When I came to write this column, I opened a blank Word page and was for the first time presented with a new start message at the top: ‘Select the icon or press Command+L to draft with Copilot’. Tempting... but no.
It’s the latest AI assistant to appear built-in to the software tools we journalists and editors use every day. AI has caught on quick. It’s hard to find an editor that hasn’t yet used it for something or other: transcription, translation, summarising, proofing or some other, probably quite dull, mechanical task. And editors are also dealing with the effects of AI, especially on their websites as more users are using AI search tools and summaries in their everyday browsing. AI is here and it’s going to get better and bigger.
However, it is not so useful yet for generating written articles. It’s yet to show its creative writing side, although I think it will in time. You can eventually wring a piece out of it but by the time you’ve given it enough prompts, further steers and corrected its hallucinations and made it talk sense, you might as well have written the piece yourself.
But I have recently been using AI for image generation for a special project and the results are interesting. Its early attempts, when generative AI first appeared in public, frankly looked weird. But it has improved.
It has huge potential in visuals and it’s already well embedded in the media business of images. The big picture libraries now have plenty of AI imagery which they are rightly labelling as such and can be easily filtered out which is what I usually choose to do. My current magazine is about the solid infrastructure around us and so that’s usually what we seek pictures of, rather than generic illustrations for more abstract themes. But that’s where it comes in more useful.
After a few scandals where AI images won some arts awards, they are now finding their own place. There are now special categories for AI images in media and photography awards although, of course, an AI image is not a photograph by definition. The best of these are fantastical, extraordinary and quite distinctive. This was always destined to be its strength. How many of us remember clustering around a light box for a feature treatment and concluding with a comment like: “What we really need for this one is a picture of a rainbow coloured camel in the desert, with a car seat on its back and it’s head in an oil drum, on a rainy day.” Or some such nonsensical imagining to try to illustrate against the odds a dry and abstract article on the latest management trend or financial tech. “But of course we don’t have the time or money for an illustrator so we can’t have that...”
Only now, with AI, you can. And AI is finding its place in the media with an approach that’s outlandish, colourful, joyful, fantastical, extraordinary and a lot of fun. Or that’s where I see it succeeding. And as we needed visions of future transport, I thought I’d try it for that.
Imagining the future
We’d never use it instead of real photographs. But photographs that haven’t been taken yet because the subject doesn’t yet exist... that’s excusable.
Getting these fantastical images though takes time to get all the prompts right, to correct the hallucinations that it can be quite stubborn about, or to just smooth out the wrinkles.
I found it’s quite good at trains, but not in understanding that railway lines usually have several sets of tracks for travel in both directions. Bicycles it really struggles with. It knows they have a rider and two wheels, but not that those wheels are almost always round, or that the handlebars always go in front of the rider, not behind. I wonder if that’s because most image sources out there are views from the front, and AI doesn’t yet get that when the rider is turned away from the viewer, so are the bars.
AI is full of strange mistakes rooted in the data it exploits. And image generation has always had these quirks, one of the most famous being the fish with fingers. Asked to portray a carp, the algorithms scraped all the images of carp from the web and decided carp had fingers on their bellies. That’s because 99% of images of carp on the internet are of fishermen holding up their prize catches, their fingers showing underneath. So, carp had fingers.
AI illustration is not instant; it does take time. But whereas for a written article, it might be quicker just to write the article yourself, for images it rarely is. Speed isn’t all of course, and where you want something artistically better, you’d usually still reach for an illustrator where budget allows. But time and budget don’t always allow.
While the media is exercised about copyright and AI illustration, the bigger problem for the public and politics is fake photography made with AI, by bad actors, dodgy nation states or people in backrooms trying to make a few easy bucks. Or just a social media name for themselves, more followers or likes.
Fakes are strongest where people want to believe. A keynote speaker at News:rewired showed us a YouTube video of a bird with a crown on its head. Not wearing a crown, but part of the bird. Look, he said, even something as absurd as this can fool the video poster’s mum who is not giving up in her determination that this crowned bird is real. This, said the keynoter, shows how far generative AI has come in image creation, and this is an amusing end result, that someone can be so sure that a bird like this can really come out of an egg.
But I wonder: how do we know the video is real? The mother could easily have been coaxed into this act in order to help her daughter make a viral video. Is it a fake about a fake? Welcome to the future of images in the age of AI. You know it’s not real. But in what way?
This article was first published in InPublishing magazine. If you would like to be added to the free mailing list to receive the magazine, please register here.
