Fun_People Archive
7 Feb
The language of AI

Date: Tue,  7 Feb 95 13:55:05 PST
From: Peter Langston <psl>
To: Fun_People
Subject: The language of AI

[One of the problems that the AI community encountered early on was that of
simplistic but over-optimistic naming; routines called PickBestMove(),
MakeGlobalWeatherForecast(), and  UnderstandSpokenGreek() somehow seemed much
more powerful than they turned out to be.  On the other hand, accurate names
would be awakward and, after all, they're just symbols...  This Scientific
American article suggests a solution to one such recurrent situation...  -psl]

Forwarded-by: bostic@CS.Berkeley.EDU (Keith Bostic)
Forwarded-by: Wendell Craig Baker <>
From: INNOVATION 2/6/95

Researchers at M.I.T.'s Media Lab are developing a menagerie of intelligent
agents, designed to roam around computer networks and make themselves
useful. These virtual critters (they're referred to as a virtual dog, a
virtual hamster, etc., depending on function) can perform simple tasks,
such as sorting e-mail and scheduling meetings. The software program
observes how its owner sorts e-mail, and after detecting a pattern, offers
to finish the task. There are a couple of drawbacks -- while the program
can learn to mimic its owner's actions, it is unable to learn skills that
the owner doesn't already possess. Also, consumers tend to consistently
overestimate the intelligence of computer programs, particularly those that
use language. One solution? Every time you see the words "intelligent
agent," substitute "trainable ant." (Scientific American Feb.'95 p.28)

[=] © 1995 Peter Langston []