SvennoJ said:
All you're describing is not AI learning, you're asking for the same polished fights yet changed up for higher difficulty, Basically that's what DS2 is, and then DS3. Same stuff basically, different attack patterns and weapons (and a new lick of paint of course) It's all still set patterns, the enemy you're fighting isn't learning. You're just learning its patterns and tells. |
Whooooaaa there, I'm addressing Souls in particular and how previous posts are mentioning the complexity and cost, or in particular, the poster that's mentioning how it can be done without using AI learning. I, myself in the video, am clearly talking about AI adaptability in which it DOES have to learn to some extent, but here on the forum I'm also addressing the current state of games like Souls and Monster Hunter and how simple changes can be highly effective since both rely heavily on cycles (read: grinding). Once again, not looking for "human" AI, but am also discussing how simply tiered skill difficulty AI routines can be applied to currently considering "difficult" games, ergo Souls. Once again, amiibo CPU fighters are a surprisingly good example of how it can be applied without AI machine learning until said AI machine learning becomes more accessible/widespread/widely adopted.
You are right that "game guides" on how to beat bosses would become meaningless, but that's a GOOD thing. Granted, on easier modes, the vast majority of players don't need guides. For people looking for a challenge, the idea that they would then also look up how to beat the boss seems counterintuitive: the skilled players are NOT looking at game guides for how to beat bosses lol... I think they would welcome the change-up!
Rubberbanding, however, is not adaptability. That is, in and of itself, simply metadata adjustments (seriously, making a car go faster, or slower, or selecting its place in the pack, that's all just conditional changes similar to the previous poster mentioning ammo/resource adjustments based on player performance). This is why I kind of chuckle at Xbox still being the overall most hated of the three as they have done backend things that most "gamers" would completely fail to appreciate, and one of those is Drivatars. Sure, there are still programmed confines to Drivatars, but I can tell you that it's implemented better than people imagine it is, and it uses simple AI learning processes where a human player is present to learn data from and apply to their digital offline AI racer. Otherwise, they kind of randomize the behaviors where no data is present. Is there still a target difficulty that the AI drivers try to achieve based on the selected difficulty by the player like what you described? Yes, I've tested it. But it's still a step forward in the right direction when cars will brake early or brake check me, or side swipe me or pit me at all what is seemingly completely random and is what keeps even a simple offline race potentially "dynamically difficult". When the AI behaves out of expectation, it's what challenges the human lmfao 
