Can AI write better poetry than most people?
With so many terrible human-generated words out there, is it any wonder that people are using machines instead.
Every year Master of Malt, the online retailer I used to work for, has a poetry competition for Burns Night. All you have to do is submit a poem related to whisky and Burns and the best one wins a bottle of whisky. In my six years at the company, it’s about the only good marketing idea I ever had. Some years we’d have less than 100 entries, sometimes up to 200. As you can imagine, the quality was variable. There was much rhyming of ‘whisky’ with ‘frisky’ but overall it was fun and we often found some genuine talent or just things that made us laugh.
But last year something changed. We had over 2,000 entries. Had our customers suddenly been bitten by a love of the bard? No, they had discovered Chat GPT. This made judging the competition nearly impossible. We did automatically disqualify any poem that used the phrase ‘amber hues’, a favourite of the machines, but we still had to read hundreds of poems between a very small team. My worry is that we missed some good authentic one in the deluge of words.
Funnily enough, the computer-aided ones were often better than the human ones (Giles Coren from The Times put on a poetry night which mixed computer-generated poetry with the real things). Not as good as best but they rhymed and scanned and trotted along at a merry pace. It was only when I read a couple that were eerily similar that I became clear what was going on. After that, it became easier to spot the tell-tale signs.
Writing poetry with a metre is clearly something that machines can learn to do. It really has more in common with maths than literature. And that made me think, what other kinds of writing can machines do better than most of us?
Long before large language models (LLM) like Chat GPT existed, people were writing formulaic, essentially meaningless stuff. Using the word ‘content’ as a catch-all term for what used to be journalism, copy or literature was a sign that the work of writers was being debased… often by themselves. I’m thinking of things like listicles, ‘top 23 bao restaurants in Neasden that you have to go to now’, which were so big in the mid 2010s. I’m not pointing the finger, I’ve written my fair share of these. For any work guided by the mysterious ever changing rules of SEO, why not cut out the middle men and just get the machines to do it?
There’s oceans of human-generated babble that could easily be outsourced to machines. Early this week I received answers to questions I sent out to a vodka brand for an article in the drinks press which could have been generated by AI - though it’s hard to tell. It might have been real hand-crafted corporate guff. I mean think of how our politicians speak, the jargon of soundbites tailored to a media landscape that doesn’t exist anymore. They don’t sound human. Would anyone notice if Britain’s current prime minister was replaced with an actual machine?
I think it goes some way to explaining the decline of Hollywood. So many of today’s actors are so ozempiced and botoxed that they look like a different species. I noticed this as long ago as 2009's ‘It’s Complicated’ with Alec Baldwin, Diane Keaton and Steve Martin. Once the funniest man in Hollywood, Martin couldn’t raise a smile because his face didn't' move. All those ‘influencers’ (glamour models) on Instagram who are losing their livelihoods to computer-generated versions of themselves already look like cartoons.
I will admit, I have used large language models (but I didn't inhale). I use Claude to check my work for articles for grammar, clarity and spelling mistakes - though I probably ignore about 80% of its suggestions. It always tries to remove my jokes. In this case it pulled me up on:
‘The example "top 23 bao restaurants in Neasden that you have to go to now" works but is oddly specific.’
Isn’t it uncanny how much like a human editor it is? Largely, I’ve found LLM useless for research because they tend to make things up though again that’s something that real human writers have done for a long time.
But even if/ when AI improves, I am very wary of using it for anything but routine technical work. Some people tell me that it’s good for ‘ideation’ but I think this is a dangerous step for a writer to take. You’re losing what makes you unique, your point of difference with the machines. Your thought process is the most important thing, humour, opinions, faces with character that move.
I've judged a number of contests as a literary editor and now work on a publishing project that incorporates AI--a product that uses AI to find and share _human_ writing. Here's what I'm (currently) certain about:
1) There was already near-infinite levels of "content" slop
2) There's now even more infinite levels of "content" slop
3) Because of how AI is built, it's inherently "backwards-looking" and focused on "generality" and thus not meant or able to be truly "creative"
4) At least at this point, I don't see any indication that AI is going to reach a level of creativity and inspiration that leads to a moment of singularity/individuality (https://www.routledge.com/The-Singularity-of-Literature/Attridge/p/book/9781138701274) OR that anyone's really working to do that
5) It's a lot of fun to have Claude generate funny poems quickly to share for occasional jokes
In any wine critic ,I want originality,attitude as long as it is tempered with some empathy, humour,an ability to tell a good story and to communicate in well turned sentences.
AI may give the correct answer ,but the words are leaden and joyless.
I welcome good wine writers into my inbox because they are reliable and I trust their judgement.
Here an example of a well turned sentence:: Wine is merely a pause on the journey from grape juice to vinegar.
Generated by me and not a Bot.
I try to aspire to the late great Clive James whose weekly TV column in the Observer was a master at the top of his game.