Re: Machine learning for a conversational system

> Once
> you've finished your beta testing phase and release the game, the machine
has
> learned all it's going to learn, and the resulting behaviour from the
player's
> perspective isn't any different from T3's OOP approach.  Does it make any
real

Correct, the system wouldn't be able to learn after beta. (Unless it can
continually provide feedback to the author, such as in a MMORPG.) Learning
from the user's inputs alone (without human intervention)  would require
unsupervised machine learning, which as you discussed later, is very
dangerous, especially if the user wants to mess things up. (Even speech
recognition systems are very cautious about this.)

As far as making any real difference: I think that given the small sope of
contemporary NPC conversations it doesn't make any sense at all to include
machine learning. I was just thinking out 10-20 years. 10 years isn't so far
out, because if I decide to unterake a graphical IF system I want an
architecture that will last awhile. I haven't yet figured out how a
machine-learned convesation system would affect the architecture, though.

My thoughts proceed as follows:
1) Back in the mid 1980's I wrote several adventure games in basic or
pascal. They worked ok.

2) When I looked over TADS/Inform, I noticed how nicely the OOP language
correlates to the rooms, game objects, and parsing. While I was able to
write an aventure without using OOP, it would have been much easier with
OOP.

3) If I think that machine learning is the way to go, then what kind of
language would facilitate both the machine learning and pattern matching?
While OOP will work (just as basic/pascal would), there might be something
better. I'm not sure what this is, but it's worth a ponder.


-- 

Mike Rozak
www.mXac.com.au