UK blogger Martyn Crucefix discusses his experiments with AI poem creation in his blog post ‘Can AI Write an Original ‘Poem’ By ‘Me’?’
On a personal note, I too (like Martyn) have been fiddling around with AI of late, seeking to create a small picture book as a family gift. It’s set in Spain in 1936 during the Spanish Civil War, & I’d been searching for images of 1930’s vintage trucks … tried google searches of well-known automotive brands for the era, but nothing popped up and figured, maybe AI can do the trick. Which it did, with ease. I’m not fussed it’s an AI generated image, the picture book – if it happens — will be a gift, a bit of fun, paying more attention to the act of gift-giving than to other considerations…. AI has its uses, in my experience.
But it has its downsides too, particularly in attempting to imitate the human – as Crucefix discovers….
Having written this I have convinced myself – even more than I had been before – that what AI has written is a ‘hide’, a cover for a poem, making use of words and ideas that it has found associated with poems (perhaps even my own poems) but which it doesn’t itself ‘get’ (how could it?). On the other side of the desk sits the reader. The question for the reader is: how well do we read any poem that comes before us? Do we accept its (often) feeble gestures towards significance as the real thing? Out of a hundred poems we read in magazines and on-line, how many of them ARE the real thing? I’d bet my AI generated poem would find its way into a UK magazine (eventually). It has the aura of a poem, it has many of the familiar gestures of a poem, it doesn’t really make proper sense (which some think is the mark of a real poem), it doesn’t have the heart of a true poem (but lots of poems I read don’t either because they too are copying, mimicking tropes and phrases from other poems).
Postscript … a podcast chat with Jamie Freestone (substack ‘The Stark Way‘) on Sarah Wilson’s site ‘This is precious‘ a couple of days ago is well worth a listen (though I assume it’s behind a paywall), discussing the projected scenario of AI in 2027 being so advanced that (in Freestone’s words), ‘you’ve got this sort of self-improving feedback loop where maybe one [AI] company finds almost to their own surprise that they’ve got something that is now much closer to an artificial general intelligence, or an artificial super intelligence, perhaps. And then you’re worried that it will break free of its constraints or start using deception to do things that you don’t want it to do and it becomes very hard to handle … I think that’s probably the kernel of the AI 2027 scenario….’
PPS not to mention the significant water concerns due to the substantial water requirements of data centres that power AI systems….