07.07.26
Composing the film soundtrack · an experiment with AI
I reconnected with my composer yesterday, it sounds like he’s emerged from a really intense period - sixteen hour days, seven days a week, for 6 weeks on a major project. I detected that he found it energising as well as exhausting, sometimes those sensations walk together, paradoxically.
He observed that there’s nothing like a pressing external deadline to bring matters into focus, and one of the difficulties that I have on this project is there is no external third party to keep me honest, which does mean that things can drift. I’m always conscious that the absence of an external deadline can become a problem, but this is a film that I have wanted to make for a while, and I want to take the time to do it justice. Although external deadlines force the discipline of completing the work, they can also mean that you make compromises in order to deliver on time, rather than to the highest standard that you are otherwise capable of. So I think an imposed deadline can be a mixed blessing; as the great Douglas Adams once said, “I love deadlines. I love the whooshing noise they make as they go by”.
My composer seems enthusiastic to get back on board, it sounded to me like he needs a lengthy holiday, but I’m grateful for his enthusiasm. We’ve organised a studio session for Friday, where we’ll review the film, talk through some ideas, and then he’ll put together some working tracks while I plough through the picture cut.
Yesterday I set aside some time for an interesting foray into AI, making use of Anthropic’s Fable model while it’s still available within a standard plan. As from tomorrow Fable will no longer be available on subscription, you’ll have to pay for it separately.
My relationship with AI is somewhat ambivalent. I’m not one of the doomsayers, who are intensely hostile towards the technology. I see the downsides and, as with a lot of the technology that has emerged over the course of this century, I’d much rather that some of it didn’t exist. But complaining about it is like complaining about gravity - it’s not going anywhere. I have a curiosity towards it, and I’m interested in exploring parts of its capability.
As much as I’m not one of the doomsayers, I’m also not one of the evangelists. I don’t think that AI is the answer to everything, certainly in its current form, or that it’s going to replace human competences wholesale. In most cases where I’ve made use of AI, I’d rather have a human on the job. I would never let it write for me, for example; not just because, for me, the act of writing is an end in itself, but because it produces a particular kind of writing that I don’t enjoy. We all see it saturating online writing, and it feels mechanical, lacks a human voice and texture.
If you ask it to evaluate any part of your own writing (I ask it to check for typos/errors), it will instantly seek to adapt your writing into a series of formal conventions which I think correspond to bad writing, and fail to reflect the rhythm and tone of your own voice. Where I’ve seen people argue that you can ask AI to write in your voice, I don’t think it can, if you have a genuinely distinctive writing voice.
So I don’t believe the claims that AI makes for itself, but neither do I catastrophise about it. I find that it can be a useful tool in assisting with busy work, basically. I find it’s very useful in terms of certain processes; I use it for research processes, I use it for formatting and organising. As an example, there’s a whole reservoir of very dull Photoshop work that I regularly have to get through - changing frame size, applying borders and presets and so on - I have set all that up in a pipeline which the AI bashes through automatically. Traditionally you might have an intern perform these tasks, so I can see that it creates a problem with training the next generation, but it is nevertheless useful in clearing out a load of hamster wheel jobs that I’d rather not do.
But what I wanted to experiment with on Fable, before Anthropic raise the drawbridge, was strategising. I find that AI is limited if you ask it to think either for itself or for you. It’s very good if you ask it to research and deliver information that exists out in the world, like an enhanced Google (although you have to be very careful in the prompting to ensure that it excavates its information from high quality sources, with proper references that you can check), and what it can be especially good for is evaluating or extracting patterns from your own thinking; I wanted to see how Fable performed in that respect.
I would never ask an AI “what should I do?” or “should I do X or Y?”. Instead, I wrote a lengthy briefing note in which I went through all of the various business ventures that I’ve embarked on over the past couple of decades, the different products that I’ve launched, the film series and catalogues that I’ve produced, and I recorded my experiences of and thoughts about each one: lessons that I’d drawn, what the outcomes were, where the successes and failures lay. I then made a list of what I perceive to be my strengths, as well as all of my weaknesses, and I wrote down all of the ideas that I’ve had over the past six months to generate revenue alongside my artistic work (fifteen altogether).
I did not ask it about my artistic work at all, I think that’s not the role of AI; I kept it to the business and revenue layer and developed a large, complex prompt. Using AI like this is not a quick process, it is not a shortcut. I was attempting to get it to do something which I might otherwise struggle to do myself, particularly now that I’m a solo operator. Previously, I would have had brainstorming sessions with colleagues and my business partner, and AI is no substitute for that, but AI can access a wider pool of information than people sitting in a room together can, and perceive patterns and nuances that might otherwise prove elusive (confirmation bias is, of course, a risk in both contexts).
In any case, I’m currently by myself so I don’t have access to regular brainstorming sessions, the AI has to perform that role for me in the immediate term. To restate: it was not a short process, I spent hours preparing briefing notes and documents, organising all the information to give to Fable, which then chewed on it for a decent amount of time.
What it did was to surface the knowledge I already had, and organise it for me. It parsed my information dump and found recurring patterns in both my successes and failures; it identified common themes and extrapolated a set of principles from them, which it then tested my current ideas against, made some recommendations, and proposed some action steps. I’ll now take those recommendations and discuss them with a human as the next stage.
It’s easy for your thoughts to become tangled and circular, particularly when they’re churning huge tides of time - businesses that I might have started twenty years ago, washing in with some drama that happened four years ago; it’s easy to become overwhelmed by all those experiences, tied as they are to memory. Getting the AI to interpret it all extrapolated patterns that I don’t necessarily perceive. All those patterns emerge from me and my experiences, not from the AI, but the AI helps to clarify them.
It was an interesting exercise, one that made me a little worried. The conclusion was that the path to making a new living from where I am today is going to be lengthy and arduous. There are no alternatives, that’s just the truth of the situation; the AI certainly did not blow smoke up my arse. Anyway, I now have some shape to my thoughts around that question.
Onwards.