NetNews Usenet Archive 1993 #3/NN_1993_3.iso/spool/comp/ai/philosop/7415

Internet Message Format | 1993-01-27 | 17.8 KB

Path: sparky!uunet!ukma!usenet.ins.cwru.edu!agate!doc.ic.ac.uk!uknet!bhamcs!axs
From: [e-mail] (Aaron Sloman)
Newsgroups: comp.ai.philosophy
Subject: Re: Colbys-Weizenbaum etc Belated Replies (2)
Message-ID: <[e-mail]>
Date: 28 Jan 93 01:05:18 GMT
Sender: [e-mail]
Organization: School of Computer Science, University of Birmingham, UK
Lines: 370
Nntp-Posting-Host: emotsun

Another belated attempt to catch up on comments on some of my earlier
postings. This has two replies to Jeff Dalton.

> From: [e-mail] (Jeff Dalton)
> Message-ID: <[e-mail]>
> Date: 19 Jan 93 18:16:51 GMT
> Organization: AIAI, University of Edinburgh, Scotland

> In article <[e-mail]> [e-mail] (Aaron Sloman) writes:
> >[e-mail] (Jeff Dalton) writes:
> >> In article <[e-mail]> [e-mail] (Aaron Sloman) writes:
[AS]
> >> >There are some people who see gods in lightning and other natural
> >> >phenomena. We don't bewail the existence of those phenomena: we
> >> >educate people.
[JD]
> >> What would people active in AI and Cog Sci say in this education?
>
> >> Would they say that thermostats are conscious (as dave Chalmers
> >> maintains)?

[AS]
> >I think he's changed his mind.
> >
> >But I would rather warn people against assuming that questions like
> >"Is X conscious" have any clear meaning, with true or false answers.
> >(I've written about this previousl so won't go on now.)

[JD]
> There's certainly a way of looking at them that makes it seem that
> way.  However, whenever I see arguments along those lines, I wonder
> why they're being made.  It doesn't look like that great a problem to
> me.  I'm a realist about consciousness.  That is, I think there's a
> fact of the matter as to whether something is conscious or not
> (although there may be borderline cases).

If you are talking about a person who has been asleep and you are now
asking whether he's awake, then I agree there's a fact of the matter.
If you are talking about someone who has been drugged and are asking
whether he's recovered, then there may be a fact of the matter if it's
the kind of drug whose effects are clear cut (like putting you to
sleep).

But there are lots of cases where its not clear that there's a fact of
the matter. For some nice examples regarding feeling pain read "Why
you can't make a computer that feels pain", in D.C.Dennett's book
Brainstorms. Roughly, I interpret his answer as being: "because the
concept of feeling pain is so incoherent and full of inconsistent
criteria that there's no way of deciding whether or not the computer
feels pain." Of course that doesn't mean that toothache doesn't exist
in humans.

Similarly, I think we can ask whether a chimp is still asleep or has
regained consciousness. More generally, there are many categories of
things of type X such that for individuals of type X there are clear
transitions between being conscious and being unconscious, though for
all of them there may also be unclear cases where our criteria come
apart (like the individual who writhes and screams during the operation,
but immediately afterwards remembers absolutely nothing). Lets call
categories like X "mindful" categories: people are mindful, and I claim
chimps are. I don't have any views regarding amoebas.

When you start asking which CATEGORIES of things are mindful, i.e. which
can be conscious and which can't, as opposed to which things in some
particular mindful category are or aren't conscious, then it is not
clear that there is any fact of the matter. Is a fly conscious? An
amoeba? A two month human embryo? A machine of type so and so?

I often compare this with the following: we can ask whether it's
noon or midnight or 5pm in London, or in New York, or Moscow, because
the question is clear cut. If you point at a bit of the moon and ask
is it noon, or midnight or 5pm there, you may get different answers,
depending on whether you use as your criterion the angular elevation of
the sun above the horizon at that point on the moon, or whether you use
a criterion based on projecting our timezones outwards perpendicular to
the earth's surface, as you probably would for someone flying a plane or
a balloon.

There's NO fact of the matter about which is the right way to decide
what time it is at a particular place on the moon. Similarly there's no
right way to decide how to project our criteria for saying whether
something is conscious or not to other types of animals or machines,
where the criteria fall apart.

> ...There's a problem about
> whether we can know which is the case,

Not when there's nothing to be the case, because the question is
muddled, incoherent, etc.

> and there's a problem of making
> it clear just what we're talking about by using the word "conscious"
> (especially since it has a number of different meanings).  But it
> seems to be that our aim ought to be to solve these problems as best
> we can rather to argue that we cannot.

Yes, I think we ought to solve the problems. Here's how: we first
produce a good theoretical overview of the types of designs that can
produce various kinds of behavioural capabilities more or less like
ours. We then analyse the different kinds of capabilities that are
capable of being generated by different kinds of mechanisms covered by
the overview. We then give technical names for these different sorts of
capabilities (and states, and processes, etc.). We could then CHOOSE one
of these as closest to the ordinary use of the word "conscious" and use
that as our new technical definition of the word. It would then be a
fact of the matter whether amoebae, flies, mice, chimps, etc. etc. were
or were not conscious.

Compare: we used to have words for different kinds of stuff, but they
had no clear, precise, definitions to cover all cases. The development
of physics and chemistry provided a good generative basis for a taxonomy
of kinds of stuff (chemical elements of various sorts, isotopes,
chemical compounds, mixtures, solutions, alloys, etc. etc.). Within this
array of kinds generated by the new science, some kinds came very close
to previous non-technical uses of words like "water" (H20), "salt"
(NaCl), "carbon", "iron", etc.

It then became a fact of the matter whether something was water or not,
or water plus impurities, etc., whereas previously there were ill
defined boundaries.

Of course, someone who thinks he knows what consciousness "REALLY" is
from first hand introspective experience, will reject all this. What
I've found is that that sort of tendency is very-wide spread, deep
rooted, and hard to argue against. In fact it's very like a religious
conviction in some cases. Philosophical therapy over a long period
sometimes cures such people. But only sometimes. (It took me a long time
to get rid that delusion.)

[JD]
> Earlier you wrote:
(as eliza might have said :->)
[AS]
>   People who use the words "conscious", "consciousness" in technical
>   discussions, generally *think* they know what they are talking
>   about, but are normally totally muddled, between all sorts of
>   different interpretations sometimes mixed up with mumbo jumbo
>   concepts to do with magical inaccessible entities and processes.

Well, I've tried to unravel some of that, in my remarks above.

>
> My impression is that people generally do know what they're talking
> about, are not totally muddled,

Well it's hard to be sure without discussing with them at length, and
asking them to define precisely what the question is that they are
asking about fleas, mice, embryos, or machines, when they ask whether
these are capable of being conscious (i.e. are "mindful" categories.)
I don't dispute that they understand the ordinary use of English
phrases like "He's regaining consciousness now".

Also the fact that people *claim* to know what they are talking about,
or *appear* to know what they are talking about, doesn't mean that they
*do* know what they are talking about. There are lots of cases where
people don't discover till long afterwards, or sometimes never at all,
that they have been using incoherent concepts. I think Aristotle thought
he knew what he was talking about when he wondered where the "natural"
place was for earth, water, fire, air, etc. Before Einstein some people
may have thought they understood the concept of continuing identity of a
spatial location (as opposed to a continuing set of spatial relations to
other things).

We are not authorities on whether we attach definite meanings to our
words, for we easily deceive ourselves.

> ...and don't mix in mumbo jumbo
> concepts or magical entities and processes.  (I'm leaving out
> "inaccessible" here.)  Of course, when there is muddle and magic,
> it's worth pointing this out.  But I'm very suspicious of this
> sort of general accusation, divorced from all concrete instances.

Point taken. I took various verbal cues in previous contributions as
evidence, because they looked like the turns of phrase I've often heard
used by people who I had talked to in more depth, and (in my view) found
to be muddled, etc. and often unwittingly committed to the existence of
some kind of special stuff divorced from the realm of scientific
investigation, inacessible from "outside", etc.etc. and somehow
(magically?) capable of producing events in the physical world.

Even distinguished scientists can fall into this sort of trap (which was
my conclusion after reading the well known book on consciousness and the
self by Eccles and Popper (whose exact title I forget).

[AS]
> >I'll backtrack a little: if the vendors know there's a good chance
> >that because of inadequacies in the current educational system users
> >of a package are likely to be fooled into attributing capabilities
> >to the machine which it really does not have, then perhaps the
> >machine should from time to time print out a suitable warning

[JD]
> Humm.  I think there's a chance that education in AI will lead
> people to attribute capabilities to the machine which it really does
> not have.  I gave an example a while back of a class in Edinburgh
> where students argued, with the support of the instructor, that
> Eliza has some real understanding, although only a small amount of
> it.  (Ie, that the difference between Eliza and, say, a human
> was essentially only a matter of degree.)

[AS]
A tendency to believe that understanding is all-or-nothing (and
similarly for consciousness, etc. etc.) often goes with the sort of
muddle I was talking about. If a detailed analysis of human
understanding of language shows that it involves 173 different
capabilities, and if Eliza has 4 of them, well then I'd say there's
nothing wrong with saying that it has a small degree of understanding.
(I've enlarged on this in my paper on understanding in IJCAI-85)

Saying it's "only" a matter of degree may imply that there's a continuum
of cases. If that's what you are objecting to, I agree. Design space is
full of DIScontinuities, and some of them make a very large difference
to the capabilities of instances of designs.

But arguing about which subset of capabilities is really required for
understanding (or even a "teeny bit of understanding") is like arguing
over which criterion is the real one for deciding on the time on the
moon. It's just a silly argument as there is no answer.

[JD]
> I think there's a very strong tendency to see some real understanding
> behind any use of natural language.

That's because there usually is: e.g. in the case of the computer
program there's the understanding of the programmers that lies behind
what the computer prints out. If you are tempted to argue about whether
the machine *itself* has "real" understanding as opposed to "fake"
understanding, or "counterfeit" understanding, or "primitive"
understanding, or "partial" understanding, etc. then I think it's a
pointless argument.

Of course there's a different point, namely that people can be deceived
by appearances, just as people have been deceived into thinking the
earth is flat, the sun goes round the earth, the gods are punishing
them, etc. etc. But I didn't think you were talking about that.

> ..We can see the other side of this
> when we consider how hard it often was for people to see how a
> computer could use symbols that are meaningful to us rather than being
> confined to numbers.  For instance, there are people in offices
> everywhere who have had to divide all recognized work activities into
> numerically designated categories and who accepted the explanation
> that they had to do this because the computer required it.

This is beyond my experience. There are, I admit, lots of people who use
numbers gratuitously instead of informative descriptions, e.g. many
would-be scientists (e.g. psychologists who suffer from physics-envy).
I don't see how teaching people about AI will confuse them about this.

> Education ought to be able to do something about this, but I think
> that education would have to be somewhat "anti-AI" (and anti-TT), at
> least as the debate is currently conducted.

Strange to read that coming from you, given that you, of all people,
know full well that computers are capable of doing all kinds of
non-numerical things, and that AI in particular has all sorts of
non-numerical techniques. But I expect I've misinterpreted you.

Now your second article:
> From: [e-mail] (Jeff Dalton)
> Subject: Re: Colbys-Weizenbaum and Overcoming Depression 2.0
> Message-ID: <[e-mail]>
> Date: 19 Jan 93 19:20:36 GMT

> In article <[e-mail]> [e-mail] (Aaron Sloman) writes:
> >[e-mail] (Angus H Rodgers) writes:
> >>
> >> I'll change the question, then, to one which you are less likely
> >> to find ambiguous: would it be necessary for the mechanical
> >> psychotherapist to be able to *feel* anything? (With apologies to
> >> Roger Penrose, and Hans Christian Andersen.)
[AS]
> >Well, the English word "feel" is almost as bad as the word
> >"conscious". Are you asking whether the machine would need to be
> >able to
> >    feel the surface it is resting on
> >    feel the breeze
> >    feel itself falling
> >    feel how far it is from the wall
> >    feel the roughness of the wall
> >    feel that it is getting close to finding an answer to a question
> >    feel that it is getting hotter
> >    feel hot
> >    feel tired
> >    feel that the weather is likely to turn wet
> >    feel sure that Fermat's last theorem is provable
> >    feel angry      (compare "be angry", which is quite different)
> >    feel depressed  (compare "be depressed")
> >    feel inclined to refuse to answer
> >    feel unwilling to give up

[JD]
> This looks to me like it might be a solvable problem.

Yes - that's my point. For a subset of very specific cases at least we
can begin to specify criteria that might be used for discriminating
those cases.

But then someone comes along and says: "Ah it may feel in that sense,
but does it REALLY feel?" And then you probe a bit and find that they
can't define what they mean except by saying "Well YOU can feel things
can't you? so you know what feeling is from your own introspective
experience", etc. and then the argument gets bogged down because they
don't appreciate that you cannot define anything by pointing at an
instance, even pointing inwardly. (Compare consciousness.)

[JD]
> BTW, doesn't Dennett have a paper on whether a robot can feel pain
> (in _Brainstorms_)?  Does he manage to pin "feel pain" down well enough?

Yes the paper is relevant -- referred to above. My interpretation is
that he shows how incoherent the concept is as soon as you move beyond
the "standard" situations in which we learn to use the word "pain".

[AS]
> >Such questions are often not so much a healthy expression of
> >intellectual curiosity as an attempt to use rhetorical questions to
> >challenge what is felt as a threat of some kind (e.g. Weizenbaum's
> >threat that taking AI seriously somehow degrades mankind).

[JD]
> I think this is a rather dubious tactic, Aaron.  You don't come
> right out and say "you're just using rhetorical questions against
> what you see as a threat" (perhaps because you don't think it's so),
> but you try to create the impression that there's something suspect
> about "such questions".

I wrote "often". I don't know what proportion of cases need this
diagnosis. It's certainly true of very many people I have argued
with in real life. It's hard to be sure from typed contributions
in comp.ai.philosophy! So perhaps I over-react to some of the familiar
sounding turns of phrase and rhetorical questions.

[AS]
> >I was referring in a rather sloppy fashion to an old philosophical
> >notion -- some people think that the only way you can possibly know
> >what "consciousness" is, is by having it, and that's because the
> >relevant states and processes are assumed not to be part of the
> >physical universe but somehow part of a different (spiritual,
> >mental) realm of being, not accessible via physical perception or
> >measurement, but only via a kind of introspection.
> >
> >Often such thinking implies a notion of a kind of magic whereby
> >these non-physical mental or spiritual entities can cause
> >occurrences in the physical world or possibly vice versa.
[JD]
> Tar them with dualism, eh?  I don't think this sort of thinking is
> actually very prevalent on the net (at least in AI newsgroups).  At
> least I haven't noticed anyone defending substance dualism (although
> Dave Chalmers defended property dualism).

Dualism of a sort often lurks unacknowledged beneath the surface of
hostility to AI.
(Though not in the case of Searle or Penrose.)

Scepticism about AI is different from hostility: I am deeply agnostic
about what AI can or cannot achieve. But I think the best way to find
out is to engage in AI research with an open mind. Currently I don't
think we even understand what the problems are, let alone whether AI can
solve them.

But doing AI is the best way to learn.

Aaron
--
--
Aaron Sloman,
School of Computer Science, The University of Birmingham, B15 2TT, England
EMAIL   [e-mail]  OR [e-mail]
Phone: [phone]       Fax:   [phone]