True RPG
by Robert Brim · in Game Design and Creative Issues · 03/13/2001 (10:18 am) · 7 replies
Not too long ago( a year or so), I read an article stating CPU power was not fast enough to build a truely interactive environament, in which the player could type (or maybe today say into a microphone) their interaction with a NPC Then the NPC would react to the character, not just give the same line over and over and over, or not be able to talk at all etc..
SO my idea is for someone to build a AI matrix that can truely react to a player's action or words.
If we could have that, we would effectively breach the event horizon on RPG games.
SO my idea is for someone to build a AI matrix that can truely react to a player's action or words.
If we could have that, we would effectively breach the event horizon on RPG games.
About the author
#2
I thought that the MMORPGs were going to be a more interesting venue for player roleplaying. Instead of walking through pre-determined stories designed by the game developer, the players would make their own stories. Unfortunately, it seems like all people are after is the next level or a power-up on most of the MMORPGs.
03/13/2001 (2:05 pm)
While I like the idea of allowing a player to use a microphone, I think this the idea of an NPC actually understanding the player is overkill for an RPG. Most players of RPGs accept that the game world has limits, and that they are not going to be able to sit around and have a unlimited conversation with NPCs. This is an acceptable limitation on game play.I thought that the MMORPGs were going to be a more interesting venue for player roleplaying. Instead of walking through pre-determined stories designed by the game developer, the players would make their own stories. Unfortunately, it seems like all people are after is the next level or a power-up on most of the MMORPGs.
#3
03/13/2001 (7:27 pm)
Technically the technology already exists to create an AI that could make conversation with a human. It's just a matter of bringing together two technologies that are already available; Speech recognition sofware and Turin chat bots (I think that's what they're called). I'm talking about the bots that are so well programmed that they're quite capable of fooling most people into thinking they're chatting with another person. Stick the two technologies together and you should have an AI that can keep up an intelligent conversation with a human. :)
#4
What Robert is talking about in the first post is the holy grail for speech research. I work at HRL Laboratories in a speech research team. A fully open dialoge system that is able to parse meaning from every utterance it recieves is just not possible right now. You must break the problem down into domains, where the scope of knowlege is limited on the computer side. Even this is a monumental problem, and is the current focus of our research. The breadth of language is too large for current computers (and humans!) to handle. Construting the grammer is exponential compared to the size and scope of the domain. True open state of the art systems have the knowlege and grammer equivilant to 3-4 year old child (I think, don't quote me, things change fast).
Don't mean to totally burst your bubble, but companies spend millions of dollars a year, just to try and solve/deal with a small domain that they are intrested in. Progress is being made, and that info does trickle down, but this is an ongoing research domain, and not a simple little problem that no one has been working on.
I think your best bet would to research directed dialoge systems, and implement them. This is what you have probably run into in most computer speech recognition areas. True, they are not open, and the boundries can be found pretty easily. This is why so much research is directed to open systems. But the are also doable, for the part time person.
If you really want to solve the open dialogue problem, get funding and do it full time. Get a research job, or get a grant, or go to school and find a PhD program that studies these things. This will give you the best chance of contributing to the field, and you'll get paid doing it too :).
Joel.
03/15/2001 (1:50 pm)
Haven't run across them, but I bet they are called Turing chat bots after Alan Turing, father of computer science and AI and all that fun stuff. Simply fooling a human for a little while and actually extracting meaning and acting on that information is two entirely different things.What Robert is talking about in the first post is the holy grail for speech research. I work at HRL Laboratories in a speech research team. A fully open dialoge system that is able to parse meaning from every utterance it recieves is just not possible right now. You must break the problem down into domains, where the scope of knowlege is limited on the computer side. Even this is a monumental problem, and is the current focus of our research. The breadth of language is too large for current computers (and humans!) to handle. Construting the grammer is exponential compared to the size and scope of the domain. True open state of the art systems have the knowlege and grammer equivilant to 3-4 year old child (I think, don't quote me, things change fast).
Don't mean to totally burst your bubble, but companies spend millions of dollars a year, just to try and solve/deal with a small domain that they are intrested in. Progress is being made, and that info does trickle down, but this is an ongoing research domain, and not a simple little problem that no one has been working on.
I think your best bet would to research directed dialoge systems, and implement them. This is what you have probably run into in most computer speech recognition areas. True, they are not open, and the boundries can be found pretty easily. This is why so much research is directed to open systems. But the are also doable, for the part time person.
If you really want to solve the open dialogue problem, get funding and do it full time. Get a research job, or get a grant, or go to school and find a PhD program that studies these things. This will give you the best chance of contributing to the field, and you'll get paid doing it too :).
Joel.
#5
I didn't realise how far we still are from making a truly 'intelligent' computer system i.e. one which can, as you said it, derive meaning from every utterance. Sounds like the holy grail of computing indeed. :)
/me runs off to get the education required to program a 'true AI'. ;)
03/16/2001 (7:30 pm)
I found out about my spelling error (Turin instead of Turing) shortly after posting my message. Woops. :)I didn't realise how far we still are from making a truly 'intelligent' computer system i.e. one which can, as you said it, derive meaning from every utterance. Sounds like the holy grail of computing indeed. :)
/me runs off to get the education required to program a 'true AI'. ;)
#6
There are multiple problems with existing technologies that we are focusing to develop upon and later release as an sdk for games and applications alike.
first and formost are current AI and chatter bot technologies. Most of the current chatterbots i find very plain and too demanding in that they want to control the conversation. While this will fool some of the people some of the time.. case based AIML reasoning is not the way to go.
I have found a lexiconal DB, like thought treasure, if developed further offers significantly realistic speach.
problem two is in most voice engines, you are going to have to reduce the vocabulary in order to better facilitate the AI. The vocabulary is just too high to accurately capture what a user is saying.
Just a few thoughts.. rambling now... just glad to see others working on the same thing i am.
03/19/2001 (12:25 am)
I too am working in the AI domain at present on voice driven expert systems.There are multiple problems with existing technologies that we are focusing to develop upon and later release as an sdk for games and applications alike.
first and formost are current AI and chatter bot technologies. Most of the current chatterbots i find very plain and too demanding in that they want to control the conversation. While this will fool some of the people some of the time.. case based AIML reasoning is not the way to go.
I have found a lexiconal DB, like thought treasure, if developed further offers significantly realistic speach.
problem two is in most voice engines, you are going to have to reduce the vocabulary in order to better facilitate the AI. The vocabulary is just too high to accurately capture what a user is saying.
Just a few thoughts.. rambling now... just glad to see others working on the same thing i am.
#7
Problem 3 would be the consideration that there currently isn't much use for such an NPC, even if you did take the time to create it. It would simply be (as already has been stated) a talking AI.
An important element in games is the fact that they nearly without exception follow the traditional storytelling method, including games with "open ended" structures. This means that NPCs when talking to you are doing so in order to bring you into the "plot" and lead you from A to B (let us for the moment forget that this can be either Conclusion B, Item B, Emotional State B, or any of the many other possibilities). In essence they are effecting you, and being effected by you.
In an interactive enviroment, this becomes a problem. Here, the story must be generated from three factors. The first factor is basic plot, in essence the fact that you can't allow the player to do -anything-. Limits exist in time and budget, you will probably limit the game to a certain planet, a certain language, a certain mentality, or a certain civilization. The second factor is that of the player's actions. If the player kills the evil Dragon then someone is going to be happy, someone is going to be mad, and the cave is going to be empty. The whole world will have to react, but it will have to react in a way that will keep the player's interest active, most probably in a way similar to the development of a classical "story". The third factor that will effect the development of the story is that of randomness. It simply wouldn't be all that much of a "interactive" world if the same actions always led to the same results, since our own world through it's complexity never allow such simple developments. Through some sort of random structure this complexity must be emulated.
The problem as I see it in a interactive enviroment (as for example an RPG) lies not in the technical issues alone but also in the story. If you consider the hours that have gone into story development of most good games out there, you will realize that it would take a Herculean effort to create a technological "emulation" of the story creating process at any level.
Then to create a subordinate NPC "dialouge" layer that would not only have to be in direct relation to the "story reality" but also the inherant "signifier-signified" relationship in the gameworld in respect to dialouge would not only mean working grammer and ways of expressing yourself, it would have to emulate the very way humans relate to their surroundings.
Of course this might be a slightly overbearing way of looking at the issue. One must not create a whole "virtual reality" just to make realistic dialouge in a game. However, one should be aware of the issues that can crop up, since this is a topic larger then one might think at a quick glance.
03/19/2001 (3:20 am)
Although it's already been said before, and by people much more knowledgable then I...Problem 3 would be the consideration that there currently isn't much use for such an NPC, even if you did take the time to create it. It would simply be (as already has been stated) a talking AI.
An important element in games is the fact that they nearly without exception follow the traditional storytelling method, including games with "open ended" structures. This means that NPCs when talking to you are doing so in order to bring you into the "plot" and lead you from A to B (let us for the moment forget that this can be either Conclusion B, Item B, Emotional State B, or any of the many other possibilities). In essence they are effecting you, and being effected by you.
In an interactive enviroment, this becomes a problem. Here, the story must be generated from three factors. The first factor is basic plot, in essence the fact that you can't allow the player to do -anything-. Limits exist in time and budget, you will probably limit the game to a certain planet, a certain language, a certain mentality, or a certain civilization. The second factor is that of the player's actions. If the player kills the evil Dragon then someone is going to be happy, someone is going to be mad, and the cave is going to be empty. The whole world will have to react, but it will have to react in a way that will keep the player's interest active, most probably in a way similar to the development of a classical "story". The third factor that will effect the development of the story is that of randomness. It simply wouldn't be all that much of a "interactive" world if the same actions always led to the same results, since our own world through it's complexity never allow such simple developments. Through some sort of random structure this complexity must be emulated.
The problem as I see it in a interactive enviroment (as for example an RPG) lies not in the technical issues alone but also in the story. If you consider the hours that have gone into story development of most good games out there, you will realize that it would take a Herculean effort to create a technological "emulation" of the story creating process at any level.
Then to create a subordinate NPC "dialouge" layer that would not only have to be in direct relation to the "story reality" but also the inherant "signifier-signified" relationship in the gameworld in respect to dialouge would not only mean working grammer and ways of expressing yourself, it would have to emulate the very way humans relate to their surroundings.
Of course this might be a slightly overbearing way of looking at the issue. One must not create a whole "virtual reality" just to make realistic dialouge in a game. However, one should be aware of the issues that can crop up, since this is a topic larger then one might think at a quick glance.
M
Don't jump head first into this one.