Path: utzoo!utgpu!water!watmath!isishq!doug
From: doug@isishq.UUCP (Doug Thompson)
Newsgroups: comp.society.futures
Subject: Re: The future of AI
Message-ID: <36.22625115@isishq.UUCP>
Date: 12 Apr 88 16:05:55 GMT
Organization: FidoNet node 221/162 - ISIS International, Waterloo ON
Lines: 150
rwojcik@bcsaic.UUCP (Rick Wojcik) writes:
RJ> Weisenbaum's defection is even better known, and his Eliza program
RJ> is
RJ> cited (but not quoted :-) in every AI textbook too. Winograd
RJ> took us a
RJ> quantum leap beyond Weisenbaum. Let's hope that there will be
RJ> people to take
RJ> us a quantum leap beyond Winograd. But if our generation lacks
RJ> the will
RJ> to tackle the problems, you can be sure that the problems will
RJ> wait
RJ> around for some other generation. They won't get solved by pessimists.
Weisenbaum's problem with AI does not fairly translate into "pessimism"
or a lack of "will". Sure, he points out that many forecasts of
breakthroughts by AI types just haven't been realized. His complaint, in
the book "Computer Power and Human Reason" seems much more about the
mentality of some AI workers, and the way problems are defined.
He disputes, for instance, the populist scientific view of man, that
homo-sapiens is just a complex machine (behaviourism). Machines depend
on "effective procedures" -- logical if-then and cause-effect
relationships. Artificial intelligence, so long as it is based on the
kind of machines we know about today, must also be based on "effective
procedures". Human decision-making, according to Weisenbaum, makes use
of "effective-procedures" at times, but is not confined to it. Human
decision-making usually involves two or more people talking about a
problem and coming to understand its meaning to them, as distinct
persons, in an historical situation. This suggests avenues for solution.
The way we talk to each other, the way we understand meaning -- this has
much to do with the experience of being a human being, and being treated
like a human being by other human beings.
How are you going to mechanize that? Either the machine will understood
pre-defined meanings "programmed" into it, or it will develop its own
meanings based on its own experience. The latter is far beyond current
capabilities, and the former is -- well -- relatively trivial. All you
end up with is a decision-making loop which is only successful it can
take account of *all* possible input and permutations. It is basically
no different than the rules of a welfare bureaucracy, for instance. The
raw data is the applicant, the applicant is examined according to
certain pre-defined criteria, and the bureaucracy decides to pay or not
pay. We all know this is unfair to some because people who don't really
need help get it, and some who really do need help don't. The real
humans applying for welfare don't always fit the pre-defined categories.
You could mechanize that process though, because it is based on clear
rules that are expressed as effective procedures. You might even call it
intelligence, but it is still not going to replace the human appeal
committee that can look at what the "machine" or the "bureaucracy"
decided and over-rule it when an exceptional case arises.
This is just one of the problems for which AI has, in Weisenbaum's
argument, no theoretical solution. People deal with new situations with
creativity, often through such things as empathy, based on their
experience of being human, and what it means to be human. Can you make a
machine think it is a human and think like a human?
Well, there are those who say you can -- there are those who say you
can't and Weisenbaum is saying you probably can't but you most certainly
*should not*.
Such a machine, if it worked, could end up imposing its creator's sense
of meaning, mostly frozen in time, on everyone subject to such a
machine. It could generate its own sense of meaning and run the world
according to what it -- not its creators or subjects -- thought was
important. Already there is lots of evidence that the limited
instruments we have today are doing this.
Ultimately the perfect AI machine would behave exactly as a human might,
and have all the capabilities that a human has. One seriously has to ask
why one would want to make a mechanical man for trillions of dollars
when we can get billions of flesh-and-blood men for pennies per hour?
I think we have here "optimism" based on a combination of ancient dreams
of "perfect slaves" and "supermen". Of course, we presume that we will
be able to control these machines once we have built them. That is the
romatic misconception. We build bureaucracies we cannot control,
institutions we cannot control, we have all written computer programs or
parts of programs that no one can control -- or properly understand.
I very much identify with Weisenbaum's basic question: "Why on earth
would we want such machines?" "What possible *good* could they do for
us?" The debate is partly technical, but mostly philosophic: grated for
the moment that you may eventually build such a thing (which is quite
doubtful) what would you use it for?
The answer, if you look about, is to make more effective military
weapons. Or at least this represents the majority of answers being found
today. Very little AI work is being directed toward reconciling human
differences, resolving world problems, feeding the poor, or bringing
justice to the oppressed. Very much AI work is being directed toward
increasing the capacity of some men, in possession of these instruments,
to control other men.
Since most AI advocacy is rooted in a behaviouristic understanding of
mankind, it is not surprising that instruments to modify behaviour
comprise most of what is being produced -- or researched. At best
though, we could "artificialize" only a very small and specific portion
of human "intelligence" by pursuing this path. Theologians define
idolatry as the worship of a sub-set of human attributes at the expense
of others, leaving an unbalanced, distorted result.
This is precisely Weisenbaum's complaint against AI. He doesn't say
there is not good to be achieved down this road, he does say that the
approach being taken in this culture is very unbalanced, and unhealthy
and therefore the net effect is negative.
His plea is not that we stop work on AI, but that we approach it in a
more balanced and wholistic way with a more civilized list of human
priorities such that the machines we create serve to benifit mankind,
and not make life more tenuous and intolerable.
At the moment, this is largely impossible for a wide variety of
political reasons. Weisenbaum was not prepared to work on AI devices to
enhance the kill power of military equipment, nor was he prepared to
work on machines to mechanize psychotherapy and remove the human doctor
from the treatment of humans. AI apologists generally are quite prepared
to work on such projects, and even hail them as great progress for the
human race. To Weisenbaum -- and me -- such things are Frankenstinian
obscenities which can only degrade human life.
RJ> Henry Ford had a good way of putting it: "If you believe you
RJ> can, or if
RJ> you believe you can't, you're right."
Well, I'm not gonna knock the power of "positive thinking" -- but Hitler
believed he could . . .
In addition to the question of "do you believe?" we must ask "in what do
you believe?" before deciding to help you or put a stop to you.
------------------------------------------------------------------------
Fido 1:221/162 -- 1:221/0 280 Phillip St.,
UUCP: !watmath!isishq!doug Unit B-3-11
Waterloo, Ontario
Bitnet: fido@water Canada N2L 3X1
Internet: doug@isishq.math.waterloo.edu (519) 746-5022
------------------------------------------------------------------------
---
* Origin: ISIS International H.Q. (II) (Opus 1:221/162)
SEEN-BY: 221/162