> Hans writes
> >>
> Russell Standish <R.Standish.domain.name.hidden>:
> > ... However, it is always possible to _artificially_ construct a
> > system to pass the Turing test that isn't concious.
>
> Balderdash!>>
>
> Hans is very succint so it's not clear if his deprecation is about building a
> Turing machine that can pass the Turing test or a Turing machine that is
> conscious or both of the above. So I may or may not agree with him.
I clarified in the previous email that I meant "to pass a particular
Turing test in which one knows all the questions beforehand". This is
the case of the Olympia story. Sorry for the confusion.
>
> A lot of discussion has been made about consciousness as if it was an
> objective, absolute quantity. And yet the only way to measure consciousness
> is by means of a relative test: the Turing test. Basically this test COMPARES
> the performance of a machine to a human.
>
> I think that the only way to meaningfully "measure" consciousness is by means
> of a Turing test. So by definition if a machine passes the Turing test then
> it is conscious. If it doesn't then it isn't. It is understood that the
> Turing test will not be a superficial examination, but a thorough evaluation
> evolving measuring intellectual as well as emotional (even irrational)
> responses to a wide variety of situations. If the machine behaves like a
> human, then it should be entitled to all human rights under the law. If it
> behaves partially as a human, say in the intellectual domain but not in the
> emotional one, then, in relativistic terms, it differs from the human in that
> respect. Its consciousness STILL EXISTS BUT IS DIFFERENT from the human.
>
I demur from this slightly. Each and every one of us experiences
directly our own conciousness. We do not perform a Turing test on
ourselves to decide whether we're concious. However, the only way to
ascribe conciousness to external objects is to perform a Turing
test. i.e. we must ask "Is our conciousness a good model for the
proerties of this external object?". It is clear that like all models,
it can be an incorrect description. I think Jacques is correct in
saying that the Turing test is neither necessary or sufficient - it is
simply the best we have. We have discussed a number of scenarios
whereby being too thorough in an investigation will reveal a concious
object as nonconcious.
> Similarly, a humans twin pretending to be the other twin, may not be capable
> of passing a Turing test aimed at finding out who he really is. This only
> proves that THE TWINS HAVE DIFFERENT CONSCIOUSNESS UNDER THIS PARTICULAR
> TURING TEST.
>
> Now the kicker!. Had the Turing test not be capable of distinguishing the
> twin from his sibling, then UNDER THIS TURING TEST, THE TWINS WOULD HAVE THE
> SAME - IDENTICAL CONSCIOUSNESS!!! They would be the same person! Is this a
> paradox? Not at all. It is the same kind of situation generated when two
> persons, one in motion with respect to the other measure the velocity of an
> arbitrary object. They conclude that the object is animated with a different
> velocity with respect to their own frame of reference.
>
> Thus a given person/machine will have a different consciousness depending on
> who is doing the observing. The "observer" could be a Turing test as well as
> a full fledged human, equipped with his own consciousness. The feeling of
> **I** is the self observing the self in an infinite recursion loop.
>
> Thus, I propose that consciousness is a full fledged relativistic quantity.
> As such I must define how to specify its frame of reference. The frames of
> reference consists of information: That contained in the mind of the observed
> object and that contained in the mind of the observer. The information of
> interest is the mutual information which is the the difference between the
> two.
>
> I would like to formalize the argument by comparing the information contained
> in the mind and in Turing tests to information constained in axiomatic
> mathematical systems. The full power of Goedel's incompleteness theorems
> could then be brought to bear. Some "truth" may not be provable within the
> system, similarly, the twins may not be distinguishable by a given Turing
> test and a machine may "appear" to be human, simply because the observing
> apparatus (human or test) does not have the power to distinguish it from a
> human.
>
> Some of you may argue that even if a test is not capable of distinguishing
> between two consciousness, these consciousness have an objective, absolute
> reality. There is no way, however, to prove such an assertion independently
> of the Turing test process.
>
> Wish me a good vacation! :-)
>
> George
>
> --part1_e65644bc.24cf8de4_boundary
> Content-Type: message/rfc822
> Content-Disposition: inline
>
> Return-Path: <everything-list-request.domain.name.hidden>
> Received: from aol.com (rly-zd05.mail.aol.com [172.31.33.229]) by
> air-zd04.mx.aol.com (v60.18) with ESMTP; Tue, 27 Jul 1999 14:24:49
> 2000
> Received: from mx1.eskimo.com (mx1.eskimo.com [204.122.16.48]) by
> rly-zd05.mx.aol.com (v60.18) with ESMTP; Tue, 27 Jul 1999 14:24:40
> -0400
> Received: (from smartlst.domain.name.hidden)
> by mx1.eskimo.com (8.9.1a/8.8.8) id LAA04227;
> Tue, 27 Jul 1999 11:21:16 -0700
> Resent-Date: Tue, 27 Jul 1999 11:21:16 -0700
> Message-Id: <199907271820.UAA09542.domain.name.hidden>
> Subject: Re: Implementation
> Date: Mar, 27 Jul 99 20:17:57 +0100
> x-sender: marchal.domain.name.hidden
> x-mailer: Claris Emailer 1.1
> From: Marchal <marchal.domain.name.hidden>
> To: <everything-list.domain.name.hidden>
> Mime-Version: 1.0
> Content-Type: text/plain; charset="US-ASCII"
> Resent-Message-ID: <"KKnug1.0.-11.RYVdt".domain.name.hidden>
> Resent-From: everything-list.domain.name.hidden
> X-Mailing-List: <everything-list.domain.name.hidden> archive/latest/1007
> X-Loop: everything-list.domain.name.hidden
> Precedence: list
> Resent-Sender: everything-list-request.domain.name.hidden
> Content-Transfer-Encoding: 7bit
>
>
> I will make some comments on the last posts. But we are entering in
> very deep
> waters and I would like to make general remarks.
>
> I said it before, but I want
> repeat it here. One of my main goal is to understand what is "the
> physical" and where does it comes from. And like Wheeler I don't think
> that this can be explain by physical laws.
>
> With Ockham razor there is no need of
> the crackpot/Maudlin argument, the UD argument (PE-omega) is quasi-enough
> to convince oneself
> that the 'universal part' of physics must be extract from computer
> science.
>
> Note also that the 'reversal' is in both Tegmark and Schmidhuber (it
> seems to me), but they
> haven't see the measure problem (do they ?), and they havent' put their
> methodology to its logical extreme.
>
> Another general remark is that, in your post, I agree sometimes whith
> what you are saying until you jump to a conclusion which I don't
> understand.
>
>
> George Levy wrote:
>
> >Similarly, the insertion of the piece of wood in the computer must be done
> >by
> >someone. Let's call that someone Maudlin's demon. Deciding what the right
> >place and the right time is to make the wood irrelevent to the thinking
> >process, in order to satisfy the counterfactual role that the wood must
> >play,
> >requires Maudlin's demon to think. Maudlin's demon then becomes part of the
> >computer's consciousness just like the subject in the chinese room
> >experiment
> >becomes a cog in the chinese room ability to speak chinese.
>
> An interesting similar (although computationalist) move has been made
> by Eric Barnes "The causal history of computational activity: Maudlin and
> Olympia. (The journal of philosophy 1991, pp 304-316.)
> What is interesting for me is that Barnes'move forces him to pretend
> having the ability to distinguish between being awake and being dreaming
> (which I doubt), (which contredict also my theetetic self-referential
> theory of knowledge where p is known when p is justified by the machine
> and true or just consistent).
>
> George Levy wrote:
>
> >In the end, I am a strong sceptic of both computationalism and physical
> >supervenience. I believe that consciousness exists only in the eyes of the
> >beholder, and is a relativistic property, based on the relativity of
> >(mutual)
> >information as defined by Claude Shannon.
>
> I am definitely open to the idea that consciousness exists only in the
> eyes of the beholder, that it is a relativistic property, based on the
> relativity of mutual information as defined by Claude Shannon, Kolmogorov,
> and as used by Everett (but see also the paper of Adami and Cerf in the
> quant-ph).
>
>
> Hans Moravec wrote:
>
> >So, deterministic machines can have just as much free will as you or
> >I. The key is that they don't know everything that's going on,
> >outside themselves or in, so often don't know what will happen next,
> >or how they will respond to it. Many-worlds may provide an
> >interesting additional "source" of ignorance, but limitations on what
> >a finite process can model already provide sufficient ignorance for
> >free will even in a fully deterministic framework.
>
> I agree. What MW or self-duplication adds is truly random uncertainty.
> This is "testify" by quantum computers.
> I don't think determinism is an "effective" problem for free-will,
> nor do I think randomization can help in making free-will possible.
> I think free-will is related with the boundary of self-knowledge.
>
>
> Hal Finney wrote:
>
> >But more than one computation can be conscious, obviously. It is
> >conceivable that the new computation, although different, is conscious
> >as well. This is a possible escape from Maudlin's argument.
>
> > [...]
>
> Are the content of consciousness different ? or the intensity are
> different ? I'm not sure I understand.
>
> HF:
> >So it seems to me we need a new argument for why the computer sans
> >counterfactuals should not be considered conscious.
>
> > [...]
>
> It seems to me that a computer without counterfactuals is like
> a doll, a teddy bear, or a sculpture.
> Unlike Hans I don't understand what would it mean to ascribe them
> consciousness.
>
> HF:
> >One of the thins which is attractive about Wei's approach, as I understand
> >it, is that it does not try to answer the question of whether a given
> >system is conscious, at least not in yes-or-no terms. Rather, it tries
> >to give a probability that a given system is conscious, and specifically
> >that it instantiates a particular consciousness, such as my own.
> >
> >This allows you to have such things as systems which are "probably"
> >conscious, or, in a sense, "partially" conscious (in the sense that
> >we can treat them as having a 10% chance of being conscious, say).
> >This interpretation makes most sense in the context of the Strong
> >Self-Selection Assumption (that we can consider our moments of experience
> >as randomly chosen from among all observer-moments). The probabilities
> >assigned to consciousness serve as a weighting factor for how much they
> >contribute to the ensemble of all observer-moments.
>
> I agree and I appreciate very much this way of seeing the things. It is
> linked (from my humble understanding) to James Higgo's anthropic
> principle/occam-razor. This 'interpretation' along with comp leads to
> a total reversal ....but in the relative way (I will not insist here).
> I do link consciousness with machine's inference about their own possible
> consistent extensions.
>
>
> David Seaman wrote:
>
> >This seems an excellent viewpoint, consciousness requires the freedom to
> >react to a reasonably wide range of circumstances in a way which is not
> >predictable to other observers. So a single execution can never confirm or
> >deny consciousness however many times it is replayed. But I'm not so sure
> >that a Turing machine cannot have free will. I'd guess that the appearance
> >of free will can emerge from a sufficiently complex TM provided that the TM
> >exists in a suitably complex environment. If a person built a machine
> >containing a TM it would be part of our MWI universe and the requirements
> >could be satisfied. This would not be an isolated TM since it would be
> >simulated by and react to its environment, and any 'randomness' requirement
> >could actually involve a sensitivity to gravitons, photons, or quantum
> >interference.
> >
> >I tend to agree that a completely isolated TM is unlikely to have free will
> >or be conscious (and in any case it would be impossible to test it). Of
> >course the program executed by an isolated TM may well be able to generate
> >a universe containing conscious subjects. In the special case of an
> >isolated TM generating a universe which contains exactly one conscious
> >subject in a suitable environment it could loosely be said that the TM's
> >program is that conscious subject. But this is different to saying that
> >the TM itself is conscious, and it would not be apparent from looking at
> >the TM that it was generating consciousness.
>
> I agree in part. It is easy to build version of dreaming Olympia.
> A dreaming machine would be, at least here and now, an isolated
> 'conscious'
> (but not necessarily awake) machine.
> I think an isolated 'conscious' machine cannot be isolate for ever for
> purely computational reasons.
>
>
> Jacques Mallah wrote:
>
> > Bruno, I think it is now abundently clear that Maudlin's paper
> >does not rule out physical computationalism, and other people on the list
> >have seen that as well.
>
> Clear would be enough. Abundently clear is a little to much.
>
> I don't understand what really means 'physical' in physical
> computationalism.
> It is clear that we have not the same primitive elements.
> I believe in numbers and number's dreams. Some dreams are deep and
> partially sharable among UTMs, those are their relative realities.
>
> I appreciate the everythinger's work on these questions, and I guess it
> is not easy to abandon the physical supervenience thesis.
>
>
> Russell Standish wrote:
>
> > > Chris, this is a well thought out reponse, and it persuades me that
> > > the difference between conciousness and nonconciousness could be as
> > > little as the "inert block of wood", precisely because it is a
> > > physically different system. It actually reminds me of the quantum 2
> > > split experiment. An interference pattern is seen, or not seen
> > > according to whether a detector placed at one of the slits is switched
> > > on or not.
>
> > [...]
>
> 1) The great programmer dovetail also on the quantum turing machines...
> 2) I think so. There is a deeper analogy between the computationalist's
> counterfactuals and the quantum. This is linked to a paper by Hardegree
> showing a formal similarity between a very natural definition of 'quantum
> implication' and Stalnaker's logic of counterfactual and my own definition
> of 'observation'. (ref. in my thesis), and the resulting arithmetical
> "quantum logic".
>
> >Thinking about this some more, I realise this is exactly what is going
> >on. Consider Olympia from a MWI point of view. For the vast majority
> >of worlds containing Olympia, Karas (I believe that is what the
> >correcting machinery is called) is active, handling the
> >counterfactuals. Only on one world line (of measure zero!) is Karas
> >inactive, and Olympia is simply replaying the previously recorded
> >data.
> >
> >Now consider what happens when Karas is turned off, or prevented from
> >operating. Then, in all world lines is Olympia simply a replay
> >device. From the MWI point of view, the simple inert piece of wood is
> >not so innocuous. It changes the systems dynamics completely.
>
> OK.
>
> >
> >Now this has bearing on a supposition I have argued earlier - that
> >conciousness requires free will, and the only way to have free will is
> >via the MWI picture.
>
> Not OK. See above.
>
> >In this context, a Turing machine can never be
> >concious, because it follows a preprogrammed path, without free
> >will. Note this is not the same as saying comp is false, unless you
> >strictly define computers to be Turing machines.
>
> I do. It is my working hypothesis.
>
> >My suspicion is that
> >adding a genuine random number generator to the machine may be
> >sufficient to endow the architecture with free will, however, of
> >course the question is unresolved.
> >
> >What does this all mean for your thesis Bruno? Alas I didn't follow
> >your argument (not because it was written in French - which I have no
> >problem with, rather because I was not familiar with the modal logic
> >you employed, and haven't raised enough enthusiasm to follow up the
> >references). Could it be implying that you have too restrictive
> >definitions of both comp and sup-phys?
>
> Church's Thesis is a vaccin against any restrictive interpretation
> of comp. Comp makes the unknown much bigger that we have ever thought.
> Even if from the archimedian point of view there are only numbers
> relationships.
>
> >
> >Quote from Bruno follows:
> >
> >> This seems rather magical to me. If only because, for a
> >> computationalist,
> >> the only role of the inert block (during the particular execution) is
> >> to
> >> explain why the machine WOULD have give a correct answer in case the
> >> inputs WOULD have been different.
> >> This mean that you don't associate consciousness with a particular
> >> physical computation but with the entire set of possible computations.
> >> But that is exactly what I do ..., and what I mean by the abandon
> >> of physical supervenience.
> >> A singular brain's activity becomes an invention of the mind.
> >
> >Could it mean that you are defining sup-phys to be supervenience on
> >the one track classical physics, rather than on the MWI style quantum
> >physics?
>
> Most people in cognitive science do but I do not care about the level of
> duplication.
>
> I think that ANY sufficiently patient self-referentially correct
> machines, either by introspection or by observation (or a mixture of
> both), will infer MWI-like physics (once observing below their level of
> duplication, for example).
>
> Bruno
>
>
>
>
>
> --part1_e65644bc.24cf8de4_boundary--
>
>
----------------------------------------------------------------------------
Dr. Russell Standish Director
High Performance Computing Support Unit,
University of NSW Phone 9385 6967
Sydney 2052 Fax 9385 6965
Australia R.Standish.domain.name.hidden
Room 2075, Red Centre
http://parallel.hpc.unsw.edu.au/rks
----------------------------------------------------------------------------
Received on Tue Jul 27 1999 - 18:01:00 PDT