From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!utgpu!cs.utexas.edu!sun-barr!olivea!uunet!psinntp!scylla!daryl Wed Feb 26 12:54:12 EST 1992
Article 3981 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!utgpu!cs.utexas.edu!sun-barr!olivea!uunet!psinntp!scylla!daryl
>From: daryl@oracorp.com
Newsgroups: comp.ai.philosophy
Subject: Re: Look-up tables
Message-ID: <1992Feb24.174823.15344@oracorp.com>
Date: 24 Feb 92 17:48:23 GMT
Organization: ORA Corporation
Lines: 29

Neil Rickert writes:

>  Case 2:  The table is prepared in advance, and deals with all possible
>       contingencies.  That means it has to have responses about
>       Jesse Jackson being elected President, about Arafat being
>       elected Prime Minister of Israel, etc.  When the actual test
>       is given what actually happened will be known, so some response
>       by the interviewee will be patently ridiculuous, even though
>       that could not have been known when the table was prepared.
>       A competent interviewer can certainly show up the robot.

Where were you when we went over all this earlier?  Anyway, you have
to assume that the robot is provided with the same interface to the
rest of the world that a human would have. A fair comparison would be:
keep both the robot and the human confined to a sound-proofed room for
the duration of the test. Any information about the outside world must
come via the teletype provided.

Alternatively, (this was suggested to me in e-mail by Hans Moravec)
you could imagine equipping the robot with the full range of human
sensory inputs. In that case, the table would have to store not
conversations, but sequences of lower-level inputs and outputs. Since
human sense organs can only process a finite number of bits per
second, it still would suffice to have a finite table to describe the
appropriate response to every conceivable input sequence.

Daryl McCullough
ORA Corp.
Ithaca, NY


