[UA] Re: UA digest, Vol 1 #601 - 20 msgs

[rowan at media.mit.edu]

> > Actually, what I was thinking about was something
> > like this.  Onscreen, there's a sort of control
> > panel -- I picture it as a circle with red at the
> > top, shading to blue at the bottom.  You use this to
> > pick the emotional inflection of what you're saying.
> >  For instance, if you're at the very top, you're
> > very passionate -- angry or desperate or whatever. 
> > At the bottom, you're very dispassionate -- maybe
> > coldly logical or just blase.  I'd want there to be
> > some meaning on the right -left axis as well --
> > maybe the right indicates that you're friendly while
> > the left is unfriendly.  (That being the case, a
> > square might work better as a controller.  Upper
> > right hand expresses a passionate urge to help the
> > person you're talking to.  Upper left hand expresses
> > anger, rage and hatred.  Lower right is casual
> > conversation.  Lower left is chilly disinterest. 
> > That sort of thing.)  
> > 
> > You'd use the mouse to control the emotional element
> > of the conversation.  With the keyboard, you'd pick
> > topics out of a list.  Whenever you learn something
> > relevant in a conversation, you get a brief summary
> > of the information ("Alex Abel hired the Freak to
> > get a Hand of Glory away from Selena." "Dirk Allen
> > has been having bowel troubles.")  Before you enter
> > a conversation with someone, you can hotkey topics. 
> > If I'm going to go talk to Jeeter, I can hotkey Dirk
> > Allen and Alex Abel and medical problems.  While
> > talking to him, it's quicker for me to bring up
> > those topics, but with a little extra time, I can
> > scroll through all the topics I've already
> > 'acquired.'
> >
> > Computer types?  How feasible is this?  It would
> > still be tree based, but it would seem (on the
> > surface) much more genuine because the emotional
> > content of a statement or question strongly
> > influences the response.  It doesn't require
> > anything like a Turing interface because the player
> > has no way to put in unexpected stuff.

I worked with these kinds of affective interfaces in grad school. The "visual 
emotion display" problem has been kicked around, and some solutions have been 
dumber than others. Nothing great yet. Here's some info from my group:
http://www.media.mit.edu/affect/AC_research/communication.html

> > Ideally, this would all be synchronized with
> > hundreds (hell, probably thousands) of sound files,
> > so you actually hear Jeeter going "Heh, yeah, ol'
> > Dirk can't seem to keep a cool stool."  Then when
> > you say something, you might time it at a natural
> > conversation pause, or you can interrupt -- timing
> > (as in real life conversation) would be important.
> > 
> > -G.

Yeah, sound files are the bottleneck for really user-responsive game storylines. 
Deus Ex is a good example of a game that tries to allow lots of leeway to the 
player, but the voice-acting effort that went into it was phenomenal. Even so, 
the storyline is pretty linear and the conversational choices very limited. You 
won't be able to do anything like what you're envisioning until some kind of 
meaningful text-to-speech system is developed with emotional and inflectional 
markups. Don't hold your breath for that kind of stuff.

I've thought about this problem a lot, and it just isn't really feasible until 
AI makes some big breakthroughs (which may never happen). As far as gaming is 
concerned, other people are going to be your best bet for meaningful 
conversation in-game for a while: I think that massively multiplayer online 
games are the way to go. For a good example of where things are going with AI 
conversation, though, check out Deus Ex... it's an interesting 
storytelling/gaming interface.

-Matt Norwood

_______________________________________________
UA mailing list
UA at lists.uchicago.edu
http://lists.uchicago.edu/mailman/listinfo/ua