AI Is A Tool, Not A Human Replacer

I've seen a lot of big gaming companies talking about AI being something they should be investing in. Not specifically for anything useful but for artistic purposes that take jobs away from people. They're trying to take the human out of development.

I'm a developer, even if not in a AAA company. I understand the benefit AI can give the industry. But if you've ever used it, you will recognize some of what I'm about to talk about. AI is flawed and repetitive and often completely unhelpful. But, because of the tech industries buzz words, companies have bought in to the idea it'll fix their money troubles by being able to cut a bunch of jobs so they can just have AI do all the artistic and writing stuff. So today I want to go over some of the problems with AI and why it is not the answer for any of the industry woes.

YOU ALREADY SAID THAT

I've used AI for one main thing in my own developer experience, debugging my code. It's been helpful for seeing some small things that I've simply not caught because I have been staring at the screen for so fucking long that it just doesn't exist to me. I can't express enough how helpful it has been to be able to just copy my code in, press enter, and it tells me what's likely wrong.

But it's not always right, sometimes it says something that is patently false or even takes my Java code and spits out an answer as if my code is in Python. That's not even the worst part, that I can just prompt it to correct and it'll fix, the worst part is the one thing you can't fix, and that's it's repetition.

For example, I've been in production for a game I'm developing right now and I'm figuring out some coding logic for some core parts of it in Unreal. Well, I've been using some AI for debugging and figuring out how I might want to structure some of the logic and functions (for non programmers, functions are essentially just something where you can write a bunch of code and then "Call" that code whenever you want, wherever you want. It's like you're able to direct the program to do something specific). During this time I've hit some walls where my code is doing almost what I want it to. Just one thing is off. And I have had significant issues figuring out why. This is pretty common in development, particularly early development as you get core systems to work. Anyway, I'll ask the AI to help and it'll give me an option and I'll try that and it won't work. So, then I get another way to do it and try that and it won't work. Then I get two more options and try them and they still don't fix my problem. Sometimes they even break it. And then I say "hey all these suggestions you provided haven't worked, what's another option?" And it'll go "Here's some other options than the ones I've already provided" and it'll list the exact same suggestions it has already suggested word for fucking word. It's fucking infuriating and I've gotten incredibly frustrated at it during my time working with it.

See, AI only knows as much as you give it. So if I know what I need to do it can help me figure out some logic, but if I need to fix something, some kind of bug, and don't know how to fix it, the damn thing is useless. For it to actually be helpful after trying all these options I have to figure out my own new approach and then ask it for some logic critiques or suggestions (which is how it should be in that developers should still be doing the main thinking for the games structure, but is not how it's being pitched by those in the AI industry). The AI isn't actually helpful for fixing big issues, it's helpful for small things. But when I need to rewrite and restructure some core logic to my game, and none of the options I've tried worked, the AI just starts going in a circle. It can't understand that there's other ways to think, that there's ways to take a step back and think bigger, think in a broader, program wide approach. To it, the only thing that exists is the information you've directly provided and you can't necessarily provide all of your code to the AI without it being overloaded and losing track (or having to pay for premium subscriptions which is a big No from me). Sometimes, it'll say things that directly contradict how you are already approaching things, even if you provided your code. It's crazy. It’ll also purge its memory every so often to keep talking because it’ll forget information that was already given earlier in a chat. It doesn’t have an actual memory, it’s just a cold machine model that can’t actually remember anything.

AI cannot accurately replicate the kind of thinking a human can do. All it can do is be a tool to help get some ideas of how to fix things, and a tool that has significant flaws at that. In order for this flawed tool to do the things the buzzwords say it will do, it needs to actually be able to do all of these things I've talked about. But it won't be able to and never will because it's not real AI, it's an LLM (Learning Language Model) that instead of thinking and then speaking and picking your next word or sentences carefully, it spits out and only thinks about the next word when it's actively choosing it. This is obvious when you provide a bunch of information to it and it acts like it's thinking. The information has details that almost contradict each other, or at least details that provide more context to your problem later on in your prompt. But instead of the AI responding to your whole prompt, it responds to the first part of your prompt that it understands. The moment it finds that spot, it stops reading and composes it's response because it is not intelligence and it never will be.

Note: AI has also helped me learn Unreal which has been so intimidating for so long for me, but because of AI I've been able to learn super quickly without annoying tutorials. So, some of the stuff I use it for that I've mentioned above are mainly just because I'm still learning my engine.

AI is not a trained artist

One of the things the Industry heads are hoping AI can do for the industry is replace artists. But...have they even seen AI art? It looks awful. Like there's some impressive stuff but it's still bad shit. It still can't replicate...anything that a normal person can. Let's examine three pieces of concept art, one made by a person and two made by AI at my direction to try and replicate the one made by a person to see the differences.

This was made by an artist at Bioware during the concept phase of the first Mass Effect. It's of the planet Noveria, an icy snow planet that has advanced tech and mountains as well as corporations that drive investment in the local settlement. You can see the care the artist put into not just the characters in the piece but the design of the windows of the building. You can see the history of the world and the cleanliness of it in a way that just screams corporations. You can see how the building directly fits into the mountain side. How it looks like it was built there by people who knew how to make buildings.

My prompt: A sci Fi scene with characters looking out over a mountainous region to the left of screen all in the foreground. The background is of on an icy planet that has a big advanced tech building built into the side of the mountain. The building is white and almost echos the image of a mountain built by intelligent beings.

I then used the same prompt but added this at the end. Make it a bit stylized as if it was drawn with a fine touch.

Do you see the problem here? Things don't look like how people would actually design stuff but in the Mass Effect image you can see a building that looks like people actually live there. You see a building that looks like it was built to be one with the mountains around it. But the AI images? Fuck dude. The first one reminds me of space shuttle stuck onto the side of a mountain that for some reason still has a bottom to it instead of it being fully built into the mountain. It looks like someone shoved it into the mountain instead of built it to fit into the side of a mountain. The second one is definitely better there but...why is there just a giant staircase in the middle of the mountains for a building that's supposed to be in the snowy mountains and have advanced tech? Why is there not just a door and a lift or elevator or something? This is advanced tech, not a fucking ancient ruin. In the AI images it has this soulless feeling to it of a machine creating something that might look good on first glance but after a second just makes no sense. Both images also have the exact same 3 people in the front who are generic as fuck and just wearing fucking boring clothes. But in Mass Effect they have fancy advanced Armor on. There's just so many issues in the AI images that an artist wouldn't have.

Now obviously this was my only two prompts, I didn't push it further to get more specific or to refine the image to get exactly what I wanted. But that's kinda the point. In the Mass Effect art you can just see the care put into it, how it's all well thought out and intriguing and gives you a feeling that this world has an existence that you just want to explore. The AI looks like something a higher up in a parent company would love but would obviously be trash if you just look at it for a couple seconds instead of a quick glance.

AI art can't make something from nothing. Can't make something new that doesn't have weird logical problems with it. The art above is just obviously AI art because of the problems. A person can see the problems with it, can see ways it needs to improve and ways it doesn't make sense. An AI can't.

The Stealing of it all

In my first section I mentioned how AI only knows as much as you give it. This goes back to how AI is trained off of things that already exist. When dealing with Unreal it takes info from things like the Unreal online documentation, but some of that isn't relevant or the AI gets confused and uses Unreal 4 stuff instead of Unreal 5. Again. This is all because AI is trained off of the internet.

When it takes in the art prompt from my second section, it uses all the images it's looked at online to inform how it should draw things, but then fucks it all up with its own misunderstandings of what makes the original art great. The AI imitates, it steals ideas from the rest of the internet instead of having an original idea of its own.

This has been talked about a lot by critics of AI, reasonably. And it does this in a sometimes unconsented fashion. So for instance, on Substack you can turn off a setting that blocks your writing from being used to train AI. You consent to it, you say "sure. Who cares." But this isn't every site, many don't tell you when your work is being used to train AI, there are times when you just find out that "oh hey. AI has stolen my work and my opinions. That's great". Having that feeling is awful.

All AI does in the creative (art and writing) field is just fucking steal. It's not a person and it never will be.

It does have uses

This is an off shoot of what I talked about the earlier where I've used AI for debugging code and making sure logic for something makes sense. I think it's important to remember that just because AI is a buzzword and that it's not able to replace all the things both what the creators of it are promising and what the game industry higher ups are desiring, that doesn't mean it doesn't have uses. For this example, let's look at a recently shown AI animated Aloy from the Horizon series.

The animation for her is... rough to say the least. You can see it's a bit too overdone and her face doesn't really react in the way you would expect a normal face to react when you're doing a bunch of talking. The possibility of AI being able to be used to improve facial animation is actually really interesting, however. It doesn't inherently take away jobs, it doesn't steal something from others, it can be refined to work in the best way possible without randomly going off in a weird direction. And it can make games feel more real.

The tech is early days but it could likely also make animating simply, quicker and cheaper. Right now games have such long development times which balloons a budget significantly. This often happens to try and adhere to massive graphically intensive games that set a standard in how much they look like they're real (also because of mismanagement causing games to end up in development hell for the better part of a decade or even a whole decade). But those things take a whole lot of time to draw and also a fuck lot of time to animate, even with facial capture. It's not an impossible use case for a game developer to use AI to help refine, not do entirely, a facial animation of theirs which halves the time needed to be able to do the work. Or even to help sync up the facial animation from facial capture with the audio recorded by the voice actor. If AI could be used to make games take even a smaller amount of time to make then that would be beneficial.

Open world games also have a shit ton of work put into random NPCs reacting in a realistic way to actions around them. For instance, in the Assassin's Creed games you have whole cities filled with people that you just run by. These people all exist in the world as objects that have predetermined walking tracks, actions, things they can say, and how they could react to the player. This takes a lot of work over a long time. With AI that time could be reduced, it could keep the developer from having to spend such significant time with small time code and having to hand code all of the walking tracks and basic actions like sitting or drinking. Freeing them up to be able to spend the vast majority of their time with higher complex logic and functions. Complexity that only a human can handle. This could only be done with internal AI systems as no currently publicly available ones are knowledgeable enough to be able to do literally any of this.

Mild tangent: At the same time, I think there's an argument for both using AI to help for some of these things, and for simply dropping others as things to even do in games at all. If playing Avowed has shown me anything it's that a game is not enhanced by the Bethesda pushed idea that every NPC needs their own daily routine where they come into work, do stuff for the day, then go home and sleep. This might be more "immersive" but all it really does is make the game more annoying. It is not fun to come to a store in the middle of the night in Skyrim only to find out/remember that "oh yeah they're closed" and then use the wait system to make it morning again. It's basically just looking for reasons to force the player to engage with that wait system instead of the game being there to let the player have fun. A small amount of games could still have this, but 99% of games don't need any of that shit. 99% of games don't need fish in rivers to make it more "realistic". Realism is a plague on the industry because it forces devs to do a fuck ton of work to make a good looking game as opposed to just a good game. But that's an argument for a different, future article. End tangent

Important note: AI Aloy was from a presentation from Sony about AI tech and did use an approximation of Ashly Birch's voice. That's not okay. Ashly is an amazing voice actor and AI being used to replace any voice actor is fucking insane. AI helping animators do their jobs faster? Fine. AI being used to get rid of a person's job that is fundamental to the personality behind a character? Fuck off.

The Conclusion

Games are supposed to be about having fun, AI has its uses there, but not in the way it's being pitched. Not in the way that it could literally replace entire departments of a game developer. The industry can use AI to a make the small things easier and less time consuming but there's also something they could do that would do the same thing. Drop it. Be like Avowed. Drop all the shit that just doesn't make a game "better" and instead weighs it down. Drop having the most realistic graphics when it'll take so much longer to make. Drop things that don't pertain to the game as a whole like having a bunch of wildlife that you can't even necessarily interact with. Drop trying to make open worlds with so many things you can do when you could just make a couple hub worlds or a linear game and have it still be an amazing experience.

The industry needs to grow. It needs to change. AI is only a tool that can be used by developers to make their jobs easier and to make their jobs faster. But the industry won't be driven to this change by an AI machine that revolutionizes development and removes human input, a Panacea if you will. It'll be driven by the people that exist in the industry right now and will exist in the industry tomorrow. By the developers who spend days or weeks animating something for it to be just right. By the developers who stare at their code for days not seeing the problem that exists in it. It'll be driven by solo indie devs who don't have other people to talk through their logic with. Developers make the games, not the computers they use.

Meow,

Cat

Thanks for reading my article about AI being a tool not a Panacea! Subscribe for free to receive new posts every Wednesday and support my work.