Claude to the rescue

What do you do when a vital cover story interview doesn’t show up?

You’ve got the venue, sorted the photography, perhaps sold some ads against it and you’re ready to go, complete with questions and recorder. But no sign of your celebrity subject.

You might panic. You might sigh and resign to running that cuttings job that won’t be half as good. Or bring forward the next issue’s cover story. A range of options would run though the mind of an editor in that situation. But would you turn to AI to help?

There are plenty of ways you might legitimately use AI around a magazine interview. You might ask it to suggest questions, or angles that haven’t been much covered. You might use it to transcribe the audio file afterwards, and then maybe search that transcription to find the quotes you remember best.

By now, editors are starting to get the hang of what AI is good for, and not so good for. Many are getting trained in it now, and they’re discovering how to use it for real, how to get better results with better prompting and so on.

The more mundane the better. It’s good at checking copy against style and suggesting improvements. It’s pretty good at editing down articles for other channels, repurposing for other media. It can sometimes come up with good headlines, standfirsts and meta data descriptions and titles for better SEO. And when it’s not turning your English audio into Welsh (yes, that’s really a thing), it’s good at transcription. Translation too, into Welsh or anything else.

Some colleagues and I recently built an experimental agent to turn print articles into digital versions. In a previous job, I made something similar using my own coding knowledge of PHP, Javascript, HTML5 and XML. It took perhaps a week in total, maybe more, and thousands of lines of code.

AI can do the same thing with one side of A4. But it does it better, because it also checks for style, suggests SEO improvements, and more. And we’ve only just begun.

Where AI does not work so well is in original writing. Give it articles to edit and tell it not to look outside that box and it does a good job. But let it out into the wild of the internet to finds facts and try to create an article and it starts hallucinating. As I’ve argued before in this column, I think that one day it may be able to reliably produce sensible articles on simple subjects, but that remains a long way off for now.

AI still always needs a human on the loop. For AI creation, that means an editor has to check every single point, which is too time consuming. By the time you’ve done that, you might as well have researched and written something better from scratch yourself.

We’re now hearing the apocryphal stories of freelancers taking a chance, submitting an AI article only to be rumbled by a sharp-eyed sub editor and losing the gig.

Plan b

But in a strange twist, now we have an example sanctioned by the editor, when Esquire Singapore’s interviewee, actor Mackenyu, failed to show. They explain what they did at the top of their article thus: “We had the photospread, but nothing directly uttered by the 29-year-old. With a driving need for a feature, we had to be inventive. Harnessing our creative license, we pulled his verbatim from previous interviews and fed them through an AI programme to formulate new responses. Are these the words we expect from Mackenyu?”

This is odd on so many levels. They weren’t passing off – they were quite up front about what they were doing.

But the weirdest thing to me about this is the strange trust in AI. Readers may remember I’ve asked AI to write my column both here and in my previous technology magazine, but the point was to test AI as media tech, rather than aiming to produce a real substitute.

This fake interview is supposed to be worthwhile in its own right. Esquire Singapore presumably wouldn’t use a human to make up an interview and say so. What would be the point? But the fact they gave it to AI and thought that would be better somehow shows an odd faith in AI, like they assume it has more insight than a human writer. Perhaps it can do. Tech giants are starting to train AI to produce avatars of their top executives so they can ‘be’ in more than one place at once.

But this Esquire Singapore interview wasn’t with an agent approved or trained by the real subject. I’m hearing from journalists concerns that email interviews are sometimes starting to come back with responses that read suspiciously like AI answers. But those at least are agreed by the subject. These weren’t.

It’s all very strange, and while they shouldn’t have published any of it at all, they certainly shouldn’t have used the bit where the AI Mackenyu makes up something about living up to the legacy of dead father, the actor and martial artist Sonny Chiba. Looks like they realised that too, as it’s now been removed.