Hello Everybody:
Thought I might mention that I''ve had a new poll up for a bit that some folks might want to take part in. It''s a generally annual poll and this is the third year for it.
The question is, "Where do you think the next innovation in game AI will come from?" Specifically, what genre of game do you think might advance game AI the next notch?
You can weigh in at www.gameai.com.
Ferretman
ferretman@gameai.com
www.gameai.com
From the High Mountains of Colorado
🎉 Celebrating 25 Years of GameDev.net! 🎉
Not many can claim 25 years on the Internet! Join us in celebrating this milestone. Learn more about our history, and thank you for being a part of our community!
Where do you think the next innovation in game AI will come from?
I''ve thought about this last time you posted the question and I''ve voted at the boards but I think this question is missing the point a bit.
I''m going to make some generalised bold statements here which aren''t completely true but are a simplified version of the truth as I see it.
First of all the majority of theoretical leaps forward in AI comes from academia. In fact I can''t think of one that hasn''t. The basic ideas, the paradigms and methodologies that act as frameworks for solving the problems pretty much exclusively originate in academia. This is mostly because the questions of what is intelligence, what is life, what is this or that biological occurence and how can we model it are far older than the computer games industry by between a few hundred and a few thousand years. Finite state machines, fuzzy logic, neural networks, evolution, theoretical biology, state-space searches, any kind of philosophical standpoint and all associated paraphernalia were originated in academia and the advances in these fields, as well as the spawning of new ones, seem likely to remain there.
The question of military AI is out of my field of knowledge but few of the advances in AI the military develop are a)new in scope (they are interesting implementations of academic ideas), b) relevant to computer games or c)likely to be anything other than national secrets for the next 50 years.
Steve, you''ve worked with the military before if I remember rightly. Am I way of the mark here?
I may seem to think lowly of computer games from the above statements and what I''ve said stands in terms of revolutionary paradigms for AI. However, in terms of amounts of implementation that occurs in AI, I doubt there is a field that has the sheer quantity of code written, implementations and reimplementations that computer games has. The number of hobbyist projects, professional games, testbeds and the like that computer games programmers create that involve AI is surely huge compared to any other medium.
So really the question is, what innovation from academia will be the next innovation for computer game AI and what genre of game will it come with.
If I was to guess about the innovation I would say it would have to be adaptable AI. AI that learns while you play and beats you at your own game or, in terms of AI assistance, helps you beat an enemy at their own game. My reason for this is that almost every other kind of AI has been implemented to some reasonable degree, wheras this kind of AI is not one I''ve seen outside of B&W.
The problem with enemy AI is that almost any programmer could write AI to do a job perfectly within a constrained world. FPS AI can be made to calculate your trajectory and fire that rocket between your eyes from the other side of the map. An rts AI simply has a level of control over every facet of their troops allowing them to adjust tactics instantly and never forget to set that next tank building, while you''re struggling to figure out where to place your first building. AI programmers can also cheat and are fully justified in doing so. There is no point in sticking to the rules if the player will gain in gameplay for your indiscretions.
So any adaptibility will have to increase gameplay by making the enemy better in some way that makes the game more fun to play. Either by impressing you that they are adapting to your tactics or by the fact that the world the AI is entering into is unknown to the AI coder and they have to adapt to that specific world quickly. Alternatively the AI you want to adapt may be those 4 other members of your Unreal Tournament team, who you want to act differently to your orders than they have been so far. There''s only so much functionality that you can allow the player to explicitly play with before an FPS becomes a micromanagement game. AI team mates that you can order to perform an action then berate or congratulate based on how they performed it in such a way that they adapted to your ideas would be one such innovation dragged from academia into computer games. As to which genre this kind of idea will enter into, I''ve no idea. That all depends on office politics, publisher demands, capability of the AI coder and trust in that capability by management and team members. Whether this happens in an rts, fps or rpg I don''t know. These genres encapsulate so many actual types of game that adaptibility may play a part in the future AI of any of them.
As I''ve said, this type of reinforcement learning can be seen in B&W, however there is still scope for unsupervised or evolutionary learning methods to translate to computer games, where the AI arrives in a world and adapts without your, or anyone''s, direction.
Just my $0.02 bet.
Mike
I''m going to make some generalised bold statements here which aren''t completely true but are a simplified version of the truth as I see it.
First of all the majority of theoretical leaps forward in AI comes from academia. In fact I can''t think of one that hasn''t. The basic ideas, the paradigms and methodologies that act as frameworks for solving the problems pretty much exclusively originate in academia. This is mostly because the questions of what is intelligence, what is life, what is this or that biological occurence and how can we model it are far older than the computer games industry by between a few hundred and a few thousand years. Finite state machines, fuzzy logic, neural networks, evolution, theoretical biology, state-space searches, any kind of philosophical standpoint and all associated paraphernalia were originated in academia and the advances in these fields, as well as the spawning of new ones, seem likely to remain there.
The question of military AI is out of my field of knowledge but few of the advances in AI the military develop are a)new in scope (they are interesting implementations of academic ideas), b) relevant to computer games or c)likely to be anything other than national secrets for the next 50 years.
Steve, you''ve worked with the military before if I remember rightly. Am I way of the mark here?
I may seem to think lowly of computer games from the above statements and what I''ve said stands in terms of revolutionary paradigms for AI. However, in terms of amounts of implementation that occurs in AI, I doubt there is a field that has the sheer quantity of code written, implementations and reimplementations that computer games has. The number of hobbyist projects, professional games, testbeds and the like that computer games programmers create that involve AI is surely huge compared to any other medium.
So really the question is, what innovation from academia will be the next innovation for computer game AI and what genre of game will it come with.
If I was to guess about the innovation I would say it would have to be adaptable AI. AI that learns while you play and beats you at your own game or, in terms of AI assistance, helps you beat an enemy at their own game. My reason for this is that almost every other kind of AI has been implemented to some reasonable degree, wheras this kind of AI is not one I''ve seen outside of B&W.
The problem with enemy AI is that almost any programmer could write AI to do a job perfectly within a constrained world. FPS AI can be made to calculate your trajectory and fire that rocket between your eyes from the other side of the map. An rts AI simply has a level of control over every facet of their troops allowing them to adjust tactics instantly and never forget to set that next tank building, while you''re struggling to figure out where to place your first building. AI programmers can also cheat and are fully justified in doing so. There is no point in sticking to the rules if the player will gain in gameplay for your indiscretions.
So any adaptibility will have to increase gameplay by making the enemy better in some way that makes the game more fun to play. Either by impressing you that they are adapting to your tactics or by the fact that the world the AI is entering into is unknown to the AI coder and they have to adapt to that specific world quickly. Alternatively the AI you want to adapt may be those 4 other members of your Unreal Tournament team, who you want to act differently to your orders than they have been so far. There''s only so much functionality that you can allow the player to explicitly play with before an FPS becomes a micromanagement game. AI team mates that you can order to perform an action then berate or congratulate based on how they performed it in such a way that they adapted to your ideas would be one such innovation dragged from academia into computer games. As to which genre this kind of idea will enter into, I''ve no idea. That all depends on office politics, publisher demands, capability of the AI coder and trust in that capability by management and team members. Whether this happens in an rts, fps or rpg I don''t know. These genres encapsulate so many actual types of game that adaptibility may play a part in the future AI of any of them.
As I''ve said, this type of reinforcement learning can be seen in B&W, however there is still scope for unsupervised or evolutionary learning methods to translate to computer games, where the AI arrives in a world and adapts without your, or anyone''s, direction.
Just my $0.02 bet.
Mike
Author
quote: Original post by MikeD
I''ve thought about this last time you posted the question and I''ve voted at the boards but I think this question is missing the point a bit.
[/quote[
Fair enough. The question originally came from the earlier GDC AI roundtables where it was asked somewhat often. I thought (still do) think it makes for a good general question, which is why I try to run a poll once a year. What interests me the most about the answers is how they seem to change a bit as the genres wax and wane from year to year.
First of all the majority of theoretical leaps forward in AI comes from academia. In fact I can''t think of one that hasn''t. The basic ideas, the paradigms and methodologies that act as frameworks for solving the problems pretty much exclusively originate in academia. This is mostly because the questions of what is intelligence, what is life, what is this or that biological occurence and how can we model it are far older than the computer games industry by between a few hundred and a few thousand years. Finite state machines, fuzzy logic, neural networks, evolution, theoretical biology, state-space searches, any kind of philosophical standpoint and all associated paraphernalia were originated in academia and the advances in these fields, as well as the spawning of new ones, seem likely to remain there.
Hmmm….I''m afraid I''ve got to disagree with this. While I think academia has done yeoman''s work, I think most of the real leaps have come from the gaming side of the house rather than academia.
Hmmmm again….I do note that you say, "theoretical leaps forward in AI", so possibly I''m misquoting you here. I guess I''d agree that most of the theory gets advanced in the academic realm, while most of the practice gets advanced in the gaming realm.
quote:
The question of military AI is out of my field of knowledge but few of the advances in AI the military develop are a)new in scope (they are interesting implementations of academic ideas), b) relevant to computer games or c)likely to be anything other than national secrets for the next 50 years.
Steve, you''ve worked with the military before if I remember rightly. Am I way of the mark here?
No, you''re absolutely right (from my point of view, anwyay); I just didn''t want to say anything around the poll so as to avoid skewing the results. But yes, in general I''d say the military has historically been the most "conservative" customer when it came to AI. Part of that is due to the general immaturity of simulators until recently; part of that has been (in the American culture, anyway) an insistence of "human in the loop" for control of lethal weaponry. That''s changing nowadays especially in the simulator world but at present it''s unlikely (IMO) we''ll see big advances come out of that realm.
quote:
I may seem to think lowly of computer games from the above statements and what I''ve said stands in terms of revolutionary paradigms for AI. However, in terms of amounts of implementation that occurs in AI, I doubt there is a field that has the sheer quantity of code written, implementations and reimplementations that computer games has. The number of hobbyist projects, professional games, testbeds and the like that computer games programmers create that involve AI is surely huge compared to any other medium.
We''re 100% agreed.
quote:
So really the question is, what innovation from academia will be the next innovation for computer game AI and what genre of game will it come with.
Here''s where I think we''ve got a disconnect. Based on my impressions from the last GDC I''m not sure many developers are actually paying attention to what''s going on in academia, nor am I convinced that academia in general is doing much to publicize what they''re doing. Both the gaming and the academic fields have great respect for each other, it seems, but if anything I''d have to say the academics are paying closer attention to the "practical" solutions of game developers than developers are paying attention to the latest "boring new theory" of academia.
quote:
If I was to guess about the innovation I would say it would have to be adaptable AI. AI that learns while you play and beats you at your own game or, in terms of AI assistance, helps you beat an enemy at their own game. My reason for this is that almost every other kind of AI has been implemented to some reasonable degree, wheras this kind of AI is not one I''ve seen outside of B&W.
There''s a lot of reason to think this, that''s for sure, and it''s a reasonable choice for where the next innovation might come from. Since you''re talking a technology more than a genre I can see why the poll isn''t quite along the lines you''d want.
For what it''s worth I did have a poll back in July, 1999 that asked, "What AI technology are you hoping will be "the thing" to make the difference in your next game?" that dealt more with technologies than genres. The results were interesting and might be more along the lines of what you''re looking for. (If you want to see that poll''s results, go to www.gameai.com, click on "Old Polls" on the left, and scroll down…it''s Poll #4, I believe.)
quote:
As I''ve said, this type of reinforcement learning can be seen in B&W, however there is still scope for unsupervised or evolutionary learning methods to translate to computer games, where the AI arrives in a world and adapts without your, or anyone''s, direction.
Agreed in principle. You open up an interesting can of worms, though, that also needs solving. What if an AI gets "too hard"…can it be reset? How do you prevent an AI from learning something "dumb"? Do we have different AIs for different players on the same computer? What if their skill levels are wildy differing? Etc.
quote:
Just my $0.02 bet.
You raise some great points, actually…..thanks Mike!
Ferretman
ferretman@gameai.com
www.gameai.com
From the High Mountains of Colorado
well, we game programmers are the only people who can push the ai technolegy forward, because we are the smartest and the most creative, and the only reason the AI development is going so slow is because our audiance are mainly kids, and kids like games with explotions, and thus the demand for AI is not as intensive as for graphics,
so maybe, in a few years, when more and more people pick up the lasted console and accept it as TV shows and movies, there will be a demand for none violent and smart game, and by then, game development groups will turn their focus on AI,
conclution, consumers are the only force that can push the AI technolegy forward
so maybe, in a few years, when more and more people pick up the lasted console and accept it as TV shows and movies, there will be a demand for none violent and smart game, and by then, game development groups will turn their focus on AI,
conclution, consumers are the only force that can push the AI technolegy forward
quote: Original post by Ferretman
Here''s where I think we''ve got a disconnect. Based on my impressions from the last GDC I''m not sure many developers are actually paying attention to what''s going on in academia, nor am I convinced that academia in general is doing much to publicize what they''re doing. Both the gaming and the academic fields have great respect for each other, it seems, but if anything I''d have to say the academics are paying closer attention to the "practical" solutions of game developers than developers are paying attention to the latest "boring new theory" of academia.
This is a problem in industrial/academic interaction in just about every field. I read a study that concluded innovations that begin in academia take 5-10 years, often more, to reach industry. There are a lot of new and astounding technologies that are ''old hat'' to academics that are just finally becoming wide-spread now. The real innovations introduced by industry (game programmers) seem to usually be just speed or space optimizations on an algorithm that is already known by academics.
There''s plenty of stuff going on in Expert Systems or Machine Learning in the universities that puts all the comercially available stuff to shame. The problem seems to lie in the fact that researchers often publish their research paper, but don''t publish their source code along with it. (University legal departments are often to blame for that.)
Isn''t it a bit incorrect to compare academic AI with game AI? After all, academic AI doesn''t have to perform in real time under the pressure of competition. Doesn''t game AI often give the illusion of AI without have a complete, true AI system?
-Kirk
-Kirk
quote:
well, we game programmers are the only people who can push the ai technolegy forward
I''m not sure if you were only joking, but in any case, I disagree with this view. I believe any person or group working with AI has the ability to push the technology forward.
quote:
Isn''t it a bit incorrect to compare academic AI with game AI? After all, academic AI doesn''t have to perform in real time under the pressure of competition. Doesn''t game AI often give the illusion of AI without have a complete, true AI system?
I think that it is ok to compare academic AI with game AI as long as you compare each AI with how well it does it''s own specific task.
Invader X
Invader''s Realm
quote: Original post by kirkd
Isn''t it a bit incorrect to compare academic AI with game AI? After all, academic AI doesn''t have to perform in real time under the pressure of competition. Doesn''t game AI often give the illusion of AI without have a complete, true AI system?
-Kirk
It is certainly not true that ''academic AI'' does not need to "perform in real time under the pressure of competition". Pressure comes from the natural selection between competing academics (competing for funding, jobs and prestige). If your method (or the one you chose to implement) doesn''t match up in speed and accuracy with some other method, then it simply dies a horrible death, unless it has some other particular quality that researchers and users like (eg, ease of implementation). On the issue of real time, while it is true that many academics doing pure research still play with ''toy worlds problems'', many of us are building systems that work in real world domains in real time. Game AI cannot claim that, as the domain of the game AI is a very constrained, well defined game world where any of the causal relationships can be varied.
As for the comment by the anonymous poster that only game programmers can push AI technology forward, I would beg to differ. In fact, I would go so far as to say that the only innovations in AI technology produced by the game industry are in implementation efficiency and in showing that a particular algorithm can be applied to a particular gaming problem. Why? Because the industry does not yet support research.
This is not to say that the game industry does not make a valuable contribution to the development and uptake of AI technology. Indeed, by using AI in their games, developers are creating an environment of natural selection for the innovations of academia.
Academia is best suited to research and development of new AI technologies. It''s their bread and butter. Industries like computer games are best suited for creating an environment of competition between these algorithms and showing the world which algorithms work best on tough problems.
I envisage that within 10 years we will see much closer ties between academic researchers in AI and game development studios. Why? Because many academics see the games industry as the next ''stage'' upon which they can develop and test theories. Small toy worlds are becoming old hat. Increased computation speed and power has enabled many academics to work on real world domains, however these still have their problems and applicable domains are few and far between. Game worlds are the ideal next candidate for domains to work with as they encapsulate most of the important relationships: agent-agent, agent-environment, agent-human, human-environment.
Anyway, enough from me.
Regards,
Timkin
I agree that academia is THE main driving force in evolution of AI, definitely not
game programmers. Simply because academics are constantly building on the latest
research made, while game programmers are just reaping 20 year old academic results.
But then ofcourse, games don''t need the latest in the field, partly because the latest
in the field is so computationally intensive. The most interesting results today are
on massively multiprocessing machines anyway.
I also get pissed off the fact (hello CheeseGrater) that the science papers I read loosely
describe this great new way to optimize rigid-body dynamics or whatever, but DON''T GIVE
ANY GODDAMN ALGORITHM DESCRIPTION! Not even pseudoalgorithms.
But I suppose we all want to eat, even scientists...
game programmers. Simply because academics are constantly building on the latest
research made, while game programmers are just reaping 20 year old academic results.
But then ofcourse, games don''t need the latest in the field, partly because the latest
in the field is so computationally intensive. The most interesting results today are
on massively multiprocessing machines anyway.
I also get pissed off the fact (hello CheeseGrater) that the science papers I read loosely
describe this great new way to optimize rigid-body dynamics or whatever, but DON''T GIVE
ANY GODDAMN ALGORITHM DESCRIPTION! Not even pseudoalgorithms.
But I suppose we all want to eat, even scientists...
Hi!
We have to differ between academia and industry in this case. Academia invents the _new formalisms_,
industri invents the _applications_ thereof.
I hardly believe that "quake IV" will invent the followup to neural nets or fuzzy logic it will happen within
academia.
The industry is still trying to cope with fuzzy logic and NNs which where the hot topic within
academic computer science when I made my master thesis ten years ago. (time flies -- sigh)
As a comparison -- it was only about 2-3 years ago the 386 processor got flight security approval,
i.e. you are allowed to use a 386 processor in your safety critical applications in an aircraft.
just my $0.02
/trysil
"A witty and slightly sarcastic quote from an unkown source"
-- The generic SIG
/trysil
We have to differ between academia and industry in this case. Academia invents the _new formalisms_,
industri invents the _applications_ thereof.
I hardly believe that "quake IV" will invent the followup to neural nets or fuzzy logic it will happen within
academia.
The industry is still trying to cope with fuzzy logic and NNs which where the hot topic within
academic computer science when I made my master thesis ten years ago. (time flies -- sigh)
As a comparison -- it was only about 2-3 years ago the 386 processor got flight security approval,
i.e. you are allowed to use a 386 processor in your safety critical applications in an aircraft.
just my $0.02
/trysil
"A witty and slightly sarcastic quote from an unkown source"
-- The generic SIG
/trysil
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
