I am a grad. student at MIT wrapping up a thesis (just a few more days now!) on AI learning methods and games. It is an examination of the past and present of AI learning methods in games, and some thoughts about the future. I was wondering what some of you might think on some of the stuff I tried to discuss in my thesis...My main question is: "What do you think is going to be the future of AI learning methods in games? And, how should people begin to think about using AI learning techniques in games?" I'd be very curious to hear what you folks might think. I've listed a few more specific questions down below, to give people a better/more specific idea of what I am wondering about. Thanks in advance! Christian Baekkelund (draco at mit dot edu) P.S., Due to the cross-disciplinary nature of these questions, I posted this here to the game design forum, to try to get more of a feel on what people think with regards to game design, but I also made a similar post on the AI forum to try to get at the technical issues. Not just blindly spamming everything... :) ----------------- If anyone wants a bit more clarification on what I'm thinking about or have a bit more time, here's some more specific additional questions: - In addition to "the future role of learning methods in games", where and how have past attempts at using AI learning techniques in games has succeeded or failed? - Where, in terms of design, could/would AI learning be mostly beneficial, and how should designers begin to think of the use AI learning techniques? - Also, from a design standpoint, is there anything very specific that a designer wishes he could "know" while designing a game? (Like "the player up to this point in the game has favored this kind of attack vs. these kind of foes", or some other specific issue that you currently do not have access to, but wish you could from a design standpoint.) - Is there a particular type of game or genre of games that the use of some learning methods would especially benefit from? - What new genres of games do folks think would come about with better and easier use of AI learning techniques in games?

I have little to contribute, but I'd like to ask if we could read your thesis once you've finished it. It sounds very interesting, and I for one would appreciate an opportunity to enjoy it.

For what it's worth, I'd like to see AI skills replace "+2 against undead" and the like in RPGs. A combat system that replaces a "level 7 sword skill" with actual AI decisions based on training and previous encounters would make "Experience" more than just a point score. It would lead to organic character development and a real sense of character history.

Quote:

Original post by Iron Chef Carnage I have little to contribute, but I'd like to ask if we could read your thesis once you've finished it. It sounds very interesting, and I for one would appreciate an opportunity to enjoy it.

Sure thing...it'll be available publically.

Quote:

For what it's worth, I'd like to see AI skills replace "+2 against undead" and the like in RPGs. A combat system that replaces a "level 7 sword skill" with actual AI decisions based on training and previous encounters would make "Experience" more than just a point score. It would lead to organic character development and a real sense of character history.

Hmm...I'm not sure how the "+2 against undead" would be changed? But the "actual experience" making a difference is a more interesting concept. The question is, would it really matter? Does it really make that much of a gameplay difference in the end? I'm still unsure...

For a good idea of what I'm thinking of, take a look at Wavinator's post on fear. Characters with a good learning ability would be able to become experienced with regard to specific circumstances and events. The first time a soldier encounters a firefight, dead child, alien or zombie, it would be traumatic, and the character would not be able to respond effectively. With time and experience, they could become not only used to the stimulus, but could build a library of knowledge regarding it. They would begin to expect convenient treasures to be boobytrapped, or learn that going for headshots against los ganados can sometimes exacerbate the problem, rather than solving it.

In a small, controlled game, this sort of thing can be done manually--"fudged" by careful worldbuilding. But in an arena where convincing NPC characters have to adapt to unforeseen circumstances--new tactics in an RTS game, or player-designed vehicles in a space shooter, for example--then AI learning systems would be a more elegant solution.

I remember something like this was used in tournament fighter games some time ago. Killer Instinct was the first time I was really impressed by an AI caharacter's ability to learn my gameplay habits, adapt to them, and force me to adjust accordingly. I think that AI "players" will begin to appear in games like Civilization and RTS games, so a player can choose an AI opponent from a roster, and face not only the AI that was programmed into it, but also the experience that the artificial player has acrued over time. Maybe chess would be a good start for this technology.

Quote:

Original post by ChristianBaekkelund Hmm...I'm not sure how the "+2 against undead" would be changed? But the "actual experience" making a difference is a more interesting concept. The question is, would it really matter? Does it really make that much of a gameplay difference in the end? I'm still unsure...

In that case, i'd say the benefits wouldn't justify the cost. There are other much better suited scenarios to spend CPU cycles. Hiding the numbers would do the trick probably... and putting some fancy words on the box. like... "Intelligent learning combat system I.L.C.S."
That'd make most folks happy =) until they found out of course. Years later.

Anyways, back on topic: What i'd like to see (or figure out) would be a way to express a 2D image as a coherent piece of information you could feed to current learning systems (most based on 1D information).

Quote:

Original post by ChristianBaekkelund My main question is: "What do you think is going to be the future of AI learning methods in games? And, how should people begin to think about using AI learning techniques in games?"

I was listening to Will Wright at the Stanford Game Design Lectures, and his take is that the 'behavior' wars have just begun (quoted on context that the graphics and the hardware wars were over).

To me, this means (and I know it is asking a lot of AI) that the more you can create (in terms of a relationship) you can put in a game (this is more than a conversation engine script thought out by a human to simulate small parts of advancing gameplay though NPC interaction/relationship) human behavior either on the choice of helping or not, hidering or not, supplying or not, detracting or not, from the avatar the player has invested themselves in, so that emotions, progress and things affecting the victory condition are on a higher level of presentation (speaking as a writer/designer) than today.

This is going to let the player make friends (in small ways), argue with, help, fight alongside or the opposite, with an entity. This is where it is going, and this is where enjoyment out of a game has amazing possibility. I know that AI is some tough stuff to design and implement, and, to make it as real as possible for the player in the gameworld is going to be quite a challenge, but this is going to be huge take away value for the player.

Quote:

If anyone wants a bit more clarification on what I'm thinking about or have a bit more time, here's some more specific additional questions: - In addition to "the future role of learning methods in games", where and how have past attempts at using AI learning techniques in games has succeeded or failed?

Speaking from a player standpoint looking for things to help me advance through a difficult game, some of the things I would have liked more is the simulation of getting information from an NPC that was not one way or another, or just repetitive clue exposition. Human beings in life (and if art imitates life) in games gather an amazing amout of intelligence and information from communicative interaction with others, for better or worse.

To be able to enhance this an order or two would make things incredibly realisitic, especially in war or detective games. If I knew what an AI was capable of doing in terms of response, I could write a scene where the player had a chance to be a CIA operative conducting an interrogation, or a detective like Sam Spade or Mike Hammer deceiving or interviewing a witness or unsuspective oppositional character, giving my player an amazing amount of enjoyment because they could effect the game with 'mental mechanics' instead of what the I/O design as a function of game/avater/NPC mechanics. This kind of entertainment value is going to change the way we play forever. Not to mention the kind of sums of money this level of interactive take away value will be worth retail.

Quote:

- Where, in terms of design, could/would AI learning be mostly beneficial, and how should designers begin to think of the use AI learning techniques?

Certainly any instructional material is going to be presented better in a visual environment and methods, but if you could go back as Neo and challenge Morpheus time and time again in philosophy besides the martial arts, you'd get replayability you'd have to program a limitation in replays into, just to get a fair value of sales to production ratio. There are a few other applications I could think of, but I can't discuss them in public due to proprietary restrictions.

Quote:

- Also, from a design standpoint, is there anything very specific that a designer wishes he could "know" while designing a game? (Like "the player up to this point in the game has favored this kind of attack vs. these kind of foes", or some other specific issue that you currently do not have access to, but wish you could from a design standpoint.)

Like I said before, a clear delineation of capabilities and responses will allow design choices we can balance in not only the game play, but in the mind play.

Quote:

- Is there a particular type of game or genre of games that the use of some learning methods would especially benefit from?

Hard to say, because I am biased towards adventure games, and not expert enough in other genres to advise well.

Quote:

- What new genres of games do folks think would come about with better and easier use of AI learning techniques in games?

That I think will be entirely dependent on what the AI could be capable of representing behaviorally, then designers can adapt the AI type or form, to the function, or challenge design of the game.

Link us to the paper, and HTH

Adventuredesign

although, its unrelated, I would like to how would i get into MIT, what I need to get in .etc as I really really want to go to MIT to study Maths and Computer Science.
Thanks.

Quote:

Original post by ChristianBaekkelund "What do you think is going to be the future of AI learning methods in games? "

Im by no means an expert in this field, however, since I will soon be asking to view the thesis (oops, just did), I feel I should offer my .02 cents.

I would think that for AI Learning specifically, we would want to mimic that of our own sense of learning. Dynamically creating questions for the AI to answer themselves, will only in turn create a more educated AI system, and thus most likely form new questions about the findings, much like humans do. I would think that the ultimate AI would indeed find the answers to the questions which are posed, and then compile a new list of questions about its recent findings.

Oh, and if you need, I can send you an email as a reminder for the thesis. ;-)