I was wondering if anyone was willing to chat with a non-coder a bit about the state of the art in AI. What kinds of things can AI do nowadays? What are some of the most interesting games these days in terms of AI, and how so? Articles or site links welcome.

What kind of AI are you talking about here? The term is notoriously overloaded and means a lot of different things to different people. What AI programmers consider "AI" is not at all what Joe Sixpack who just saw Bicentennial Man considers "AI".

"Interesting" is also a point of view. There are some interesting developments in games like Go, where AI is still very primitive compared to even novice players; and there are other interesting things going on where games like Checkers are completely solved for all possible games; and still more interesting things in the middle ground where people work on stuff like sports AIs and whatever else.


In short: can you narrow down the question a bit? :-)

Sure. I was thinking in terms of what we can do with bots in shooters and the like.

Not game AI per se but this was kinda interesting:

AI robot: how machine intelligence is evolving

AiGameDev is probably your first stop, then :-)

There are some interesting developments in games like Go, where AI is still very primitive compared to even novice players; and there are other interesting things going on where games like Checkers are completely solved for all possible games [...]

Sorry to nitpick, but go programs are at the level of very strong amateurs (see http://dcook.org/gobet/), and checkers has not been solved for all possible games, but has only been weakly solved (which means we know the value of the root under perfect play (draw)).

In the commercial games these days the most advanced things seem to be Behaviour Trees and planing systems (GOAP or some variation).
When I say advanced I don't really mean the newest discoveries. For example planning systems go as back as 1971(STRIPS), and even earlier.
There is a lot of difference in the purpose between Game AI and Academic AI. If you are interested in the above subjects, as mentioned before, aigamedev is a really good place to start reading.

For bots in shooters I remember liking what they did with killzone's AI:
http://www.cgf-ai.co...killzone_ai.pdf

In call of duty and the likes most of it is scripted and not really interesting or special.

The state of the art is IBMs Watson. Nothing else comes close to it's achievements.

Second to that is Microsoft Kinect. It detects people and estimates poses in real time, even against cluttered backgrounds. Very artificially intelligent.

Most of the best AI is in things you don't notice... Path estimation to reduce the appearance of lag, algorithms to ensure realistic poses and animations, navigation algorithms, etc...

Most video game NPCs use scripted behavior or simple hard coded behaviors. You will find the most interesting AI where a machine needs to interface with reality, but at that point it tends to become very transparent.

Some UAVs have pretty slick AI for target tracking and line-of-sight planning. Googles self driving car is another example.

The state of the art is IBMs Watson. Nothing else comes close to it's achievements.

Second to that is Microsoft Kinect. It detects people and estimates poses in real time, even against cluttered backgrounds. Very artificially intelligent.

Most of the best AI is in things you don't notice... Path estimation to reduce the appearance of lag, algorithms to ensure realistic poses and animations, navigation algorithms, etc...

Most video game NPCs use scripted behavior or simple hard coded behaviors. You will find the most interesting AI where a machine needs to interface with reality, but at that point it tends to become very transparent.

Some UAVs have pretty slick AI for target tracking and line-of-sight planning. Googles self driving car is another example.

I assume this is what ApochPiQ was referring to when he described the term AI as overloaded. IMHO, most of the examples you give are not intelligence at all, rather, they are heuristic or algorithmic solutions to problems. Personally, I would prefer to reserve the term intelligence for solvers, not solutions. I think this is really where the divide between academic and industrial/practical AI is most noticeable. Most research and development tends to focus on solvers, or at least generalized solutions.

I can't disagree with you about the state of game AI. There are some exceptions, but most games seem to prioritize predictable fun over emergent behavior.

That being said, the things that most games do with AI and the things that games could do with AI are very different.

Why does the OP want to know?