The AI gaming revolution is lamer than I thought
So, Microsoft's AI-generated Quake II demo has been making the rounds since yesterday, and while it's somewhat fun to mess around in, it's also a perfect case study in how the very executives within the game development space still don't understand games, what they are, how they work, or why people care about them.
How did we get here?
Alright, it's story time with Uncle Mors. A few months ago, there was a playable AI generated version of Minecraft also making the rounds. It was a really weird and somewhat fascinating project, because it was essentially AI hallucinating frames based on your input. There was no gameplay logic or anything, just AI trying to predict what comes next based on millions of frames of training data. As a result, the whole thing felt like a fever dream, where the terrain shifted constantly, and you kept teleporting to different locations the moment you looked away.
It got a lot of flack at the time which kind of annoyed me, because to me it was clear that this tech wasn't actually intended for gaming. It was intended for real time video generation, like say you want to generate B-roll footage of a drone and you wanna control it live or something. Even for that kind of stuff though, the tech had loooooong ways to go. And for gaming? It was practically unusable, and wouldn't ever become usable. Again, it's just the AI predicting what image on the screen comes next. It's not advanced enough to simulate the hundreds of interlocking systems a game like Minecraft consists of, especially with perfect accuracy.
Anyway, over a month ago Xbox also announced their own generative AI gaming model, Muse (nothing to do with the band people can't stop comparing to Radiohead). Looking at the announcement post, this is clearly Microsoft's own take on the Minecraft AI demo, using very similar tech with very similar results. The difference here is that this one is very specifically made for gaming.
Alongside the announcement post, they also released an extremely baffling video that I feel like not enough people talked about. In it we hear Phil Spencer say some incredible stuff like:
“You know, one of the things we care a lot about at Xbox is game preservation.”
Huh? Preservation? What's that got to do with anything?
“But you could imagine a world where, from gameplay data and video, that a model could learn old games and really make them portable to any platform where these models could run.”
Phil what in the FUCK are you talking about? This is just AI video generation. How is it supposed to replicate games perfectly? That makes literally no sense, and should be obvious to ANYONE who has any idea about how generative AI works.
Does he genuinely think we can train these models to just emulate games? I had initially thought he was going to talk about AI post processing to remaster games, which is a dogshit idea too, but it's at least possible. But no, he's proposing an emulator of some kind. Genuinely one of the dumbest things I've heard an executive say in a long time.
“These models and their ability to learn completely how a game plays without the necessity of the original engine running on the original hardware I think opens up a ton of opportunity.”
Like, this sounds REALLY cool if you have no idea how any of this shit works. But if you do, it's just a load of nonsense.
Now, back to present
And now, a month later, we can finally try Muse for ourselves, not with an open world procedurally generated game, but with Quake II. And the result? I mean… It looks a bit more coherent from a first glance, but when you start messing with it you realize that yeah, it really is exactly as broken as you'd expect.
This thing looks like Quake II, but it doesn't play like Quake II. It doesn't even play like a game, because it's not a game. It's a sequence of stitched together hallucinations with a weak grasp of continuity. It doesn't simulate any of the foundational elements that make Quake II an actual game. It mimics the appearance of gameplay, without any of the structure underneath.
Move forward in this AI demo, and the hallway will warp around you. Turn around, and suddenly you're somewhere else. Darkness isn't an absence of light, it's a screen wipe into another disconnected space. Oh what's that? You're shot by an enemy? It's fine, your health meter doesn't even function properly, it just wobbles around. You're not playing a game, you're watching an unstable, fucked up deepfake of Quake II try and fail to keep a coherent thought together.
The fundamental issue here is that this model doesn't really understand what a “game” is. It doesn't simulate anything. It doesn't contain systems. It stitches together a plausible visual approximation of gameplay based on pattern recognition. That's it. It's not reasoning about how health works, how enemies behave, or how space connects. It's recognizing, for example, that when a hallway ends, sometimes an enemy appears, so it spits out an enemy. But it doesn't know exactly what to do with that enemy because the Quake II enemy AI is way too complicated for this AI to replicate.
But the biggest tragedy here isn't just that this tech doesn't work. It's that Microsoft, with all its money and engineering talent, thought this demo was worth releasing in the first place. They looked at this mess and said, "Yup, this represents our vision!".
That's horrifying!!! That reveals a catastrophic level of detachment from what makes games work, from what makes them worth playing. Like, games are fun because you can actually play them! They are more than just some images moving around. They function!!! CONSISTENTLY!!!
The sad truth is that tech executives, and much of the broader tech media in general still treat games as nothing more than just novelty items. Shiny toys for teenagers to play with, and YouTubers to make money off of. Like, they don't understand video games because they don't even play them in the first place. And they most certainly don't appreciate the creative and technical craft that goes into making one. This is why they see this as some kind of revolutionary leap forward for game development. They genuinely believe that this is a tool developers will one day use to prototype games faster. Yes, they actually said that. They genuinely think that this is a step toward a game development revolution, a tool for universal emulation, or god forbid, preservation.
No. It's not. This isn't a game dev tool. It's a gimmick. A toy that got mistaken for a tool. A flashy prototype built on misaligned incentives, aimed at out of touch shareholders rather than players or actual game developers. A marketing ploy dressed up as vision. A stunning lack of respect for both game development and preservation. A sign of an industry careening toward a future it doesn't understand, powered by executives who still think game development is mostly about making pretty pictures that move when you press a button.
Okay… Then what's next?
Nothing. This tech is a dead end. As I said earlier, there is some potential for it to be used for interactive video generation (which I'm also not a huge fan of), and it is interesting to mess around in, but it's NOT the future of video games like Phil Spencer wants us (or rather wants shareholders) to think.
Look, I know only 5 people in total are going to read this post, but...in the off chance that you're an executive reading this, maybe you should stop watching keynote sizzle reels and start talking to the people who actually make games. If we keep pretending that generative AI can “just learn” how games work from video and player input, we're going to waste years and billions of dollars chasing ghosts.
And if you're in the press, stop treating this like it's a breakthrough. It's not. It's a cautionary tale. Stop reporting on it so directly. Yes, I'm talking about you, Geoff “Mr. Video Games” Keighley. Shut the fuck up.
Now if you excuse me I'm gonna go back to playing Sonic Generations for the 27th time.
Keep Reading
Games are made by individuals
Why is it that you don't know that many directors for games?
