A blend of a idle gladiator auto-battle + real machine learning (AI) game.
The game tracks ~100 (or whatever) gladiators that have rpg like attributes - strength, durabilty, speed etc and the gladiators auto-battle on a turn-based hex map. 1 vs 1. As a slave master, you get to buy a gladiator, or more, then watch it autobattle. It might win. It might lose. It might die. You don't care if it dies. You'll buy another one.
The thing is each gladiator is a genuine, unique, learning AI. It starts off knowning nothing. It is dumb. It is a rookie. It has to fight another rookie and it slowly learns. First it stumbles around the place. Then it knows how to find a target. Then it knows how to swing and win. Or - it will if learns fast. It will die if it learns slow.
100 gladiators = 100 little neural nets all learning at it's own pace. Fast learners strive and become hero gladiators. Those are the ones you want to own but can you scout those early on when they are rookes? Can you spot the neural networks that show potential? Can you spot, and buy, the cheap rookies that are destined to become expensive hero's?
This is a "roster management" game where you scout and buy gladiators that show potential and then you watch them auto-battle to the top. Each neural network will grow in different ways and different speeds giving each gladiator a truely different behavior. There will be many many lemons and there will be the occasional hero.
Top down, 2D, turn based, hex map. Watch your gladiator take actions each turn and cheer him on to victory so you get the spoils of tournament victories.
They all die in the end - but perhaps not yours - not today.
