Earlier quoted context omitted.
I would love to see someone use modern machine learning techniques to make a kick ass Civ AI.
Genuine question - would that be amenable to fast AI? It's less of a problem on modern PCs running Civ4, but on contemporary systems late-game large maps with many AI/units could really drag during turn-processing.
The customization available in IV makes it basically infinitely replayable, but the AI makes the trajectory of each game too predictable if you understand the mechanics well enough.
Lots of old strategy games have been revived by introducing new factions that change the game’s meta; imagine if this process was automated by training the AI on recorded games from the entire playerbase, or on games recorded locally to adapt to the user’s unique style of play.