AI making 'games'
(deepmind.google)
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (23)
sorted by:
AI for game logic is a fools errand. There's no way it could ever be reliable.
Graphics inferencing will totally be a thing though, possibly very soon. Render a basic scene, skip all the shader stuff and go straight to generating the graphics style you want. AI models have also proven to be very good at approximating light transport and particle simulations, so water and smoke and diffraction and transparency could all just happen for the the same linear time cost.
AI with interaction like this would be much more suited to VFX and creative applications where you don't want strict rules. Being able to adjust a scene with real time feedback would be very useful.
I’ve always thought generative AI could be best used in something like a sports game. Take college football the video game. I play NCAA 11, and each week of the season it would create headline blurbs out of the simmed results, using player stats, game stats and team/conference records. Using those stats, I think it would be cool for AI to generate a nice little article about each of the big games and the team you’re coaching in Dynasty mode. It would add a lot of immersion and also make it easier to keep up with the wider season as it is quite boring looking at stat spreadsheets after a while.
Scenes are good start. That's where I am starting at for my personal learning.