Turing, AI and ESP (1/2)

[Seb Thirlway]

From: jporter <jp4321 at idt.net>


[snip]
>>The thing that strikes me most obviously in this argument is
2.,
("machines can't have be capable of ESP")
>>which seems to be a completely unwarranted assumption.   OK it
>>does seem plausible, if we all imagine for a moment that we do
>>believe in ESP, that a machine would not have this faculty.
>>However, replace "has an ESP faculty" with "can enjoy
strawberry
>>ice-cream", "can fall in love", or any of the other standard
>>attributes of humans in this type of argument, and then the
>>argument seems quite lame - in fact, an argument addressed and
>>disposed of elsewhere in the article.
>
>You cannot so simply replace "has an E.S.P. faculty" with "can
enjoy
>strawberries and cream-" which is a static datum. The machine
could be
>programmed to respond randomly to "food" questions- either like
or dislike
>to varying degrees- and store the reply in memory, so as not to
contradict
>itself in case it were asked again. E.S.P., apparently, is
dynamic and
>synchronous and cannot be randomly faked.


Agree completely that ESP and liking types of food are in
different categories.  I can imagine (as you have) a way of
programming a machine to simulate all kinds of "human" behaviour,
but I can't imagine a way to program it to perform ESP: I don't
know how to perform ESP myself, and the impression I get is that
if I did I still wouldn't understand _how_ I did it - in such a
way that I could program a machine to do the same.
I think what I meant to get at is that the ESP argument is of the
same form as arguments about other more common human
characteristics: it attempts to draw a line round a set of
behaviours/capabilities and say "to be capable of these is to be
human - and this behaviour or that one couldn't be simulated by a
machine".  Turing shows elsewhere in the article that many of
these behaviours in the "human" set _could_ (theoretically, for
him, but quite conceivably, for us now) be simulated by a
machine.  ESP is the sticking point for him.
The straw man I was attempting to demolish was an argument
something like

"ESP is the sticking point because in ESP Turing has identified
THE identifying characteristic of human as opposed to artificial
intelligences, or at least a major identifying characteristic.
(sigh of relief, peasants extinguish burning torches, put down
pitchforks and disperse down the pub)."

and I agree that in the Turing game, ESP would probably be a
problem for a machine, assuming that humans _do_ have an ESP
faculty.  My point was that this doesn't make ESP mysterious or
special (any more mysterious than it already is).  ESP is only a
sticking point because it requires the machine to do more than
say things about _itself_.  The nice thing about the Turing game
is how it shows that there's an assumption that a "mind" is
operating whenever anyone(thing?) says something to us: this
"mind" assumption carries with it the notions of privileged and
incorrigible access, and a machine can use this weighty
assumption against us, to fool us that it is human.  ESP would be
a problem for the machine playing the game, but not for the same
reasons as many of the other problems, which are all to do with
making easily-simulable statements and relying on the observer's
"mind" assumption to do the rest of the work.
For example, the machine could be programmed to wait until a
realistically human level of intimacy had been reached with the
observer, and then start complaining about a "love-life" it had,
asking for advice etc.  The observer's assumptions about entities
that talk in an articulate fashion would fill in all the blanks
for the machine, imagining a real romantic partner, feelings etc.
All this from a few programmable phrases!

My argument is that ESP only presents difficulties because it's
not like this, it's not a matter of saying whatever the hell you
like with no fear of contradiction.  Answering ESP questions is
not incorrigible talk about your own mind.  But then a typical
machine could just as well be knocked out of Turing's game by
failing to respond well to "Could you go down the shops and see
if Camels are still at reduced price?" - because the designers
had neglected to give it legs or wheels!  Sure, the machine could
be programmed to respond cleverly to this sort of thing (e.g.
"I'm recovering from a stroke and can't walk without
assistance"), but then it could respond just as well to ESP
questions (e.g. "I've never shown any psychic abilities").
ESP is only a sticking point because answering ESP questions is
like _really_ going down to the shops and reporting the price of
Camels: it's making a corrigible statement: in contrast to a
statement like "I feel down today", which would make the observer
nervous of responding "no you don't, you're just a machine
programmed to say that" - nervous because it just _might_ be a
human saying that, and breaking the convention of treating
others' statements about their own state as privileged and
incorrigible is rude and hurtful.  Breaking that convention is
almost equivalent to ceasing to treat the other as human.  It's
this convention that the machine in the Turing game uses against
the observer.
So my argument is that ESP is only a problem for Turing because
it's requiring the machine to play an additional game that humans
can play: over and above the "incorrigible reports of state"
game, humans can also play the "move about the world and report
on it" game.  That's what I meant about ESP being outside the
rules of the Turing game: I put this (or thought this) badly -
it's not really outside the rules, since Turing defined the goal
open-endedly as "passing for a human", but it is addressing a
completely different aspect of being human from the other ones
considered in the article.
Getting the machine to answer ESP questions is only more
problematic than getting it to go down the shops and see how much
Camels are because giving the machine legs or wheels is easy (or
at least, a solvable engineering problem), whereas no-one knows
how to give it an ESP faculty.  Turing's mention of ESP raises
more than one issue at once:

a) getting the machine to play a game quite different from the
one addressed in most of the article
b) getting the machine to have senses (i.e. to be able to report
on the world)
c) getting the machine to have a sense which we don't understand.

My argument was that (c) doesn't matter, isn't needed to explain
why ESP is a problem for Turing - the first two are sufficient.

>If it can, there's
>>another question raised as to what exactly constitutes _really_
>>being capable of falling in love, liking ice-cream etc (easily
>>glossed as "what constitutes being _really_ human") beyond such
>>consistent behaviour.
>
>One gets the feeling that he is challenging the way we take
these concepts
>for granted amongst ourselves, as much as making the theoretical
case for
>their imitation by a universal Turing machine, now known as a
programmable
>digital computer,


Yep, a great article for that sort of nervousness....

[snip]
> At that time, E.S.P., if it is a real "human only"
>phenomenon, would be a thorny obstacle for the machine. In
Turing's view,
>the true nature of our cognitive talents- whether or not they
are
>determined and capable of being wholly imitated by discrete
state machines-
>might only be discerned when dsm's are powerful enough to
imitate us with
>sufficient accuracy to make such subtle differences between us
and "them"
>apparent. This would mean, of course, that E.S.P. is not a
deterministic,
>programmable phenomenon, and so not in the repertoire of a dsm.


rj was asking whether the term "extra-sensory" is appropriate.
If the statistical evidence for ESP is or does become
overwhelming, there's a problem with what happens next, to do
with the tension between "real" and "human only".  Any "proof" or
general consensus that ESP is both of these would be quite a
major change, because of the general tendency to reduce the
"human only" to the "real" (i.e. the reproducible, that uses a
known mechanism rather than just "happening" - very close to
"possible as a programmable faculty that a machine could be
given").  "extra-sensory" in ESP can also be made to stand for
"extra-scientific" or "extra-deterministic" - and that makes it
very handily "human only" but also inevitably unlikely to be
accepted as "real".
This is a special thing about ESP: it's not just not understood,
it carries a whole lot of baggage that makes it something you
might hope is never understood.  I think Turing's mention of it
is confusing because of all this baggage.  Being capable of ESP
could be made just as unproblematic in the Turing game as having
legs and being capable of going down to the shops, _if_ ESP was
understood.  But if it was, something else would have to take its
place as the rallying point for the intuition that no machine can
fully reproduce a human being.

Reminds me of that argument between Pointy, Geza Roszavolgyi and
Paul de la Nuit

'But the Reverend Dr Paul de la Nuit is not fond of the MMPI.
"Rosie, are there scales for measuring interpersonal traits?"
Hawk's nose probing, probing, eyes lowered in politic meekness,
"_Human_ values?  Trust, honesty, love?  Is there - forgive me
the special pleading - a religious scale, by any chance?"'
(GR p 81)

Great reading on mind if anyone's interested (_not_ that quoting
the titles gives me authority - after all, I could be lying about
having read them)  :)

Richard Rorty - Philosophy and the Mirror of Nature
Dennett - Mindstorms.

regards





seb