Newsgroups: comp.ai.games
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!zombie.ncsc.mil!news.duke.edu!news-feed-1.peachnet.edu!gatech!purdue!news.cs.indiana.edu!babblefish.cs.indiana.edu!jla
From: jla@cs.indiana.edu (jason almeter)
Subject: Re: Message to All AI programmers
In-Reply-To: scott@santafe.edu's message of 16 Mar 1995 16:52:05 GMT
Message-ID: <JLA.95Mar20163255@babblefish.cs.indiana.edu>
Organization: Indiana University Computer Science
References: <3k2pj7$gib@ccshst05.cs.uoguelph.ca> <3k9f09$hsc@serra.unipi.it>
	<SCOTT.95Mar16095205@sfi.santafe.edu>
Date: 20 Mar 1995 21:32:55 GMT
Lines: 24

>>>>> "Scott" == Scott D Yelich <scott@santafe.edu> writes:

[snip]

 Scott> Personally, I like playing humans because I know that there
 Scott> will always be someone who can be better then I am.... on the
 Scott> other hand, computer players come in two classes: (1) those
 Scott> which just don't get any better and at their best are too easy
 Scott> to play against and (2) those that get so good or are part of
 Scott> a game that expects too much from a human that they're
 Scott> impossible to defeat.

 Scott> I don't like either... so I prefer to play human players.

It seems like one would want the game to adapt to the player, but not
at a 'perfect' level.  Instead of tweaking production rates (or
back-prop'ing a neural net, or whatever you do to adjust difficulty in
your game) to crush the human, you should adjust it to some win
percentage...  I.e. at difficulty level 1 the AI tries for a 10% win
ratio and at level 9 it tries for 90%.  The human can set the
frustration, err difficulty, level they want.

-jason

