It’s been a couple of years now since AI really took off, with ChatGPT becoming this widespread, easily available tool available for free* (of course, with ourselves being the product) for people to use. Although early reactions to AI were – and continue to be – mixed, I also find it hard not to believe that we are on the precipice of something that will fundamentally change how we interact with technology. In the same way the internet revolutionised information access and socialising, I think AI could revolutionise creating. Although I played around with AI when it first came out, I stopped using it pretty quickly. I wasn’t really interested in using it and I didn’t think it could do anything that I couldn’t do myself. I still think this is a valid viewpoint, but I’m beginning to find a way to integrate AI into my workflows that makes my work better, without being overly reliant on AI. In this post, I’ll first detail my initial criticisms of AI, then discuss my new workflow, and finally where I think the potential for AI lies.

Criticisms of AI

I have two critiques of AI which seem to be the dominant critiques online also. I will call these the artistic critique and the environmental critique.

The artistic critique argues that using AI hurts artists and using it inherently worsens a creative product. Breaking this down further, let’s start with how AI hurts artists. AI is trained on human work. For ChatGPT, for instance, this means a computer program was fed millions of text-based data. I haven’t seen this data, so I can only guess at what was in it, but it would almost certainly contain things like classic novels and scripts, and I assume also things like forum posts, modern novels, academic papers, legal findings, and so on. If you ask ChatGPT for a summary of a book, for example, and it gives you one that matches what happens in the book, it had to have found that out from somewhere. Unlike humans, ChatGPT can’t think for itself. All it is is your phone’s predictive text feature ramped up: it chooses words based on what percentage it thinks that word is most likely to come next in the sentence. It’s not actually “thinking” like we do, it’s attempting to copy what someone is likely to say in such a situation. Because of this, ChatGPT is an amalgamation of almost everything – and certainly anything important – humanity has created. That means when you use ChatGPT, or any AI software, you are using something that was trained on the works of artists but the artists do not get paid for that use.
Now, the easy rebuttal to this is: isn’t every human artist trained on everything we experience too? After all, my writing style is directly informed by literally everything I have ever read, and you could say that I too am just predicting what word best comes next in this sentence. Try not to think about that too hard! I am bound by the existing structures of the English language, and I can’t just write a sentence like hello glue spark bang because that doesn’t mean anything. It’s only by placing the words with further meaning that they make sense. While I am trained on everything I have read, there’s an important distinction here: I have not read everything under the sun. I quite literally do not have the time for that, it is impossible for humans to read everything ever – we’re just not that quick. So my writing style and thoughts and ideas are unique precisely because they are informed by only what I have read and experienced. I am sure I am similar to some people, but no-one is exactly like me. ChatGPT, however, has experienced everything, and so is no longer unique. Ironically, because it has so much exposure to everything, that makes it harder for it to break out into something unique. You could argue that every artwork is derivative: but ChatGPT cannot think for itself, so it can only ever be derivative, while humans can at least make derivative works that push off in interesting directions.
The second artistic critique is that using AI inherently worsens a creative project. This is tied into the above, because artists are and will continue to lose jobs to AI. AI is cheaper than a human and in most cases is going to produce something “good enough” for whatever a major corporation needs. People feel that AI projects lack soul, and in general I think people will struggle to connect to artists products made by an AI. I also think that in the future, it will be almost impossible to tell whether a film was made by an AI or not, as AI models continue to become more complex. I don’t think this necessarily harms artists, though, but I do think artists will need to adapt. Years ago, virtual studio instruments (VSTs) became incredibly popular in music. These are digital recordings of instruments that you can use on a computer without needing live musicians. Like AI now, there was a lot of concern that this was going to put instrumentalists out of a job. But the market adapted: instrumentalists now actually get quite good paydays and royalties from VST recording sessions, and while almost every major film, TV show, and game has definitely used VSTs, you wouldn’t have noticed. Having said that, even though VSTs are really good, I still prefer the interplay between musician and instrument. So while VSTs are “good enough” in many cases, there is still a preference for real instrumentalists at every stage.

The second critique is the environmental critique. This stems from how much energy use AI requires, which is a lot, to be fair. But what is often overlooked in these discussions is this relative to other power outputs. Training an AI to begin with is the biggest single power draw, and after that, just using an AI doesn’t sap too much juice. Secondly, while the numbers for AI are quite scary, once you dig into them you realise that, relative to other power-sucking things in society, AI is not that high up there. It’s a lot compared to the average house, yes. But the average house is not the main power draw in society anyway. I don’t know that the solution to AI and the environment is to stop using AI. I think there’s other more relevant things available to us. I will also say, there are differences depending in what you are doing with the AI. Asking it to generate a 4K image is a lot more power-hungry than having a conversation. To conclude, while I think the environmental impact is worth discussing, I don’t know that using it as an excuse to ban AI is a smart idea.

AI Uses

I see three main uses for AI: planning, grunt-work, and finishing. What this means is that most of the work is still done by the human, but the AI helps the human to reach their full potential. Let’s explore this in more detail:

AI for planning

AI is a dream for planning. I suck at planning. I hate it. I am much more a go-with-the-flow type person, which shocked me when I realised that too. I hate planning out tasks, I would rather just jump right into the task. At the start of my PhD my task list looked something like: Write PhD, that’s how bad I am.

AI helps a lot with planning. Recently, I’ve been toying around with a novel idea in my head. Going to ChatGPT, I asked it to help me do some planning for it. One of the best pieces of advice I got for ChatGPT is to give it a role in the prompt: for example, starting the prompt with “as a publisher,” or “as an editor,” and that puts it into those locations when it draws on its responses, compared to the more general one it might give you normally. As an editor, ChatGPT helped talk me through my plot, gave me a few ideas of which I discarded most, but then those ideas would spark an idea in me for something else. We even talked about themes in the novel and how I might bring these out. What had started as a basic “this could be fun” novel idea developed into a pretty solid plan where I feel like I could actually write something decent.

Importantly: for curiosity, I asked ChatGPT to do a chapter by chapter plan detailing what would happen in each chapter. And it was terrible! It forgot everything we talked about, it was slow and plodding and yuck. This shows that ChatGPT is not going to be good at doing that further work. But it helps as a brainstorming tool for me to get my thoughts out. It’s like talking about ideas with another person. It doesn’t mean anything, you can reject anything it comes up with, but it’s helpful to just explore it further. Most people I know are already writing out plans: you may as well write it out with someone to give you feedback in case it sparks another idea for you.

Grunt Work

This is the obvious use of AI. AI is great at menial tasks that save you time. This is I think where AI is going to take off. In my academic work, I use Reader. It’s a company that saves your highlights and emails them to you each morning. I love Readwise, and their Reader is an all-in-one document reader inbuilt into their system. It means I can highlight academic articles and these highlights are saved for me automatically should I ever want to visit them quickly, and it syncs automatically to Obsidian. Reader has ChatGPT integration, which mainly comes in handy when I upload an article or document, as it will summarise the document into a short explanation covering the main themes for me. I do think summarising is a useful skill, but it’s not one I want to do every time or for every document I upload.

I’ve seen ChatGPT used to help solve maths problems quickly, provide input on code, and draft email responses (which is great because let’s be honest, most emails are really generative responses anyway built in a particular type of language). This saves you time, so you can get back to the stuff that really matters.

Finishing

Finishing is my final use case for AI. Earlier this month, I submitted an abstract to a major international conference. This conference was a bit outside my area of music studies, so I wanted to make sure my abstract was general enough to be relevant to this broader audience. Having never attended this conference before, I grabbed last years program (which includes the abstracts from last year) and gave it to ChatGPT, before giving it my abstract and the themes of this year’s conference. I asked it how my abstract would fair in this general audience, and it gave me some advice. Did I follow all of this advice? No way! I didn’t agree with some of it and I felt some of it was bad. But some of it was useful – for example, it pointed out a really obvious way for my talk to fit in with one of the conference themes. It does feed the abstract back to you re-written to its liking, but I always ignore this and just look through my abstract and edit it. If we ever get “track changes” AI, then that would be perfect because that’s how I prefer to work, rather than copying and pasting. what it gives me.

The Future

I’m definitely going to be using ChatGPT a bit more this year in my creative pursuits, and I think the above is really all its going to come down to. You’ll notice nothing in the above takes away from my creative agency at all. I am in control at each step of the process. All ChatGPT is doing is: helping me get my thoughts down and expand upon them; doing menial tasks that I could do but are much quicker to do with ChatGPT with the same result; and providing some feedback on my work. At no point does ChatGPT take over or become the dominant voice in the creative process. This helps with the artistic critique. Although ChatGPT is still trained on artists, the fact that it’s not contributing directly to my work helps remove a lot of that impact as the core ideas are still my own unique ones. ChatGPT doesn’t replace anyone either in this workflow, but supplements existing structures. For example, I can see myself getting feedback from ChatGPT on thesis drafts, which should pick up any obvious errors and problems before I pass it on to my supervisors, which should save them time too. The environmental critique remains relevant, but given that this is text-based and only used briefly at these points, it should not contribute significantly to ongoing AI costs. In short, I think this is a use of AI that effectively meets creatives’ needs while engaging appropriately with new technologies.