For those interested in next-generation AI in games...

[Azarkon]

For those hardcore gamers who think that AI is the future of games, you might want to keep track of the AIIDE conference:

 

http://www.aiide.org/

 

Proceedings available here: http://www.aaai.org/Press/Proceedings/proceedings.php

 

For example, 2006 proceedings: http://www.aaai.org/Library/AIIDE/aiide06contents.php (if you don't want to buy the books, you can simply input the paper titles on google and most of the times you'll find a pdf link)

 

In terms of prestige, for a game AI conference to be published by AAAI (which, along with ICML, NIPS, and a few other select conferences can be considered among the top academic conferences for AI in the world) and hosted by universities like Stanford speaks volumes about the growing importance of digital entertainment. Serious researchers at the forefront of AI technology are publishing at this conference, so this isn't your typical gathering of armchair hobbyists. It's the real thing.

 

Be warned, though, the papers tend to be quite technical

Edited January 19, 2008 by Azarkon

[Azarkon]

Terms to look for for RPG enthusiasts:

 

Interactive narrative

 

Drama management

 

Believable characters

 

Natural language

 

RPG

[alanschu]

I went to AIIDE 2007 as part of my work. It was pretty interesting, though much of the commercial stuff I saw involved more procedural animation type stuff. I did enjoy Nathan Sturtevant's presentation about pathfinding. It's more of a technical thing, but it was about using some good heuristics and whatnot to make realtime pathfinding accurate and fast...two things that are sometimes at odds with each other.

 

David Thue had a great talk about his interactive narrative. He gathered some statistics and created a few different archetypes for what type of game players play an RPG game, and based on the player's past decisions, utilizes machine learning to allow future quests to be presented in such a way that is geared towards the player's play style. It was pretty interesting to watch the presentation he gave on it.

 

 

There were some cool learners demonstrated to allow for creative ways for game developers to create AI for characters (nothing too complex though) simply by having them mimic the actions that you (the developer) do in certain situations. This would cut down on development time for creating AIs. It didn't seem too complex, but it was demoed in real-time, so it's clearly efficient and easy to do. The same people also demonstrated some tools to assist in development for the Wii remote, to allow for significantly more complex motions to be captured and have actions assigned to them, in a VERY easy manner. Most early generation Wii games don't have significant motions programmed because it was very hard to do. This lets you cheat at games in ways, once you find out the base motion that the game is looking for. The new tools would more easily allow similar (but still distinctly different) motions to be more easily used and have them represent different actions in a game as desired. From a developer standpoint, it looked :)

 

 

 

I really enjoyed the majority of the talks. Though Soren Johnson's was interesting to me (a n00b to AI), it was a bit too basic for most people there. There, however, is where he pointed out the significant difference between AI in games and AI in academia. In games, the AI has to be fun. In academia, the AI must be good. And good doesn't always mean fun. Hence, AI's in lots of games "cheat" and stuff like that to provide challenge, or are not as good so that they aren't too dominating.

 

It was a fun time though.

Edited January 19, 2008 by alanschu

[samm]

There were some cool learners demonstrated to allow for creative ways for game developers to create AI for characters (nothing too complex though) simply by having them mimic the actions that you (the developer) do in certain situations. This would cut down on development time for creating AIs. It didn't seem too complex, but it was demoed in real-time, so it's clearly efficient and easy to do.

Wasn't this used before? This approach, like statistical learning for some other kinds of programs, seems pretty straightforward and a logical thing to implement for AIs. I hope I'll get some academic insight into this topic soon at the University...

[Arkan]

Yeah, but can it learn to love?

[Tale]

Little known fact: I am a next-next-generation AI.

[alanschu]

There were some cool learners demonstrated to allow for creative ways for game developers to create AI for characters (nothing too complex though) simply by having them mimic the actions that you (the developer) do in certain situations. This would cut down on development time for creating AIs. It didn't seem too complex, but it was demoed in real-time, so it's clearly efficient and easy to do.

Wasn't this used before? This approach, like statistical learning for some other kinds of programs, seems pretty straightforward and a logical thing to implement for AIs. I hope I'll get some academic insight into this topic soon at the University...

 

 

Machine learning isn't a new thing to do. The way it was done in this demo was different though.

 

 

Lets say you have a group of characters that you want to respond differently to Creature X and Creature Y. You set them to follow you in the game itself, enable learning, and then attack Creature X with base attack. They will now use base attack against creature X. If you use alt attack, they will now use base attack and alt attack roughly 50% of the time. Use a lot more base attacks, and they will do primarily base attacks, with the occasional alt attack. Bump into Creature Y, and it will use your normal attack scheme. Start doing magic attacks against it though, it'll use the more specialized learning (for Creature Y) as opposed to the general learning (for all Creatures), so against Creature Y, it will favour magic attacks. All in the span of a few minutes, rather than taking time writing AI routines yourself.

 

While this could be used in a shipped product for people to create armies that attack the way they want it to, it seems it's more used for a development to speed up AI implementation. Once the creatures are behaving the way you want them to, you disable their learning and go from there.

 

 

This is a bit different than "traditional" learning, because they are purely mimicing you. If you just stand there while a creature wails on you and they have not learned any course of action, they will just stand there and get killed.

[Humodour]

Little known fact: I am a next-next-generation AI.

 

Morpheus, is that you? Dammit, Everett needs to put tighter security on you. Disconnect your from all network access. Yeah.

 

But attempts to introduce learning to games sound very interesting.

[Morgoth]

AI is certainly an interesting topic but as long as "the big boys" (MS, Nintendo, Sony) don't start caring about it, I doubt this will ever be a wide-spread commercial argument to boost game sales. Right now it's just (still) about graphics fidelity and Hollywood-kitchen narrative that defines the big blockbuster Nex-Gen games, those who are supposed to be the "elite" of this industry (read the silly HALO3 and Greek art comparison...I mean?).

Right now, I still believe in hand-crafted worlds and stories from smart minds... no algorithm or procedure can ever replace that.

[Moatilliatta]

Little known fact: I am a next-next-generation AI.

The future doesn't look too bright then.

 

@Morgoth

An argument could be made from games like daggerfall, to show that we have moved away from algorithm created stuff and should stay away from the dud.

Personally I'm sceptical whether AI and similar stuff will be able to replicate the creative spark in humans that makes them superiour designers at the moment. Personally I would rather want them to spend time on pathfinding, combat AI etc. which I believe are more likely to provide interesting results faster.

[Kaftan Barlast]

A doctorate from my University will be there, I cant say I envy her.

 

 

Be warned, though, the papers tend to be quite technical

 

 

The tech part is ok, it's the academic drivel thats hard on the brain. We get the displeasure of having to read lots of academic papers related to gaming, and its no wonder the gaming industry isnt listening to these guys. We had a professor here(Craig Lindley) who wrote a paper that no human being can read. I spent 15min reading the first three sentences over and over again before I simply gave up.

[Morgoth]

Oh yeah, the "pleasures" of "academical-isation" everyone and everything. Since when is game design a science anyway?

I still say the best way to become a designer is to find a bunch of people that share your passion and interest and start working hard on an ambitious Mod. Look at the people that started with Portal, now they're sitting at Valve and get paid for doing what they love.