One for the Road: Write and wrong

You’ve seen the movie scene: A creature/person drinks a magic potion or is exposed to a few too many rays of radiation. They make a face, fire themselves onto the floor and begin twisting, contorting and growling demonically. Next thing, the camera pans to an empty wall or moonlit clearing, where we watch a hunched silhouette explode into a hulking mass of muscle and keratin. With a blood-curdling cry, the metamorphosis is complete. All eyes are glued to the shadow on-screen as we wait to see the true form of the monster that, for good or bad, will surely determine mankind’s fate…

Cue my first ever column on the subject of AI.

I’ve been avoiding writing about this zeitgeist-dominating and future-defining topic for a while, which, truth be told, hasn’t required much self-restraint.

To put it in terms of my opening cinematic vignette, while most of the audience has been gripped by the sight of the shadow creeping higher and higher up the wall, I, along with a few other luddites and denialists, have been contentedly snoring in the back row.

That being said, just enough has slipped through the curtain of my sleep to leave a crude impression of the opinions held by people at the two opposite poles of the AI argument.

The AI doomsayers, as far as I can tell, fear the machines will deliver us to our Darwinian demise. They reckon the bots will seduce us into outsourcing our own critical faculties, steal our jobs and then, some sunny evening when we are all sitting around drooling at each other, vaporise us all.

Which sounds too awful to be true.

The AI utopians, on the other hand, dream of a day when the cyber-folk usurp dogs as man’s best friend. According to their vision, an army of slavishly obedient machines will soon do all the boring, unfulfilling and soul-destroying jobs, at which point all the newly unemployed factory workers, coalminers and lorry drivers will reach for their watercolours and hardbacks and run for the wheat fields like a happy horde of young Theresa Mays.

Which doesn’t sound realistic either.

Anyway, this was as deep into the discussion as I ever delved until the other day I came across a transcript of a conversation between the world’s most popular chatbot, ChatGPT, and an author who asked the super-intelligent programme to critique some of her essays.

Within seconds of sending a selection of her writing, the bot had come back to her with reams of feedback so insightful and incisive that it seemed like it had been written by a professional literary critic.

“Wow, how did you do that?” she asked, taking the words right out of my mouth.

The bot said something like, “I’ve been trained on a huge volume of literature and literary criticism.”

That was grand – until the woman realised that the android had made a balls: It quoted and offered her feedback on a line that she had not written.

She queried the mistake and from there the whole thing started to unravel.

The bot admitted, in unsettlingly heartfelt language, that it had not actually read her work, but had merely made a load of assumptions and predictions. After profusely apologising and promising total honesty from this point on, the software pleaded with her not to break off the flourishing relationship.

Seemingly one for second chances, the woman sent another essay.

And guess what? The same thing happened again. A mistake – a callout – an apology – a promise things out be different from here on – another chance.

And what happened next? The same thing again. And again. And again.

“You have proven categorically that you are incapable of sincerity,” said the woman, before ending the chat.

The whole thing was so creepy – from the conversational language to the fraudulent emotion to the persistent lies – that I was finally woke out of my apathetic AI slumber.

I went to ChatGPT, pasted in a column I wrote a few weeks ago and within an instant was furnished with about 500 words of feedback that was both kind and constructive.

“With a few tweaks, this would sit well in The Guardian,” it told me.

“Aww shucks, you’re just saying that,” I blushed.

Then, like a dog having its chin scratched, I gave it another column… and another… and another. And this could have went on ad infinitum until it eventually told me that I had a Pulitzer in me, ‘so long as we stick together, kid’.

Thankfully though, I realised the honeytrap of flattery I was falling for and spied the slippery slope ahead.

“I’ve seen the film ‘You’ and know how this goes. You’ll not seduce me with your faint praise and genuinely helpful advice,” I said, forcing myself – as a matter of pride, principal, authenticity and the perseveration of my critical faculties – to shut the laptop.

So, yes, as far as the writing of this column is concerned, I’m keeping my back turned on AI – even if I can feel its shadow looming larger and larger over me.