its a very new approach in artificial intelegence in games.
Not really, dynamic difficulty adjustment has been around since at least the 80s (see for example Zanac) and quite possibly earlier. In an industry that moves as quickly as video gaming the idea of dynamically adjusting difficulty is positively ancient.
The good thing about this is that it means there's actually been a lot written about it, about games that use it, and about different approaches over the years.
Have you read up on Valve's "AI Director"?
It forms the heart of the Left 4 Dead franchise, and makes an appearance in several other games (i.e. the top-down survival horror title, Alien Swarm). It is effectively a dynamical AI presence which reacts to the player's progress and status, doling out ammo and health in some situations, and punishing waves of enemies in others.
Even if those games aren't exactly your cup of tea, they are worth checking out for the AI Director alone...
+1
L4D handles it pretty cool, you can read it up over here (look around 2009). It is one of the more advanced adaptive systems which are nicely integrated into the existing game.
The level that "got easier after the first couple times you died on it" is a very specific kind of dynamic difficulty adjustment, and in most situations not a good one:
A situation in which specific levels can unpredictably become roadblocks, for example because many general skills (e.g. timing shots and judging distances for jumping) and game-specific skills (e.g. exploiting koopa shells and finding secret passages in Super Mario Bros.) might or might not be aligned between player and level, can be approached with suggestions ("Don't forget to use your flamethrower. It can be activated with button 2"), offers to run special tutorial levels ("Improve your aim at the firing range tutorial"), and difficulty settings, possibly allowing them to be set for each level.
If the difficulty settings affect something simple and obvious, like increasing a time limit or making the character less vulnerable, instead of "corrupting" level features like enemy placement or obstacles, the player will be more confident at switching to harder difficulty when ready..
A situation in which specific levels can unpredictably become roadblocks, for example because many general skills (e.g. timing shots and judging distances for jumping) and game-specific skills (e.g. exploiting koopa shells and finding secret passages in Super Mario Bros.) might or might not be aligned between player and level, can be approached with suggestions ("Don't forget to use your flamethrower. It can be activated with button 2"), offers to run special tutorial levels ("Improve your aim at the firing range tutorial"), and difficulty settings, possibly allowing them to be set for each level.
Or there can be a “secret” passage that let's the player bypass the roadblock. Players will feel smart or happy for discovering this easier route, while hardcore players will love to take on the challenge. It can basically be a Skippable Boss, or an Easy Level Trick.
Since I'm already linking to TV Tropes, check out the Dynamic Difficulty page (including the subtropes) if you haven't yet!
Since I'm already linking to TV Tropes, check out the Dynamic Difficulty page (including the subtropes) if you haven't yet!
Haha. They point out a lot of ways dynamic difficulty can fail. They don't seem to like the idea that much.
Haha. They point out a lot of ways dynamic difficulty can fail. They don't seem to like the idea that much.
Or maybe there are just many ways dynamic difficulty can fail. Tropes—including dynamic difficulty—are tools. They're neither good nor bad.
I believe the most important questions to ask yourself still are (a) if dynamic difficulty truly supports your vision for your game better than other means (some of which were mentioned in this thread), and (b) how dynamic difficulty can support your vision.
Dynamic difficulty only exists in context, so when talking about it out of context no one can really say if it's good or bad. Resident Evil 4 uses dynamic difficulty, and the game is universally praised, so apparently it worked very well for it. On the other hand, there are many games in which it would hurt the game (e.g., Ninja Gaiden Black).
So... to be more precise; look objectively at ALL the difficulty tools you have available—including, but not exclusively, dynamic difficulty—analyze the effect each tool has on games (TV Tropes honestly is a great help for that with many games listed for each tool/trope), and then select the tools that best support your vision for your game.
I like a system where areas have relative difficulty indicators (not necessarily linear) and that player skill factors in, but does not direct the curve.
For example, an experience such as this:
1 - 1 - 1 - 2 - 1 - 2 - 1 - 3
where ...
...1s are "easy" areas
...2s are "medium" areas
...3s are "hard" areas
Would play differently based on player skill.
If the player did fine in one area, the next (or the current area) could shift gears up to +1/-1 depending.
So the player experience might end up this:
1 - 2 - 2 - 2 - 2 - 2 - 1 - 3
The player fared well in the first area, making the next one +1 harder than its natural value. Then, the system deems that +1 was an ok fit for the player and has the 3rd area with a +1 to difficulty. The player did "ok", so the next area (a 2) remains at 2 because the player has not demonstrated that beating that area was too easy, etc.
You could the player level in as well (assuming RPG power curve) where:
Final Area Difficulty = Natural Area Difficulty +/- Difficulty modifier + (Level / 10)
(All placeholder values). 10 might be too strong a divider, so the unit system would have to be balanced efficiently (I'm leaning towards a + 1 difficulty ever 3-4 levels instinctively, so the natural area difficulty and modifier could be larger numbers).
A situation in which specific levels can unpredictably become roadblocks, for example because many general skills (e.g. timing shots and judging distances for jumping) and game-specific skills (e.g. exploiting koopa shells and finding secret passages in Super Mario Bros.) might or might not be aligned between player and level, can be approached with suggestions ("Don't forget to use your flamethrower. It can be activated with button 2"), offers to run special tutorial levels ("Improve your aim at the firing range tutorial"), and difficulty settings, possibly allowing them to be set for each level.
Or there can be a “secret” passage that let's the player bypass the roadblock. Players will feel smart or happy for discovering this easier route, while hardcore players will love to take on the challenge. It can basically be a Skippable Boss, or an Easy Level Trick.
More simply, hard levels can be not roadblocks by not being on the critical path.
For example, Gemcraft: Chasing Shadows, a tower defense game, unlocks levels after beating other levels, lets one play any unlocked level at any difficulty setting, limits possible grinding, and "ends" after beating a certain level. Whenever the player is stuck there is a vast choice of other, quite different new levels and old levels that allow gaining experience point up by replaying on higher difficulty; eventually the player learns some tricks and/or levels up to be strong enough, and a level that was set aside is beaten.
Many games, mostly puzzle or action ones, adopt the even simpler scheme of freely choosing any level, or any small sequence of levels.
In New Super Mario Bros the game developers (Nintendo) decided to introduce a feature where if you die three times in a row, a Luigi block appears (they call it a "Super Guide Block"). If you break the Luigi block, you can watch Luigi show you how to overcome the challenge. This is a good idea.
To make sure struggling players actually notice the Luigi block, Nintendo makes it shake and make a loud dinging noise. In theory, this might be a good idea.
To make sure expert players have challenges, each level has bonus items that are difficult to collect, some of them requiring great feats of skill. This is a good idea.
To make sure to annoy those experienced players, each time they fail at the challenge (some of which took me +50 tries to accomplish), Nintendo thinks they failed at the entire level because they are novices, pops up the the generic Luigi block for showing how to get through the level normally, and rings the loud dinging noise.
This means, every time you fail at the intentionally higher-level bonus challenges, the game mocks you with a loud dinging noise and basically says, "Do you need help getting through the level the regular non-challenging way without the bonus items you're trying to get?". They might as well pop Microsoft Word's Clippy up and ask, "It looks like you're sucking at this game, would you like some help on an unrelated task?"
These even occur if you've already beaten the level on that save file. 
So here's my recent take on Dynamic Difficulty:
https://itunes.apple.com/us/app/arcane-castle/id971218980
http://www.gamedev.net/topic/666951-arcane-castle/
It's an endless runner game, with specific spells that you use to get past specific enemies. And there's a story mode that's not endless. It's common for endless runner games to get faster and more difficult as you go, but for some, when you lose it's difficulty is set back to the start and could take a while for an experienced player to get to the intense part. In Arcane Castle endless mode, each time you play the difficulty is adjusted by the previous game. The score increases exponentially with the difficulty, and it accounts for the points you would have gotten had you started at a lower difficulty and gotten there with no damage.
I decided to focus the game on this challenging feeling that makes you unable to speak in full sentences until the level is over. The game gets more difficult, the more life you have left over at the end of a level, and easier depending on the amount of progress you made in the level before losing. The idea is if you just barely won a level or just barely lost, the difficulty barely changes, leaving you at that difficulty level of hanging on by a thread.
However, maybe some people would prefer a more casual experience? I could add an option that adjusts how difficult the levels get. It would still be dynamic, but adjusted to get difficult slower, and easy faster. Maybe this would be enough of a feeling of control over the difficulty, even though it's primarily dynamic?
