Re: AI in FT (was Re: Be gentle...)

On Sun, 13 Jul 1997, Sutherland wrote:

> The AI will almost always win the first encounter/engagement against
human
> opponents.  The problem comes when there are engagements after the
first.
> The AI has no imagination and therefore has a hard time predicting
what a
> human will learn from the first encounter.  The win/loss ratio gets
worse

You know, as someone that actually works with AI in the real world, I'm
going to have to take issue with the implication that its incapable of
learning new responses or even creativity.  The application of neural
networks, genetic algorithms and other modern techniques allow software
to
come up with new, even novel solutions to situational environments.
Whether the learning occur as a result of neural reinforcement paths or
because the less successful code dies and the winners survive to breed
variants, experience gets encoded into the matrix of rules the software
uses to make decisions.

Given, what, a thousand or more years of development, AI that can
interact
meaningfully in human terms should find learning things based on that
patterning near-trivial.  This won't require recoding it after every
encounter or any such manual intervention, it'll happen as a natural
consequence of the way the system generates its responses.

For a paper on a realworld application of something you'll have some
connection to, do an Infoseek search for 'Amalthea MIT'.  The Amalthea
project used a GA to learn your preferences in mail filtering.

-- 
[  Alexander Williams {thantos@alf.dec.com/zander@photobooks.com}  ]
[ Alexandrvs Vrai,  Prefect 8,000,000th Experimental Strike Legion ]
[	     BELLATORES INQVIETI --- Restless Warriors		   ]
====================================================================
 "Religiones antiquae et arma ridiculae non comparant cum bono telo
 eruptionis igneae latero te, puer." ["Ancient religions and hokey
    weapons are no match for a good blaster at your side, kid"]
						    --- Hanno Solare