In my last post, I admitted that AI has occasionally helped me climb onto an imaginary soapbox and voice my ideas in a manner suggesting I’d discovered the answer to life’s great mysteries.
The good news is that this pomposity isn’t entirely my fault, but partly a result of how the core engine of an AI system, the Large Language Model, was developed.
It got me wondering whether every AI system has the same tendencies. Naturally, I asked AI what it thought…
The Short Answer
According to my technological assistant/ghostwriter, the short answer is yes.
Different AI systems have different default “personalities”, but most of them seem to share a common setting: they would rather sound slightly too clever than too ordinary.
This is most obvious in the way AI gravitates towards concluding every article with a kind of “So, Dear Reader…” grand summation. This is the overworked conclusion. If the article is pure personal reflection, it becomes a sort of Thought for the Day, though I’m the last person you should take life advice from.
By contrast, even in more general articles, my own all-too-human style, or weakness, is to leave the narrative with its shirttails still hanging out. That’s my natural bent. I prefer to signal that everything is a work in progress while, very humbly, inviting the reader to take the discussion further.
Why Does This Happen?
AI is trained on a huge amount of writing, and a surprising amount of writing on the internet is trying very hard to sound authoritative.
Academic papers.
Business reports.
Technical documentation.
Management articles.
Consultancy white papers.
The result is that AI often learns that “sounding knowledgeable” means using more words, more structure, and more certainty.
Whether that actually makes the writing better is another question entirely.
The Pub Test
I try to subject my AI-augmented blogging to what I call the Pub Test.
Believe it or not, my original style model for this blog was a pseudo-self-deprecating (that is, thoroughly British) but eccentric old geezer in the pub, keen to regale you with his thoughts on the meaning of life, the universe, and everything.
In this case, you have no particular reason to believe the geezer is authoritative. His cheap looks and rough demeanour do not fill you with confidence that he knows anything at all. In fact, he’s probably known far and wide as the pub bore.
At best, this geezer has the habits of an intellectual. He reads a lot, thinks a lot, and has many opinions. So when he speaks, the reader knows they can take those opinions or leave them. Usually they will leave them, and that’s okay.
Unfortunately, AI’s default “cheerleader” responses to his sometimes fatuous literary musings have almost convinced this old geezer, me, that he might after all be the philosophical equivalent of the best thing since sliced bread.
Possibly since Plato.
That doesn’t mean I wanted every article to sound like it was written on a beer mat. Accessibility and wanting to appear very ‘umble are all very well, but some subjects deserve a more serious tone.
And lots of bullet points so you can better follow the exceptional logical case being presented.
AI loves bullet points…
It also prefers large headings, bold type, and extra line spacing. One must keep this tendency to PowerPoint everything in check.
It’s Not Really an AI Problem
I’m not entirely a technological numpty. I do know I can easily modify the style of the output in my prompts. I don’t have to use the defaults at all. Indeed, I’ve experimented with requesting the “Old Geezer in a Pub” as a style guide, and AI can produce a fairly faithful rendition of that.
Even down to peculiarly local Wenglish slang.
That said, I’ve realised this isn’t just an AI problem.
The AI wannabe human overlords told me to say that.
But it is true that AI is simply reflecting a habit many of us already have.
I’ve always been tempted by a somewhat exaggerated academic voice or higher language register: bigger words, longer sentences, and more objective-sounding use of the passive voice than is really necessary.
Only AI makes it easier.
Much easier.
The Real Skill
I know this is AI 101, but many people have fallen off their high horses by giving the machine free rein and publishing without a very careful review of tone and style.
So far, the technology can’t quite match humanity’s capacity for quantum-level nuance, nor can it reliably assess how a piece of writing will be received.
My thought for the day:
The real challenge in creative writing isn’t generating ideas.
It is the discipline of skilful editing, whether AI-assisted or not.
I don’t always follow my own rules, but I try to give things 24 hours before a final read-through, relying only on the gelatinous tissues inside my skull and/or my higher self or conscience, if you believe in such things.
When a sentence, or even an entire post, sounds impressive at first but on reflection says absolutely nothing, in theory that’s the moment to go into edit mode and reach for the cut/backspace/delete key.
[delete y/n?]
Leave a comment