AI-generated story if we carry on in this direction.