Adaptive AI (Artifical Intelligence)

Question

I was thinking to make my arcade game match the players strength in the hard level mode. Like checking the score and if the human player's score is above computer's for certain number of units then the computer player would automatically increase the speed of its reactions, just like a human companion who gets more determined to fight back.

Does this sound a good idea to you? Are there other ways to approach this, so to make a game, like air hockey for a crude example, permanently challenging for anybody?

EDIT: I agree with the answers below. I believe that most people like to win about 70% time over the computer in order to be motivated. Hence an adaptive intelligence must not be become too superior to the player and depending on human player's concentration and some random elements in the game (good 'old' quantum RND), the outcome should be always uncertain.

A nice thing about this is that such a system could handle week and strong players and would adjust to the human's learing curve.

Answer 1

I believe it is a good idea, but to use wisely. Adaptive difficulty can be frustrating to players who wish to beat a game with a fixed set of rules and are suddenly faced with a different difficulty without asking for it. It will depend what your game exactly is about, though.

Answer 2

I think it's similar to the so called rubber-banding in race games which is fairly common, if done correctly it can make games more fun to play.

However when done incorrectly it can lead to the player feeling like he (or she) isn't getting better at the game or to the player getting frustrated that they aren't making any progress.

Answer 3

To be honest, I think most players like AI that can out smart them; the difficulty curve should be included in part of the level design, so that the player can still manage the challenge, but it is difficult. You want to keep the player constantly thinking, unsure of what the AI will do. Predictability and AI 'quirks' kill immersion and game play, so they should be the highest priorities.

Second to that, if they become too smart, lower their capabilities, for example in an FPS, you may hard code a bullet dispersion for AI to decrease their 'aim bot' abilities.

Answer 4

I believe not only is this a good idea but should be mandatory for most games.
Make your system not as simple as if the score is higher than some arbitrary number than send enemies with more health, that is the wrong kind of difficulty increase. Also, don't limit it to the hard difficulty, start easy and let the system adjust the difficulty according to what the player does.

For your example of air hockey I would divide the elements of gameplay into several parts through a brainstrom. Say, you start with the full game, that divides into offence and defence, and defence divides into keeping your goal safe and being quick enough to react. keep the player's score for each of those and then adjust it (according to a function, not some random number, I'm pretty sure there's some research for that) for each of the fields of the AI. Infinite adaptive mario is a great example of what you're trying to do, you should look at that.

Answer 5

I dislike that idea, because the change in difficulty is not obvious to the player, making him feel he has not improved, interfering with any achievement mechanism such as a high score table.

Instead, I prefer if the player gets to choose the difficulty level, but has an incentive to go as high with the difficulty level as possible, for instance because in-game rewards (score, powerups, whatever you have) increase with difficulty.

This doesn't have to break immersion or game flow; in an arcade top-scrolling shooter, the player could pick up a score multiplier (honey), thereby enraging the bee hive :D