Azdul
Magister
I'd rather play Daggerfall than anything created by AI trained to please the players. Implementing the feedback from focus groups in automated way guarantees that the end result will be crap. People don't know what they want, so AI will find local maximum, rather then explore whole spectrum of possibilities.Yes, procedural content cannot replace a game designer, at least not for now, the problem is that many games use procedural content as a replacement for the level designer when it shouldnt be, Daggerfall included. There are a few experiments of using machine learning with data collected from thousands of players to create a game designer Ai that follow the parameters from the data of what the players find fun in general and this Ai would generate the levels taking that in consideration. You could actually select the best Ais with different personalities and have them to play the level and rebuild the level thousands of times until reaching a decent result.
Just imagine the same idea used to generate dialogues for NPCs - you'll find that the players prefer fart and butt jokes to sequences built using random words - but you'll never find anything resembling actual good script or conversation.
Advanced AI can improve game development - but it should be trained on millions of examples from the real world - and not use subjective definition of "fun" as a feedback loop. Besides, no human has a patience to provide meaningful feedback on 10000 of similar, but slightly different dungeons.
IMO the most important use case for AI is providing voice acting and facial animations based on limited number of samples. We have linear plots and limited dialogue options mainly because recording voices and creating corresponding facial animations requires hours and hours of manual work.