Originally Posted by Tuco
Originally Posted by mrfuji3
Potentially this is a thing that can't be simply fixed by options in settings though, since Larian makes use of a learning AI to govern enemy behavior.

If the default game-mode has SuperShove, the AI will (and clearly has) learn strategies that depend on it. So if Shove is changed to an action, reduced in distance, etc, then that might break the AI behavior. E.g., the AI could default to using Action-Shove to no effect.

Since you mention this tangent, I have to say that I've been made vaguely curious by this. Let's say right now Larian changes Shove to an action without any other additional change. OR a modder does (there are already few mod that force this change now)... What happens with the AI? It maintains the same exact "learned behavior" but wastes an attack to shove just as frequently? Or does "weight" the modified shove as an action, decides it's less appealing that it used to be (outside of specific circumstances) and makes a different decision?

Basically I guess you could summarize the question as "To what degree is the AI based on calculating on the spot and to what gets the best result and to what degree it just acquires a learned behavior running thousands of simulations over time?"

Ideally, Larian trains the AI for each different rule set they will include in the game. Then we will have smart AI for BA shove and action shove.

In case of modding, then most likely AI will act as if it spend its shove action already. It is going to be very difficult for AI to assess the situation where shove is an action and competes with other actions (if Larian does not prepare the AI for this scenario and modders change shove to action). Probably AI won't use shove at all or use it as if it is BA and waste its action doing so.