I wasn't talking about just picking words from a list. I agree with you
that's unrealistic. My idea was to have certain categories of vocabulary and
strategy knowledge for the bot which the player could set with a number
between 1 and 100 and tweak to their liking. Then before the bot made a play
it could look at the numerical strength for the bot in relevant areas and
through some formula figure out if the bot would actually "know" play given
it's assigned strengths in relevant categories. If not it would do the same
for the next play down on the list  ETC.

 

 

Mike

 

From: [email protected] [mailto:[EMAIL PROTECTED] On Behalf Of
Martin DeMello
Sent: Wednesday, September 05, 2007 1:07 AM
To: [email protected]
Subject: Re: [wgp] Re: [quackle] turning Quackle down

 

On 9/5/07, Michael Quinn <[EMAIL PROTECTED] <mailto:onyx%40vianet.ca> > wrote:
> and fun opponent to play. I don't see how you can get around the fact that
you have to
> stop the computer from making the optimal play every turn so it won't
really be "trying its
> best"

It's the difference between finding the n top plays and choosing
something a bit down the list (very naive way of dumbing down; feels
unrealistic because sometimes the optimal play is obvious to the
1400est of players) and first altering the vocabulary and base
strategies, then picking the top play given those strategies. The
latter should mimic a weaker human player much better, and has the
advantage of "learning" as it goes along, by dint of reintroducing
words and algorithms to ramp up the skill level.

martin

 

Reply via email to