My son and I have been thinking about what video games could be with generative AI.

The north star is OASIS from Ready Player One, that is some kind of game framework in which any kind of game could be constructed easily. VR content isn't going to become mainstream until authoring costs drop a lot.

Horizon Worlds wants to be OASIS but falls far short. Not because it's a bad idea but because it's a difficult problem with difficult trade-offs. A game like Grand Theft Auto: San Andreas or Wrath of Asgard 2 makes large 3-d worlds look easy but when I develop content based on my photography I realize quickly that it is easy to fill the memory of my Quest 3 and when I think about it, I realize that GTA doesn't have any textures as detailed as my photos. There is a trade off between ease of development and runtime performance, and something like OASIS means a revolution in the first while maintaining excellent performance in the second.

So far I've been talking about graphics but story is just as interesting.

In the game Danganronpa for instance, you need to solve a string of murder mysteries to win the game. It's kinda disappointing that "who did it" is fixed from playthrough to playthrough. Danganronpa has relatively simple 2-d graphics and is a different problem than 3-d games, but one can see how a game like that could be a lot bigger than it actually is.

Similarly there is the interactive fiction world of

https://ganelson.github.io/inform-website/

Where you can "write" an adventure game in a scripting language that looks a lot like English. It's a neat trick but it's lacking the real interactivity that is lacking from all video games today which is dialogue. Cast members at places like Disneyworld and Colonial Williamburg can have a real conversation with you and so should video game characters.