I suppose I need to talk about AI. Everyone else is. Reactions seem to vary between panic (“We’re all going to lose our jobs!”) and panic (“I don’t understand any of this!”) Well, I’m here to help.
Despite my advancing years, I’m a great believer in technology. It enables me to edit a 4k series at home. It enables me to send my footage around the world. It enables me to research and develop projects without travelling. Tech’s great.
AI as a term has been creeping into the tech conversation for some while, particularly in graphics programmes such as Photoshop. And now it’s everywhere, what with your ChatGPTs, your Midjourneys and your Stable Diffusions.
As it turns out, none of this is actually AI. It’s not intelligent at all. What it is is machine learning. ChatGPT and its chums have read a lot of stuff (and some of the variants can read the Internet as well). And then they tot up everything to do with the words you put down and deliver the most probable answer they can. That’s really clever, but it’s not intelligent.
Likewise with image generation. I’ve been dabbling with Stable Diffusion, one of the more flexible text-to-image generators out there. Again it’s all down to probabilities. It takes a load of noise and reverse engineers an image out of what it could have been before it got turned into noise (according to everything it’s already seen), and your prompts give it a steer. It’s getting quite good at what it does – if what you want it do is make static portraits of people in various styles, or produce some single anime-style images. Or a steampunk cat. I’ve made a lot of steampunk cats. But look a bit more closely – and everything is subtly wrong, even in that image that won a photography prize. For now it’s not all that great, but I’m sure it’ll get better. Can it take on 4k video? That means producing 25 sensible, consistent frames per second. We’re a way off that.
Even when it comes closer to reality, I don’t think it’s going to threaten those of us in factual TV all that much. Maybe I’m missing something, but what’s interesting about factual is that the things we show are, well, factual. Perhaps we’ll all be able to afford an AI Morgan Freeman to present our programmes, but frankly a real presenter is probably going to be easier. So long as our USP is authenticity – and if it ain’t real, it ain’t factual – then AI just isn’t a problem. Real stories about real people, that’s us! And any fakers out there, well, who wants to be the Milli Vanilli of the documentary world?
That’s not to say AI won’t come in handy in the process. We’ll be able to use it to fill in the edge of a shot where a boom pole has intruded – we can actually do that already without AI, but perhaps AI will allow us to do it better and more easily. It’ll provide us with a draft pitch, a draft script or a draft edit. That will be a help, of course: rewriting is always easier than filling a blank page. It’ll give us a first pass at a colour grade. Nothing wrong with that: for years senior production staff have used juniors for the job: maybe assistant producers, researchers, assistant editors will miss out on some valuable experience. But for the process of making programmes as a whole, it’s going to make things easier, quicker and most likely cheaper. That can only be a good thing.
Jonathan Schütz
