What do you believe is the future of game AI ?

#1 Sorry misunderstood that
1) no

2) yes

3) yes

4) Because the current AI will always do the same thing when put into a certain situation an air of randomness would help.

5) Some games would benefit from them like fps's and strategy games but others it would cost to much energy for not enough of a reward.

6) That would depend what you considered to be intellegent an AI posing as a hill billy shouldn't be Einstien.

7) how they react to the same situation ex. if one runs into a rocket fight with a pistol and the other sticks back and looks for a rocket he would be smarter.

#2 1) No.

2) Yes.

3) Most definitely.

4) Because of this and this. (You can find the videos and more articles here)

5) In any situation where not all events would be entirely or almost entirely scripted.

6) Yes.

7) The best measure of intelligence in this case would be ability to properly adapt to new scenarios that had not been considered in the original programming.

#3 1) Yes, vaguely.

2) Yes, vaguely.

3) Yes.

4) Most technologies are designed to replace the currently used methods which have drawbacks. It's only logical that it would be an imporvement, unless their intent is trivial. (sorry, no details)

5) I do not understand the techniques enough to get into detail about the best places of integration.

6) Yes.

7) The human-like qualities of AI can be measured by the dynamics of its reactions.

7) Sidenote: I wouldn't usually fit "human" and "intelligent" into the same category of measurement (humans make stupid decisions based on emotion), but I'll let this slide since of course we are speaking in the context of AI. ;)

#4 1) Yes
2) Yes
3) Yes
4) This would allow AI to have a more flexible 'brain' and allow them to make non-predetermined decisions
5) Robotics, mechnical and medical exploration
6) Yes, perhaps too much (overkill) for a casual game
7) Set a number of equal tasks (goals) for each AI, and record which ones complete them, how long it took for each, and how much permutation was required to achieve the goals...

#5 When you are talking about game AI, it's important to distinguish between "intelligent" and "fun". Most people making games are concerned with "fun". If they want the AI to be more intelligent, it's cheaper to cut corners and give the AI knowledge it's not supposed to have.

Questions of intelligence apply to games like chess at a very high level... but at a low level, "fun" or "educational" is more important.

That said:
1) yes
2) yes (see this post
3) "better" is a useless word.
4) Because it depends on the purpose of the AI
5) Where appropriate -- generally those techniques are not used very often in games because game AIs usually don't need to be incredibly complex or accurate, they just have to have the appearance of intelligence and accuracy
6) no
7) Depends upon the purpose of the AI. If I took two robots that were intended to attach the car doors of a Nissan, then the one that was better at it would be more intelligent (though probably not more human)

#6 1) Yes

2) Yes

3) In a non game environment sure, however a game AI needs something different.

4) An proper game needs AI to have certain information and ignore certain other things, or it loses the "fun aspect"

5) Those techniques are good for prediction models of common behavior, as well as for puzzle solving skills, however they are not for engagement and interaction with humans

6) No as a matter of fact the most "human" AI I have ever met is named Eliza, she is a chatterbot and simply looks at what you are saying and spits it back out in some form of question or another, randomly throwing in reassurance etc. Not at all a complicated algorithm.

7) You are confusing Intelligence with Humanity, people don't identify with at all with intelligence in an adversarial position for instance a game. I would even venture to say that people don't identify with intelligence, they identify with emotion. For example, the most emotional game moment I ever had was in Dink Smallwood when Dinks home burns and it kills his mother. He wasn't around to save her, and blamed himself for the fire. This struck a chord with me and I cried, first time and only time I ever felt tears flow for a "game moment". This game was a farce, a parody of an RPG and yet that moment still chokes me up. It was entirely scripted though, not at all AI based. Contrast this to the best game AI's around. How many people get frustrated at the string parsed AI in Everquest, where if you say something the NPC is supposed to respond, and they usually respond in extremely unexpected ways. This tends to be more frustrating than enjoyable and is one reason WoW is so much more popular than EQ. In WoW, NPC interaction is simplified to a series of guided choices.

#7 Lol - How can you compare AI in EQ to WOW? Verant wrote that in 97-99 and it was / is damn good for it's time.

I think AI is just going to get better and better. I know looking short term I would like to see one of the AI Packs
for TGE completed. Hopefully Phil's.

#8 Because I'm not. I'm saying it was frustrating to deal with then, it's frustrating to deal with now.
WoW had the capability to do even better than EQ especially with the algorithms and processing power that are available now. They instead decided to go with a much, more simplified way of interacting, verant could easily have done the same back then.

#9 Here's my 5 cents:

>> 1) Are you aquainted with the term Computational Inteligence ( or Non-deterministic AI ) ?
No

>> 2) Have you ever heard about Neural Networks, Fuzzy Logic or Genetic Algorithms ?
Yes

>> 3) Do you believe that using any of the techniques above would yield better intelligence than a traditional AI approach

No (and yes)

>> 4) Why ?

Game AI isn't (much) about real intelligent, but more about fooling the gamer to believe or rather feel.
It's like game physics, it's not about real physics but the illusion of physics.
Using technics to simulate real intelligence might not be the best approach to the problem.

>> 5) Where you would use such techniques ?

Where you would use such techniques ?
Information management software (data mining, automatic data gathering and response tools)
Not to say this can be incorporated in a game (hence the No and Yes answer to question 3).


6) Do you consider that a more "human" NPC would also be more intelligent ?

No. For the same reason as above and that stated by others in this thread, human behaviour isn't the same as intelligence.

I think a humanoid should portray personality rather than intelligence.


7) How can you measure if an agent (NPC) is more intelligent ( "human" ) than other ?

That's a philosophical question, and in my opinion not very interesting for the applications your after.

A more interesting question would be, how does a real person perceive a NPC, like a machine or like a person?

#10

Quote:...and yet that moment still chokes me up. It was entirely scripted though, not at all AI based. Contrast this to the best game AI's around.

As you said, it was scripted. You can have as many as these scenes as you want without having anything to do with the AI. The AI could be simple, or extremely complex but the scene would be the same wouldn't it?

And to make my post somewhat worthy to the original poster:

1) No

2) Yes

3, 4, 5) I'm currently experimenting with this so I can't currently answer these. My teammate has some fuzzy logic code from his thesis work and we're in the process of applying this to Torque and our game to see if it turns out better. It's a horror/RPG by the way.

6) Yes

7) Alex and Dee provided great answers.

Nick

#11 Hmmm fuzzy logic in a horror/rpg should work, seems to in the movies.
Just remember the rules though. Never say you'll be right back you won't etc.

#12 Thanks, I hope it will cause we have some great code. What do you mean by "Never say you'll be right back you won't etc."?

#13 It's from "The rules" scene of Wes Cravens Movie "Scream".
More information

#14 Now I get it:)

#15 So this is for your master degree....uh ok.

if you ask me, intelligence in games is different from real-world intelligence. in a game, if a guy hides from gunfire hes considered intelligent, whereas in the real world thats just common sense. in a game, if a character is smoking, it makes the character seem more intelligent, whereas in the real world smoking is considered stupid. ha!

1. no
2. yes
3. no
4. I believe a realistic simulation of intelligence using today's technology would be based heavily on the situation as we are pretty limited when it comes to the interface. constructing a reasonable ai response to the given situation would be the best display of intelligence IMO.
5. i think these techniques are suited for games based on character interaction, rather than games which just feature character interaction.
6.yes
7.i think a good measure of intelligence is in the response ai charactrer shave

#16 My first job out of college was as an "AI Programmer" for a videogame company. I actually ended up spending as much time on weapons and special effects graphics coding as AI, but that was my original role. I haven't been keeping that up-to-date on the field, though.

1) Are you aquainted with the term Computational Inteligence ( or Non-deterministic AI ) ?
That's the new term for the collection of non-traditional AI techniques mentioned in #2, right?

2) Have you ever heard about Neural Networks, Fuzzy Logic or Genetic Algorithms ?
Heard of and to some degree implemented.

3) Do you beleive that using any of the techniques above would yield better inteligence than a traditional AI approach
In CERTAIN CIRCUMSTANCES, yes. I believe a neural network is being used for the recent indie release "Democracy." But they aren't a cure-all.

4) Why ?
I'm gonna say, "Why not:"
* GA's (Genetic Algorithms) are best-suited for solving static problem sets. Last I heard, some research had been done (and I played with some ideas myself) for creating problem-solving STRATEGIES in a dynamic environment, but I don't know that these have yielded any great success when compared to other techniques.
* Fuzzy Logic: I can't say about games, but I know that at one point Expert Systems developers found out that straightforward Bayesian math performed better than fuzzy. I haven't seen anything done with Fuzzy logic that in any way surprised me in comparison to more traditional techniques, so I've not had much interest in it.
* Neural Networks: These can and are being used now. The results are cool, however, they can be a maintenance headache. Like GA's, they will often require a significant amount of re-training when the problem set is modified (for play balance, or whatever). It's also hard to debug. But still intriguing.

5) Where you would use such techniques ?
So far strategy games show the greatest amount of potential for Neural Networks.

There are certain problem sets where traditional AI techniques really suck. The classic example is the boardgame "Go." It just can't break down to traditional brute-force techniques, and pure heuristic-based approaches can't deal with the scale and complexity of the problem. That's a good opportunity for these other techniques to shine - POTENTIALLY.

And of course, if you want AI to actually LEARN from each player and adapt to a player's strategy may require this kind of AI.

(CONTINUED NEXT POST)

#17 CONTINUED FROM LAST POST

6) Do you consider that a more "human" NPC would also be more inteligent ?
No.

A more "Human" NPC / opponent should behave more like a human. Whether that's more or less intelligent, skillful, or whatever depends upon the situation and the game. My AI Professor in college once taught us that 90% of AI programming is teaching the computer to "avoid stupid mistakes." Which is to say that the AI shouldn't make the kinds of mistakes a human is UNLIKELY to make. One other thing I learned doing games is that the AI needs to make "believable" mistakes.

In working on Void War, I found at one point that the enemy ships were WAY too smart. They could bullseye a flitting mosquito at 500 meters no problem. But was that fun? Heck no. The player would die by the time he could find whoever was shooting at him. The trick to making them feel more "human" was to make them MISS BELIEVABLY.

Players don't want opponents who are unbelievably intelligent and thoroughly capable of kicking their butt. What players want is AI that is *DRAMATIC*. But it's usually NOT the purpose of the AI to "win" the game - even in games like Chess (If it was, why would you ever play at a skill level below the highest?) The purpose of the AI is to give the player an EXCITING GAME. That's an artistic goal, not a mathematical one.


7) How can you measure if an agent (NPC) is more inteligent ( "human" ) than other ?
Turing Test. :)

See my response to #6. I don't think that is the goal. But in general, you'd measure it by their success - the same way you'd measure two players to see which one is the better player.

#18 1) Yes. well I think I may have come across the term in one of the programming gems books.
2) Oh, yeah. even my washing machine is supposed to have fuzzy logic. (though I think it is more likely Lint Covered Circuit.
3 & 4) Possibly, I don't follow AI news that much. All it takes is one clever person with a clever idea, and you'll get some damn clever algorithm.
5) Not fully experienced with the places, skills a techniques needed to impliment AI.
6 & 7) No, intelligence does not make us human (IMO), look around there are some pretty stupid people around, but they are classed as human, and from what various researchers have commented on is that some animals are quite intelligent, yet they are not human.
What I think seperates us as humans is(whether the mentioned Algorithms can do it), the ability to Learn collecting experiences (which most AI try to do), a moral compass giving us sense of right or wrong, and "gut feeling"(does this feel like the right choice).

#19 1) Are you aquainted with the term Computational Inteligence ( or Non-deterministic AI ) ?

Yes

2) Have you ever heard about Neural Networks, Fuzzy Logic or Genetic Algorithms ?

I have written examples of all three, so yes

3) Do you beleive that using any of the techniques above would yield better inteligence than a traditional AI approach

It depends entirely on the AI being implemented, and how the techniques are used. For example, a neural network can be great for machine learning, but it's not going to do a good job of making a character run around and jump in a 3D world without more traditional AI to accompany it.

4) Why ?

These techniques are just tools to create an artificial intelligence. The only difference between a neural network and the dot product is how they are used to create the correct behavior. I think AI programmers should stop thinking in terms of "can neural networks be smarter than traditional AI" and instead think of which techniques (or combination of techniques) can help achieve the intelligence you are looking for.

5) Where you would use such techniques ?

In systems that have too many factors to be able to fully model in another type of intelligence. For neural nets, they should be used for pattern recognition (which is what they excel at), but can also be used in situations such as vehicles with realistic physics. This is because a realistic physics system can be very difficult to predict with other standard AI approaches, and each time the system is tweaked the AI has to be re-tweaked. When using a neural net, you can instead just re-train it to take the new physics into account.

Genetics should be used whenever we need a non-deterministic system that should still inherit certain attributes. This can be in level generation, or learning systems for AI characters.

And "fuzzy logic" is just a fancy term for random numbers (yes, I realize I'm generalizing, but it's true to an extent). I think AI people just like to throw it around because it sounds better than random.

6) Do you consider that a more "human" NPC would also be more inteligent ?

I don't know, I've met a few humans that... (sorry, had to get that out). I think neural nets and such can be used to make an npc more "human", but I also think that alot of humanity is the small touches that happen when emotions take hold. This may never be properly modeled in a neural net, so instead it should be a balance between non-deterministic approaches to handle the logic, and designer-created actions to model the emotions.

7) How can you measure if an agent (NPC) is more inteligent ( "human" ) than other ?

I don't consider "Intelligent" and "Human" to be synonyms. An intelligent AI will be able to take the information from it's environment and make an informed decision based on it. A "human" AI will be able to take the information, extrapolate from it using its own knowledge and experiences, and then make an informed decision that is also weighted toward their current emotional state and all of their short and long-term goals. For example, the human sex drive actually affects most, if not all, of our decisions in a way that the computer can never model. When I'm thinking "look behind me", somewhere deep inside my sex drive is helping make that decision. So is my subconscious, my past experiences, the future experiences that I would like to have and not like to have, what I was told in a coffee shop three days ago, etc... Without models for all of these, it cannot be "human".

#20

Quote:1) Are you aquainted with the term Computational Inteligence ( or Non-deterministic AI ) ?

Only vaguely.

Quote:2) Have you ever heard about Neural Networks, Fuzzy Logic or Genetic Algorithms ?

Yes.

Quote:3) Do you beleive that using any of the techniques above would yield better inteligence than a traditional AI approach?

Possibly, but it depends entirely on the application of these approaches, and of course the specific implementation of each.

Quote:4) Why ?

Improved interaction between user/ player and digital entity essentially. The idea, as I see it is that humans don't operate on a finite decision set, it's why 100 people all posed with the same situation and circumstances can all arrive at different conclusions ranging from similar with mild differences to wildly different.

Quote:5) Where you would use such techniques ?

In theory the applications of such advanced digital "logic" woud be virtually limitless, but since this is a game -design forum, I'll asume you're asking in a game development specific context... anywhere that such an addition could stand to improve the overall experience for the end-user.

Whether using it to create truly dynamic, story driven content that was responsive to the players actions, or simply to teach enemy characters to duck for cover when they're standing next to a pile of crates and the player is hosing them with a machine gun.

Quote:6) Do you consider that a more "human" NPC would also be more inteligent ?

For the most part, yes, but not necessarily. In order to make a digital entity truly seem human, you'd have to give it the ability to make collosally stupid decisions from time to time that were driven by a motivating factor other than intellect or logic... like emotion. =\

Quote:7) How can you measure if an agent (NPC) is more inteligent ( "human" ) than other ?

As previously stated, the measurement would be largely situational, and a bit subjective in the end. It woud ultimately boil down IMHO to the defined purpose of the entity and how effectively they performed that task... meaning of course, soemthing more complicated than "add these numbers and give me the average" type of thing.