What AI can't do - how Nicola Sturgeon and Robert Burns mash-ups prove artificial intelligence is 'slop'
AI is writing biographies supposedly produced by the likes of former first minister Nicola Sturgeon - which shows how much this ‘artificial intelligence’ can’t do, writes Ewan Morrison.
It was only a matter of time until some bright spark tried to make a fast buck using AI chatbots to write biographies and histories of Scotland’s politicians.
So, a biography of Nicola Sturgeon, two on John Swinney and a book on Humza Yousuf have all been rapidly published on Amazon and then equally rapidly pulled from sale for violating Amazon’s guidelines.
They contained “bizarre sentence formulations”, they were “error strewn” and full of false claims. Who knew, for example, that Yousaf came from poverty?


Well, what did the “publishers” expect? They were using the “slop” generating machine that is called artificial intelligence (AI).
I’m going to refer to these as large language models (LLMs) in what follows, as AI is a misnomer. These algorithms are not intelligent and all they really do is predict patterns of words through “training” on vast data sets of books and texts they have “scraped”.
These algorithms don’t know what the words they use mean; they don’t understand reason or lies. As they have no mind, it’s also false to call their errors “hallucinations”. The non-sense they churn out can’t ever be fixed as it is built-in to what LLMs actually are.
When Chat GPT 4 first came out, I ran a test to see what its limits were vis-à-vis historical truth. I devised my test, knowing that LLMs have been programmed to flatter the user. So, I told it a half-truth and asked it to expand on it, to see if it would take the bait. An unfair prank? Not really, as LLMs don’t have emotions.
The test went like this. Nine of my books have been published, and so I asked the LLM to give me a list of the twelve books I’d written. I guessed it would fabricate three new titles, to attempt to please me.
Lo and behold, in the bibliography it made for me, I had 12 titles, all with convincing looking publication dates, no less.
But wait, one of the books was titled “The Birth of a Nation” - a name the LLM took from the 1915 American silent drama film directed by DW Griffith - now widely seen as a racist glorification of the KKK. Odd.
And the mashup-machine had only just warmed up because the next falsely named book title was, wait for it –“Nine Inch Will Please a Lady”.
Now, the Robert Burns aficionados among you will know that this is the title of one of the ‘Bawdy verses’ attributed to Burns. It contains such memorably saucy lines as:
Suffice to say, I’ve never written a book with this title or content, just as I wasn’t alive in the 1910s to write a novelisation of a silent movie dedicated to the KKK.


After laughing this off, I studied how LLMs function. I worked out the algorithm had discovered there were three books with bi-sexual content in my backlist, and that I was from Scotland.
Putting these facts together, without brain, humour or malice, it had matched them with the Bard’s title, also mistakenly depicting ‘Nine Inch’ as a novel. You can see why I refuse to call it “artificial intelligence”.
Why do LLMs mix-up the authors of texts? No doubt, it’s because LLMs are trained on copyrighted material, which has been “scraped without consent. For LLMS there are no authors, only billions of word patterns spread across the net, and all fair game. There are currently 39 copyright lawsuits raging against AI companies in the USA.
Anyway, I felt more fun was to be had from the awful algorithm and so I asked it to generate a synopsis for my non-existent “Nine Inch…” novel, just to see if it could “hallucinate” any more non-sense.
And voila, within ten seconds I was told this acclaimed 2015 novel had a protagonist by the name of Tom - a male stripper endowed with great talents, who is afraid to come out of the closet to his parents, even though he’s qualified for the nation’s biggest gay-stripper contest. Sound familiar? Scrutinising the synopsis I detected that it was a mash-up of The Full Monty with Billy Elliot, set within the stomping grounds of Irvine Welsh novels.
I showed it to a friend who said “damn, if you wrote that it’d be a bestseller”. I thought not, and was pretty appalled at how this mashup-machine that feeds on stolen text automatically destroys the truth.


Over the past few months, we’ve seen how this “slop” machine is not just contaminating Amazon publishing; not just filling up YouTube with misleading videos and replacing historical photos with fake Generative AI images; but it’s infecting historical knowledge and the news.
This week Google AI cited that JFK used Facebook and Mother Theresa used Reddit. Last month, the Iranian government was caught using generative AI photos of downed Israeli F-35 jets and B2 Bombers with oddly incorrect dimensions.
I don’t care if people think I’ve written a novel about a closeted stripper, but it would matter if someone were to write a degree paper citing that non-existent book, now that 92 per cent of students are reported to use LLMs to write their essays. Quoting non-existent books makes for a non-existent education.
Not only that, but recent studies show that Open AI’s LLMs have only got worse, with the four newest models showing failure rates on factual accuracy tests of between 37 per cent to 80 per cent.
The slop generated by LLMs might seem harmless, but it’s like sewage. Consider our society as a glass of water then add one drop of sewage to it. Would you drink it? Would you want to incorporate these sewage generating technologies into your education or healthcare system?
Would you ask a so-called AI if you should?
- Ewan Morrison is a multi award-winning Scottish author. His ninth book, For Emma, was published in paperback on June 25, by Leamington Books.
Comments
Want to join the conversation? Please or to comment on this article.