Isn''t this the same old problem that happened with robots. People imagined they would be lumbering around with two legs and two arms and an ariel sticking out the top of their heads. But in fact robots are build to do a particular job and not to be an imitation human. Surely its the same with AI. Its there to do a job not to copy the human brain.

In a sense you are correct but when discussing AI in relation to gaming then what people want is something as human as possible. Something that can play a good game and give you an enjoyable experience but one that also makes mistakes(and learns from them) and is reasonably unpredictable. Games players don''t want an unbeatable sterile experience, they want something they can empathise with.

IMO an AI''s job (in most computer games)*is* to mimick a human.





Stimulate

quote: Original post by gaussuk
Isn''t this the same old problem that happened with robots. People imagined they would be lumbering around with two legs and two arms and an ariel sticking out the top of their heads. But in fact robots are build to do a particular job and not to be an imitation human. Surely its the same with AI. Its there to do a job not to copy the human brain.

You have obviously never met ASIMO , have you!? It works as a receptionist at Honda''s Tokyo Headquarters. ASIMO is bipedal, has two arms and generally looks like a human in a space suit... although only 120cm (that''s to make it non-threatening to people... and is also the minimum height to perform many tasks in the world... turning on/off lights, setting tables, etc). ASIMO is not the future of robotics... it is the present! It is extremely smooth and stable when walking and doesn''t have that step-pause(stabilise)-step-pause-step behaviour of other bipedal robots!

Anyway, enough of me... if you''re ever in Tokyo, go visit ASIMO and say ''hi''! You might be surprised at the response you get.

Cheers,

Timkin

quote: Original post by fup
when discussing AI in relation to gaming then what people want is something as human as possible. Something that can play a good game and give you an enjoyable experience but one that also makes mistakes(and learns from them) and is reasonably unpredictable. Games players don''t want an unbeatable sterile experience, they want something they can empathise with.

IMO an AI''s job (in most computer games)*is* to mimick a human.

When I talk to experienced game developers and game players at GDC, and read the comments
of experienced game players on dedicated game forums, and listen to newbie game players at
local LAN parties, and finally talk to computer software store sales people about what they
think their customers want, then I hear a different list of wants than the above.

First, the job of AI in computer games is to provide an entertaining opponent.

An entertaining opponent has nothing to do with mimicing a human. Can you really say that
most people want to play an opponent that acts like most on-line game opponents that are
encountered? You know, the ones that drop out of the game when losing, or curses you
when you win, or allies with you just to help his buddies you are playing against, or rushes
despite an agreement not to, or who plays the same old optimum memorized tactic every
single game? I sure don''t, and I only hear a few developers say that is what customers want.

What I do hear they want, is an adaptive opponent. One that adapts to their playing style
and provides sufficient challenge for them to have to try hard to win, but then in the end
is not too challenging, so that the human wins by a slim margin. That AI is certainly not
like any human player I know. A human player would try to crush you if at all possible.

Making mistakes are relative. The behavior that might appear to be error-prone, could
actually be the right thing to do, and simply perceived as being a mistake. Since the
human player does not have the complete information that the AI opponent is basing
its decisions on, then this is impossible to detect by the human player. IMO, writing an
AI to "try to make mistakes" is a mistake! Computer game AI is sufficiently error prone
to begin with, then (as Yoda says) "there is no [need to] try".

On your next to last point, I will agree. "Games players don''t want an unbeatable sterile
experience". But I disagree with the why. Players are not seeking empathic relationships
with computer AI opponents, they are seeking to beat them. And to enjoy doing it.

Eric

quote: Original post by Geta
First, the job of AI in computer games is to provide an entertaining opponent.

I couldn''t agree more

quote: Original post by Geta
An entertaining opponent has nothing to do with mimicing a human. Can you really say that
most people want to play an opponent that acts like most on-line game opponents that are
encountered? You know, the ones that drop out of the game when losing, or curses you
when you win, or allies with you just to help his buddies you are playing against, or rushes
despite an agreement not to, or who plays the same old optimum memorized tactic every
single game? I sure don''t, and I only hear a few developers say that is what customers want.

It depends on the type of game. In a FPS game then being humanlike is the Holy Grail of AI. Find me a serious FPS player who would say otherwise and I''ll chop my arms off and eat them (I''d need help there of course ;0) In an RTS maybe not. I don''t play much RTS but when I have played them I always wished the opponents were more humanlike.
I don''t think your last few sentences here are called for Eric. They sound patronizing. Of course people don''t want an opponent who acts like that. That doesn''t need saying does it? When I said we want AI''s that play like humans that''s what I meant. That when playing the AI''s gameplay is humanlike. Not anything else.

quote: Original post by Geta
What I do hear they want, is an adaptive opponent. One that adapts to their playing style
and provides sufficient challenge for them to have to try hard to win, but then in the end
is not too challenging, so that the human wins by a slim margin. That AI is certainly not
like any human player I know. A human player would try to crush you if at all possible.

I think people *do* want an opponent they *think* is challenging. Where''s the satisfaction in winning otherwise? A good AI IMO will always make the player think hes being pushed to the edge of his skills to win. Other than that, I''m agreed that adaptive behavior is an essential ingredient to the ultimate gaming experience.

quote: Original post by Geta
Making mistakes are relative. The behavior that might appear to be error-prone, could
actually be the right thing to do, and simply perceived as being a mistake. Since the
human player does not have the complete information that the AI opponent is basing
its decisions on, then this is impossible to detect by the human player. IMO, writing an
AI to "try to make mistakes" is a mistake! Computer game AI is sufficiently error prone
to begin with, then (as Yoda says) "there is no [need to] try".

I never said a programmer should write an AI to deliberately make mistakes. As you say, AI players are mistake prone anyway. But, the AI should be able to learn from its mistakes.

quote: Original post by Geta
On your next to last point, I will agree. "Games players don''t want an unbeatable sterile
experience". But I disagree with the why. Players are not seeking empathic relationships
with computer AI opponents, they are seeking to beat them. And to enjoy doing it.

Eric

I was undoubtedly wrong using the word ''empathise'' in this context because it''s given you the wrong impression of what I think good AI is. Maybe I should have said players want to identify a little with the AI, they want to *believe* in it. If it becomes a real *entity* in their mind the player derives more satisfaction from beating it.






Stimulate

quote: Original post by Geta
First, the job of AI in computer games is to provide an entertaining opponent.

quote: Original post by fup
I couldn''t agree more

Then we agree that humanlike has nothing to do with it?

quote: Original post by Geta
An entertaining opponent has nothing to do with mimicing a human. Can you really say that
most people want to play an opponent that acts like most on-line game opponents that are
encountered? You know, the ones that drop out of the game when losing, or curses you
when you win, or allies with you just to help his buddies you are playing against, or rushes
despite an agreement not to, or who plays the same old optimum memorized tactic every
single game? I sure don''t, and I only hear a few developers say that is what customers want.

quote: Original post by fup
It depends on the type of game. In a FPS game then being humanlike is the Holy Grail of AI. Find me a serious FPS player who would say otherwise and I''ll chop my arms off and eat them (I''d need help there of course ;0) In an RTS maybe not. I don''t play much RTS but when I have played them I always wished the opponents were more humanlike.

Everything depends on the type of game!

In your wish for more "humanlike" are you sure you are not wishing for more adaptibility?

quote: Original post by fup
I don''t think your last few sentences here are called for Eric. They sound patronizing. Of course people don''t want an opponent who acts like that. That doesn''t need saying does it? When I said we want AI''s that play like humans that''s what I meant. That when playing the AI''s gameplay is humanlike. Not anything else.

I was not trying to patronize. I was trying to accurately describe the majority of
the people I have encountered when playing computer games on-line over the years,
and to convey a sense of how I had heard them described by others. The fact is,
alot of humans that play computer games on-line are not much fun to play against.

quote: Original post by Geta
What I do hear they want, is an adaptive opponent. One that adapts to their playing style
and provides sufficient challenge for them to have to try hard to win, but then in the end
is not too challenging, so that the human wins by a slim margin. That AI is certainly not
like any human player I know. A human player would try to crush you if at all possible.

quote: Original post by fup
I think people *do* want an opponent they *think* is challenging. Where''s the satisfaction in winning otherwise? A good AI IMO will always make the player think hes being pushed to the edge of his skills to win. Other than that, I''m agreed that adaptive behavior is an essential ingredient to the ultimate gaming experience.

Ok, have we now agreed on adaptation and entertaining as being more
essential than "human-like"?

quote: Original post by Geta
Making mistakes are relative. The behavior that might appear to be error-prone, could
actually be the right thing to do, and simply perceived as being a mistake. Since the
human player does not have the complete information that the AI opponent is basing
its decisions on, then this is impossible to detect by the human player. IMO, writing an
AI to "try to make mistakes" is a mistake! Computer game AI is sufficiently error prone
to begin with, then (as Yoda says) "there is no [need to] try".

quote: Original post by fup
I never said a programmer should write an AI to deliberately make mistakes. As you say, AI players are mistake prone anyway. But, the AI should be able to learn from its mistakes.

Sorry, I thought you said AI should make mistakes (ie. be made to make mistakes)?
That is different than learning from the mistakes it makes. However, those mistakes,
may not be mistakes. How about if we agree that a computer game AI must learn
to adapt ?

quote: Original post by Geta
On your next to last point, I will agree. "Games players don''t want an unbeatable sterile
experience". But I disagree with the why. Players are not seeking empathic relationships
with computer AI opponents, they are seeking to beat them. And to enjoy doing it.

Eric

quote: Original post by fup
I was undoubtedly wrong using the word ''empathise'' in this context because it''s given you the wrong impression of what I think good AI is. Maybe I should have said players want to identify a little with the AI, they want to *believe* in it. If it becomes a real *entity* in their mind the player derives more satisfaction from beating it.

Many game players tend to attribute "humanlike" qualities to a computer game AI by
filling in the blanks so to speak. They see a behavior, and do not know why it occurred,
and then attribute that behavior as being "humanlike" on their own.

In an FPS I worked on, players used to comment that the ''bots would often flank around
behind them and attack from the rear, while other ''bots pinned them with covering fire.
They thought that was really "humanlike". It wasn''t. It was having a few of the ''bots
find Non-Line-Of-Sight paths to the player and follow them, while others stayed in place.
The Non-LOS paths were paths in which each node was out of LOS of the player, thus
producing a path that flanked around buildings and cover and approached from the rear.

The intent and purpose of that technique was to be entertaining, not "humanlike".

I am not suggesting that we (as computer game AI programmers) forego attempts
to get our artificial constructs to demonstrate such behavior. Instead I am suggesting
that we will be more successful getting the player''s attributions of that quality of
behavior, by concentrating on generating behavior that is adaptive and entertaining
(especially the entertaining part) and not worry about how humanlike it seems. The
players will do that part for us.

Eric

quote: Original post by Geta
I am not suggesting that we (as computer game AI programmers) forego attempts
to get our artificial constructs to demonstrate such behavior. Instead I am suggesting
that we will be more successful getting the player''s attributions of that quality of
behavior, by concentrating on generating behavior that is adaptive and entertaining
(especially the entertaining part) and not worry about how humanlike it seems. The
players will do that part for us.

I think basically we are agreed on the important points Eric. And I was wrong about some of the other ones. Having thought a little more about it, humanlike may probably be a bad idea for singleplayer only games. (My mind has been focussed recently on creating offline bots to entertain players who usually only play online, so my previous comments were ever so slightly biased in that direction ;0))If you had swarms of enemies with that level of intelligence coming after you you''d be *****d good n'' proper. Also, it can be sometimes especially satisfying just to splatter the brains of incredibly stupid foes ala Serious Sam. Bring them on.


So, in short, yes: adaptive and entertaining are the words we should all have at the formost of our minds when developing AI.

ps. How did you get words to come up in bold? (I know not the ways of HTML)





Stimulate

I think one important thing to consider is something that Eric alluded to several posts back...

quote: Excerpted from Geta''s post
What I do hear they want, is an adaptive opponent. One that adapts to their playing style
and provides sufficient challenge for them to have to try hard to win, but then in the end
is not too challenging, so that the human wins by a slim margin.

This is just my opinion, but I sense that game players fall into two categories: a) those that like winning; and b) those that like playing. Many of those in category b) are also in category a), while many in category a) see playing as an environment in which to win and hence prove themselves (worthy?) to their friends, family, on-line friends/foes???

I particularly think that many people in category b) and some in category a) love to be challenged to their limits but ultimately want to win. That''s the source of their enjoyment... knowing that in the end, their mind will triumph over their foes.

No one really wants a stupid game AI. Maybe one thats easy to beat (making the gamer feel more superior) but not really stupid. People generally equate these two but I see them as different. I think easy to beat means "appropriate, yet simple strategy". Stupid means "inappropriate and simple strategy".

Anyway, I''m going on too long...

My point was that ultimately people want to be challenged in a competitive game situation, whether it is in a single player environment, a team environment or multiplayer death-match style. In the end, most people would like to win... or at least not lose every time.

Cheers,

Timkin

quote: Original post by fup

ps. How did you get words to come up in bold? (I know not the ways of HTML)

Bold is done like italics are done except substitute a ''b'' for the ''i''.

Eric

quote: Original post by Timkin
My point was that ultimately people want to be challenged in a competitive game situation, whether it is in a single player environment, a team environment or multiplayer death-match style. In the end, most people would like to win... or at least not lose every time.

Ok, so how do we do that? How does a computer game AI playing on Expert adapt
to you for a particular game that we both know how to play, when you (as an expert) are
a better player than I am (as an expert). In other words, we are both experts at playing
the game, but you are better than me in every game. Thus, how does the AI adapt to the
two of us??

Eric