Is a truly adaptive, learning AI actually useful in game design, or is it a waste of resources?
It depends on the goal design of the game. If it's power fantasy, I believe dumb and slightly smart NPCs would be better to convey that.
If it's challenge, smarter NPCs could answer that.
The "Goldfish" vs. Narrative Balance:
I think the answer depends on how AI is learning or adapting if you will. If it's adapting while adhering to defined personality, it should be okay.
Exactly, it really does depend on the game's core design goal. A pure power fantasy needs predictable fodder, whereas a tactical challenge needs something smarter.
I wanted to clarify the bigger picture for this project, though, because it's actually designed to go way beyond just enemy combat or boss fights. The goal is to build a completely all-purpose, multi-genre AI tool.
It's meant to adapt entirely to whatever personality and constraints the developer sets up. For a power fantasy game, a developer could easily code an NPC to be a total coward who panics and runs away from the player, making the player feel even more powerful.
Because the backend handles general-purpose actions (moving, interacting, talking), it can be applied to almost anything. It could be a friendly companion that keeps you company and talks to you organically without relying on a rigid, hardcoded dialogue tree. It could be an unpredictable driver in a racing game, or even a non-human entity like a pet dog or cat that learns your habits.
Ultimately, it's a tool meant to give devs a break from scripting endless "if/then" scenarios, letting them just define the character's core traits and letting the AI adapt to the game world naturally!
