[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [xmca] Message in a Bottle


I think that if the detractors, supporters and users of AI all began to just think of things like chatbots as what you call 'sophisticated objects' or perhaps what we might call mediating artefacts, we would avoid the sort of ever repeated argument in which claims are made for them being human like which are then repudiated by others, both referring to abstract models of what it is to be human. Users would then have a better understanding of what they can or cannot do and those who see them as teaching assistants would have fewer legs to stand on. (Philosophers might lose an area of debate, though.)

In this respect, though I could agree with a lot of his argument, I don't think Friesen's article gets us much further towards understanding chatbots as artefacts - though to be fair that probably wasn't what he was trying to do. The example of dialogue he gave just makes the rather obvious point that they cannot converse like humans because they only have a restricted domain of operation. There are a lot more interesting questions about what might be needed to make chatbots more *useful* and what their potential and limitations are in this respect.

I also thought of ELIZA after reading the paper. The gullibility of its users led Weizenbaum, its creator, to give up work in AI and instead become its critic. Maybe it only worked because Rogerian therapy is restricted and stereotyped too inits responses so that he'd picked a domain in which it was relatively easy to create something that could appear intelligent.

As someone who did write programs in this area in the 80s, I don't know if I had a wicked sense of humour, but I did learn that users can be remarkably cruel, always looking for something that would cause the program to crash or give a silly answer...


----- Original Message ----- From: "Steve Gabosch" <stevegabosch@me.com>
To: "eXtended Mind, Culture, Activity" <xmca@weber.ucsd.edu>
Sent: Wednesday, June 03, 2009 11:54 AM
Subject: Re: [xmca] Message in a Bottle

Your thought on chatbots copied here has had me thinking a little,

On May 26, 2009, at 6:49 PM, David Kellogg wrote:

[For voluntary communication to take place] ... there has to be
exactly what is missing when a human pretends to communicate with a
chatbot ... : there has to be a theory of reciprocal willingness to
communicate based on the assumption that the other is a subject like
oneself. That is the key distinction between subject-subject
relations and subject-object relations that I think Leontiev ignored.

The chatbot example is a very good one to make your point.  As long as
you play along and act as though (or perhaps even believe) that the
computer program behind a chatbot represents a reciprocal willingness
to communicate as a real person, you can keep up a real dialogue.

I remember a few years ago playing with the Eliza program, a chatbot
developed in the 1960's that is alive and well on the internet.  This
automated Rogerian-style therapist asks things like "how do you feel
about that?"  It repeats things you say back in question formats that
are designed to elicit you to talk more about yourself.  As long as
you play along, it works surprisingly well, especially if you don't
try to give it trick questions.  Doing this is an application of that
subjective thing we so often do in the movies, the "suspension of
disbelief."  At first, one may feel inclined give the chatbot the
benefit of the doubt, and actually try to seriously talk to it.  This
kind of dialogue could even be a little therapeutic!  Maybe you could
use a few moments to describe how you feel about something ...

But as soon as you become exasperated with your interlocuter being
just a computer program, the communication breaks down.  And what
happens next is just what you suggest: you no longer communicate as
though there is reciprocal willingness from a fellow subject.  You now
talk only as though you are speaking with a sophisticated object.  You
may even get the impulse to devise ways to trick it into acting like
the dumb machine you know it really is!  That is when you may discover
that programmers can have a wicked sense of humor about these things ...

Your generalization about Leontiev makes me want to read where he
spoke about subject-subject relations.  Given the general, mediational
character of human activity, I am wondering, from a CHAT framework,
what a "subject-subject" relation actually is.  Isn't culture
(objects, artifacts, words, bodies, etc. etc.) always in the middle?

- Steve

xmca mailing list


No virus found in this incoming message.
Checked by AVG - www.avg.com
Version: 8.5.339 / Virus Database: 270.12.51/2151 - Release Date: 06/02/09 17:53:00

xmca mailing list