The future of game development... has no game engine? Ai generated Doom

Another future resident for the Google graveyard, I see.

When you see how Stadia ended ...

I've read a news article about this around the time it came out. A similar thing was created for the original Mario too.

The problem is, like with all other generative AI, is the AI is purely making a best guess and has no idea what's actually going on. It saw hundreds of hours of Doom footage and it guesses what should happen when the player turns, shoots an enemy, and moves to a new area.

It has no concept of memory, it doesn't actually keep track of player and enemy position, their health, ammo, etc. It's all guesswork that happens to work okay after XXX hours of training. When the player shoots, it guesses the ammo counter has gone down by 1. But like all gen AI, it is susceptible to bullshitting hallucinating, where it will either guess incorrectly or make an uncertain (wrong) guess intentionally because it's been trained to always give an answer, not an accurate answer.

I'm sure it will improve more, it will look better without so many artifacts, maybe fewer hallucinations to better pretend it has a memory, but it will still be an extremely inefficient generative AI which has to be trained on a game before it can create one. It can't create completely novel concepts, it can only use what it saw and modify it. AI development could be the future, but it's going to have to take on a completely different form than generative AI. I think supplementation is a far more successful application of gen AI, like creating textures and basic models, writing tedious code.

In my opinion, this is another example of tech bros hamfisting AI into anything they can think of to show how it's the "future". It doesn't matter that it's incomprehensibly inefficient, slow, inaccurate, and inferior to standard practices, all it needs to do is impress other tech bros and people with limited knowledge of the subject, so that investors keep throwing $$ at AI.

Generative AI presents solutions to problems that don't need solutions. Besides, this is another one of those things that aims to replace human creativity rather than replacing human tedium so that the humans can do the creative stuff.

Silicon valley is in love with the sheer power with generative AI for everything- and a lot of normies buying on that hype- but that power is very flawed and very limited, and they keep just throwing it at things blindly to see how much comes up by itself... wich is terrible. I mean it is surprising it can guess and reach that point, but that isnt really usable. Most of everything AI today are like tech demos, i mean alpha tech demos but because of investor money theyre treating that as final...

Anyway, THIS wont fly. Not even in the future, not even if it ran fast.

BUT theres a LOT of room for AI with games done proper.
Instead of expecting generative AI to guess everything and on the freaking fly- ridiculous- we can(and likely will) make specialized agents... very specialized ones. One is making world models, another enemies, another animations... But eventually specialized agents, even being just generative will be able to make things like mechanics or even bosses- not all at once, but different agents and combining things.
Humans will probably have a role filtering- so AI make a bunch of junk, a bunch of usable stuff but not quite fun or not quite balanced... but filter throught it...

It can become quite like procedural tech weve seen and had for decades, just expanded. The kinds of procedural generation that made dungeons in text based rogue-likes can have AI agent similars making the dungeons, enemies, loot, new spell effects, quests and their questlines and the npc dialogue...

Theres potential to go even beyond like specializing agents with engines, with codes and modules... to the point that telling the bunch of agents to make a dozen guns each with 'creative energy projectiles' will do just that in the engine- a model for each, firing animation, vfx, the code, etc.
Thats far different then rendering in real time a video that pretends to be a game that reacts to your button presses