I need to write in order to know what I think.
I got that idea from the Idle Thumbs podcast back in the day. I really miss that podcast. But yeah, I just kinda wanted to process out loud about things, specifically about generative AI.
I use generative AI for coding. I say this on my site. I've used it for my site and for my game. As a programmer, an artist, a writer --- and most importantly, someone exercising a basic sense of taste --- I've tested to see what it (chatgpt) could do. This is my experience:
To me, using generative AI on very standard coding is borderline where I'd say generative AI is "the future". I personally think that people don't always realize how boilerplate coding stuff at an industry standard is. As I was learning software development patterns (basically one step above FOR loops, lol) and implementing them for my game in gamemaker, I started to realize that every app that aims to function at a commercial level is the same thing: central controller, event handler, data-driven refactors, etc. I'm not talking about indie games that might be exploring cool and interesting aspects of code. I'm saying that it's almost like being a jingle writer for a major corporation: if you want to write a bossa nova, there's literal books spelling out the essential anatomy. It's why I'm not surprised that AI-generated music works with people, because I don't think people realize or understand how much industrial production of music is purely psychologically functional. Similar thing with code.
I also think people also might not understand how much computer infrastructure is built out of smaller pieces. Programmers just think like this in general. You can think of it like the world's biggest factory simulation game. You make machines to process parts; you can use those parts to make machines that make the machines that process parts. Layers and layers of that. When I took programming classes, we learned how to write in C. I don't even konw if programming 101 students still learn C. Basically it's like having high level control over the writing head on a hard drive. You have to code how much space you want your program to take up and how the writing head will look for data. Compare this to python, where you just declare that a variable equals an integer value, and you can use built-in functions to convert that integer into a floating point number without caring about how that affects the extra space that takes up. I feel like programmers are very, very used to thinking about "progress" as "I made the pieces so that I can operate at a higher level". I feel like I see and know programmers who use generative AI for programming that are operating at that level. And generative AI works for that.
Anyway, now that my glazing has chased away anyone who hates AI, I'm ready to vent about how much it sucks.
Generative AI is so, SO bad at making anything with any kind of authorial intent. Art, writing, music, it is SO bad at delivering what you want. If you're in any kind of conversation or connection with artists, it just can't operate at the level of operational granularity. And lord knows I've tried. I've tried a few times to have a basic conversation with it about anything like fixing up some writing or reviewing basic pixel art. Its output is SO lame. I mean that with my full 90s heart, it's the most basic cringe shit I've ever seen.
The only real use I've had with generative AI for actual creative stuff is using it like a reverse google search. So if I'm trying to find some particular style, it's able to get the gist of what I'm going for and give me something plausible back. For example, I was trying to trace the source of the rapid cuts that characterize the Monogatari anime series, and it helped me connect that to French New Wave. But this isn't something unique to generative AI; people on reddit does that. And even then, I've tried to use it for brainstorming ideas and it, again, really sucks. Anything it finds is just way, way too literal, even when it's being as metaphorical as possible. At first I thought it was going to help a lot because it could provide a bunch of examples of different things for me to consider and select from --- me leaning fully into Fountain by Marcel Duchamp --- but this DIDN'T EVEN FUCKING WORK. I never found that it gave me reasonable or interesting options. It was the process of searching and selecting that gave me the options that worked.
Overall, trying to use generative AI and having it fail to meet my standards has made me realize that art, writing, and music are unique skills in that the idiosyncratic process of creation has fundamental effects on the work itself. Someone's lineart is going to be affected entirely by how they hold a pencil, and that will be portrayed in the outcome. The ultimate reality to me is that human prompting is not capable of capturing that level of total universe simulation...without just stealing.
Ultimately, after my experiments with AI, I scrapped anything I got from it. I don't use it anymore for that, because it literally does not function. I think that someone like my dad who is retirement age and who has no artistic background might say that it looks "nice" but it does not function as art. It genuinely takes less time, less frustration, and is less damaging to my artistic process to draw and write my own shit.
Having said all that, I genuinely feel ambivalent about generative AI. The programmer side of me genuinely feels like it's such a relief that after I build a well-known conceptual model, I don't have to do it manually again. Like, I spent the past 2-3 years learning about software development and game development patterns, and I was implementing them manually in gamemaker for my ARPG. I learned SO much from doing that...but also, some part of me dies when I think about having to do another data-driven refactor by hand. Basically, it's relatively easy to write code to, say, control a particular character to move them using a specific set of buttons. But if you want to be able to swap between characters in the game, or if you want to have remappable inputs, you have to refactor the code to work for ANY character for ANY buttons. It's like 10x the amount of infrastructure; I could set up a "move x character around" in gamemaker in 5 minutes by hand, but it would take me hours to restructure that into a data-driven form.
But the artist and writer part of me just sees something that is functionally useless threatening to meaninglessly transform crafted work because a hoard of idiots with no taste thinks it's doing a competent job. I opened with the comment about writing to know what you think because I fully see now that the process of artistic creation isn't just iterative like a program -- it's transformative. The process of creating changes you. Generative AI is not able to change you on your behalf.
It sucks because, tbh, I feel very cut off from any kind of art or creative community right now. I've tried testing using chatgpt just because...I desperately want someone to talk to and share ideas who is able to give me something useful for me. Part of the cost of using generative AI is that it cuts you off from the people who could help you via the interactions you'd seek when searching for information. But sometimes...it feels like there's no choice. I feel like I've tried to connect with people for something as broadly connectable as my anime art learning and process for years and I was never able to make a connection with anyone I'd consider a creative friend on any social media. I don't want to have shallow conversations and I'm tired of trying across social media and discord and shit.
So yeah.
I'll be honest, a very viscerally frustrating thing to me about AI discourse is the us-vs-them purity camps perpetuated by algorithms. To me, it's the newest big wave of algorithm-created controversy for engagement farming. It is impossible to talk about generative AI without it being a battleground for threat detection and safety schooling. What I mean is that if you mention AI at all on social media, you're entering into a conversation where people are constantly judging you and evaluating you for moral safety. The algorithm-driven Bible has a long list of topics for which people must be sorted into camps where they are either deeply and fundamentally bad or good people, and bad people must be purged. That is the only modus. In a landscape where everyone is trying to prove they are safe, there is no possibility for collective action.
It doesn't help that I have some professional stake in accurately assessing the environmental impact of something like generative AI. Environmental psychologists study the psychology around what leads people to adopt environmental behaviors and how that affects environmental initiatives, for example, what kinds of environmental behaviors people perceive to be environmentally impactful versus which behaviors are actually impactful. Truelove & Gillis 2018 found that laypeople broadly misperceive the environmental impact of behaviors, misaligning with expert assessments. I work on research projects with environmental psychologists trying to understand why these misperceptions happen, this work was part of my dissertation, and I will be publishing papers on the research I've done. When I look into the specific numbers on generative-AI, I'm concerned but not doom and gloom like the algorithmic take is. For example, I don't think that generative-AI is definitely the most harmful and impactful environmental concern today. I find people's claims that generative-AI is THE ONE thing which will DEFINITELY destroy the environment to be overblown. I honestly wonder if those takes aren't being seeded by oil companies, and then perpetuated by people's need to make hardline stances in order to gain traction on social media. I don't think people realize the scale of carbon emissions generated by Amazon trying to get teenagers their newest Shein buy. So I find it really hard to buy into the kinds of takes I've seen online that are anti-AI, even if I'm sympathetic to the cause.
Despite all this, I'm optimistic that the social structure of society will evolve around AI, regardless what happens with AI itself. I think that people also don't realize just how much humans adapt to things. When generative AI was starting to get good enough to make plausibly nice look images years ago, people were freaking out about it being the end of trustable images. But I immediately thought about how photoshop has existed for a long time and people adjusted to perceiving it. I heard that young people are now becoming hyper-aware of what they personally know, and they go out of their way to demonstrate it, because they have to distinguish themselves from AI. It turns out that people have motives for truth, and for social standing, which can lead people to become MORE, not less, informed.
As with anything, the things I wrote above are my opinions right now.
Oh yeah, one last note: I've been using chatgpt heavily to prototype my game in html/js/css/php. But I reached a point where I'm no longer directly asking for the specific architecture I want, because I hit the limits of my own personal knowledge about how these things are structured. And as much as I use AI, I can't mentally get myself to just ask for features without being sure I can review the code and understand the setup. It's just too risky. I've starting shifting to a godot implementation because I ultimately want to release my game as a mobile game. As a result, I'm taking a step back and trying to redo things in godot. And I'm frankly more excited about learning about how to implement things in godot, since the prototype has sufficed to demonstrate to me that the essential concept is possible. We'll see how it goes.