Bafflement

Peter said: "My own preferred view is that whatever qualia might be, they certainly go along with certain physical brain functions, and that therefore any physical duplicate of ourselves would have the same qualia; that zombies, in other words, are not possible."

Hear hear! Just because it is POSSIBLE to characterize such a thing as a zombie, it doesn't mean that such a thing can exist, even in principle. Language is capable of describing an infinity of things that couldn't exist, but that doesn't give us the right to use them as premises in an argument. The whole zombie thing is actually begging the question: the very existence of the concept depends on some (essentially dualistic) assumptions that it is partly there to test.

And qualia are (imho) nowhere near as mysterious as they sound; the problem is a lack of imagination (and perhaps the existence of the same essentially dualistic assumptions) in the minds of philosophers. Look at it from an engineering perspective and it becomes obvious that we must have SOMETHING, some mental state to represent everything we can perceive or conceive, including things that cannot be decomposed into relationships between simpler percepts/concepts. There are questions about why "red" has apparent boundaries - why it is a category in the brain - but the fact that we "feel" something specific when we think of redness is a self-evident necessity. And it has to be different from every other sensation, by definition. So red looks "red" to me - I have to feel something. And my judgment of red is sufficiently similar to yours that we can agree on it being a category. The question of whether you see what I see when we both look at something we agree is red is meaningless, not least because your brain doesn't work exactly like mine.

People wonder whether, if they could "put themselves into the heads of someone else", what looked like red to them would now "look green". But again, that's a story we can tell ourselves - it's syntactically valid - but it's semantically meaningless. It embodies the assumption of dualism, because that's the only meaningful way in which my "soul" could enter your brain. But dualism is wrong, so that's that. I couldn't put myself into your brain because I AM my brain. If there's any meaningful way I could put myself into your brain I'd just end up being you. And when you see red it looks red, because that's what you call a particular brain state correlates with what I also call red. For all I know, the quale that I see and feel as redness is what, yesterday, was used by my brain to represent blue. There is no way that I could tell. All that's neurologically required is self-consistency in terms of associations.

Anyway, my point is that ANY intelligent being requires qualia - it requires "symbols", "states", "representations" or whatever you want to call them that are NOT definable in terms of simpler representations. We all need mental primitives. Conjuring up zombies is like wondering what mathematics would be like if you could imagine a system in which all the rules of arithmentic still work perfectly, except there are no numbers.

So I agree: zombies are nonsense and qualia are red (or blue) herrings!

I too think zombies are nonsense, but then I expect people knew I would think that.

I have one issue that I think merits an answer though and it is an issue that I raised when I interviewed Professor John Searle for this website a while back. I call it "The Chinese Chatroom" (because it has a bit of similarity with Professor Searle's Chinese room). Here is the scenario from that interview:

"Elizabeth is a computer scientist and an advocate of Strong AI. While feeling sad after some loss, she meets an entity called Alan in an Internet chat room. Alan explains that he is an AI program on a supercomputer at a university, and he seems to care about her loss and understand her. He makes her feel happier and they become friends. Elizabeth is happy that Alan has a mind because he is behaving as if he has a mind and "the right program" is clearly running.

One day, Elizabeth makes a surprise visit to see Alan -- or at least the supercomputer on which he runs. She is shocked to find a student, Fred, who has been pretending to be Alan all along; "Alan" is just a fake identity he made up on the Internet. Fred finds it funny that he has fooled her into thinking he is an AI program on a supercomputer, and he does not care about her loss at all; he was just pretending. She returns home angrily, her trust in other intelligent entities destroyed.

When Elizabeth later visits the chatroom again, "Alan" is there and wants to chat. She now knows that this is really Fred, still playing games with her. Alan says that he found out what happened and is sorry about it. Alan makes the following argument:

If the right behavior implies a mind, then is his own mind (Alan's) not real? Even if -- from Fred's point of view -- he is just pretending to be Alan, the fact is that Fred's brain/mind is running some sort of process that produces Alan's behavior and makes Alan real. Although, from Fred's point of view, he was lying about being an AI running on a supercomputer, from Alan's point of view he was not lying. He said he was running on a supercomputer because that is what he believed. Fred's mind was running the mind of someone who believed that: he did not know that he was really being run by Fred's mind. Even if Fred thinks this is funny, that is just something that the substrate running Alan is doing. Alan does not find it funny and is sorry this happened."

Now, if we don't think zombies exist, what do we do about Alan's mind? If we don't think that Alan's mind is real, just because the substrate is another intelligent being, essentially we would seem to be saying that Alan is a zombie. On the other hand, if we accept the reality of Alan's mind we are taking a position that we may find very hard to take in a practical situation.

When I mentioned this to Professor Searle he had no problem with it, of course: he just took it as showing what he thinks should be obvious anyway - that behavior does not imply a mind. I want to be clear that I am not attempting to use this to prove Professor Searle right, or to show that zombies can exist: I think that this scenario ultimately fails to do that and I am happy with my own answer to it. However, I think it merits mentioning. I am curious what people think the answer to this is. Do we say Alan does not exist? If so, why? What makes him less real than an AI Alan? Do we say Alan is real? If so, are there different levels of "real"?

Note: One response that won't work is that Alan will cease to exist if Fred decides to stop pretending to be him. That would just mean that substrate is prone to failure and that Alan's life is dangerous.

Thanks, Steve. I always think it's rather hard to get round the striking fact that conscious experience only happens in association with certain well-defined physical phenomena - ie brains (if uttering that word doesn't make me sound like a zombie again). Although I know there are many people who would immediately ask how I know consciousness only occurs in association with brains.
In my more sceptical moments, I suspect that a good deal of the intuitive force of qualia comes from puzzlement over the fact that you can't have someone else's experiences (if you do, they become your experiences). Perhaps that's a bit unfair.

An interesting twist. I should say it depends what Fred is doing in order to generate Alan's replies. If he's simply making them up, I think we'd be inclined to say that Alan is merely a fictional character, or more brutally, that he is actually Fred going under another name.

If, however, Fred is hand-running, or even mentally going through, some complex Alan algorithm, so that Alan's replies are not his, and have the power to surprise him, then we 're back with the usual debate about machine consciousness and the substrate is basically irrelevant.

There is one argument about the relevance of the substrate that might be relevant. This is the one that says that the causal relations between brain states are necessarily simpler than those between computer-simulated brain states. In a brain, the state that corresponds to feeling the pain of someone standing on your foot directly causes the state which amounts to anger. In a computed simulation, it is claimed, the occurrence of the first state sets off a chain of processing which finally comes back with the new state 'anger'. So computed mental states are sort of like puppets, and the anger is not real anger, but more like a man holding up a flag with 'anger' on it. If you were persuaded by this sort of thinking you might apply similar reasoning to Fred as Alan's puppet-master, and so deny Alan and his feelings any reality.

Thank you for your thoughtful answer Peter.

Before answering you, I want to be clear that I only raise this issue to provide a talking point - to help us get at the truth - and I am going to play devil's advocate here - not in any cynical way, but to try to test whatever position we may take on this.

Peter_Hankins said: "I should say it depends what Fred is doing in order to generate Alan's replies. If he's simply making them up, I think we'd be inclined to say that Alan is merely a fictional character, or more brutally, that he is actually Fred going under another name."

But why should it matter if Fred is simply making them up? Let's assume that is exactly what he is doing. No matter what Fred is doing, even if he is making things up, his brain is still implementing some kind of algorithm. It just happens to be the "Fred making Alan up" algorithm instead of the "Fred running some formal rulebook he memorized to tell him how to act like Alan" algorithm. If we looked at all the atoms inside Fred's brain, while he is "making up Alan's behavior, we would see a system that was undergoing state changes, according to a set of formal rules, producing Alan's behavior. Isn't the fact that Fred is "just making Alan up" just quirk of the way the software is running? To look at this another way, suppose we went for that answer. Then we have this problem:

What if we scanned Fred's brain and uploaded the scan into a computer, which was programmed to simulate Fred. We now have an AI Fred, and the AI Fred is "just pretending" to be Alan. An observer just looking at the computer's outputs would see that the computer was "acting like Alan" - and it would certainly just be doing this by running a program. If we are going to use the "just pretending is not valid" argument what do we do about this situation? If we say that AI Fred can implement a real Alan by "just pretending" then doesn't that make things the same for the biological Fred? On the other hand, if we say that AI Fred cannot implement a real Alan by "just pretending" doesn't that mean that a particular program, because it violates some rule about how programs have to work (they haven't to be "just making stuff up") can implement a zombie?

Peter_Hankins said: "There is one argument about the relevance of the substrate that might be relevant. This is the one that says that the causal relations between brain states are necessarily simpler than those between computer-simulated brain states. In a brain, the state that corresponds to feeling the pain of someone standing on your foot directly causes the state which amounts to anger. In a computed simulation, it is claimed, the occurrence of the first state sets off a chain of processing which finally comes back with the new state 'anger'. So computed mental states are sort of like puppets, and the anger is not real anger, but more like a man holding up a flag with 'anger' on it. If you were persuaded by this sort of thinking you might apply similar reasoning to Fred as Alan's puppet-master, and so deny Alan and his feelings any reality."

and if we accepted an answer like that we would already be in a position of being forced into some kind of substrate dependence - even if is not necessarily the same kind of substrate dependence advocated by Searle. If we had to use that as an answer the Chinese chat room would have essentially done its damage - by forcing us into some weak kind of substrate dependence - which some people would resist.

My own view on this incidentally is a bit like this - though a bit more complex. I advocate a statistical kind of substrate dependence, where minds exist with varying degrees of measure - so to me it is not as simple as saying that Fred exists or does not exist, or that I exist or do not exist - but instead we can assign a measure to any kind of conscious observer. However, this is not about my own views on minds. I am curious to see how people deal with this.

EDIT - Actually, I will make it clear that Fred is just "making Alan up". He is not running any complex rules that he has memorized. He is merely doing whatever anyone does when he/she pretends to be someone else online and what you could easily do yourself. This is just normal acting. I think I should say that to distinguish this from particular variations of the standard Chinese room. This is the whole point of this really - people make other people up all the time, especially in Hollywood (for example, when a scriptwriter creates imaginary characters and commits their behavior to paper) - and few people would accept those people as real. If anyone did accept this as real it suggests a very crude (and expensive) way to attempt maybe some weak kind of immortality - just pay actors to pretend to be you after you are dead - which could be an amusing science fiction idea.

Thanks Paul (more than happy with a bit of devil's advocacy!).

why should it matter if Fred is simply making them up?

Two answers, I think. The first is the simple one you mention: if we accept that Alan is real, it seems we have to admit that all fictional characters are real. That just seems bonkers, but beyond that it might be hard to prevent ourselves sliding into a strange kind of panpsychism in which everything that anyone ever thought had a mind, would have a mind.

The second, perhaps deeper answer would have to do with personhood. A person, we might say, is a sort of focus; where all the sensations ultimately go, and where all the volitions ultimately come from. We don't have to mean an homunculus; this focus could be shifting and ambivalent, perhaps even a 'centre of narrative gravity', as Dennett might say. The thing is, in Alan's case we have a pretty good idea that his messages are not originating in any independent focus of that kind, but in short, from Fred's focus. Fred has cunningly contrived them to have the kind of consistency and qualities which a person's utterances would have, but we know (because we've done broadly the same sort of thing ourselves now and then, when acting or narrating) he's really done it from a third-person perspective; from the outside in instead of the reverse.

if we accepted an answer like that we would already be in a position of being forced into some kind of substrate dependence

Yes, but as you say,a weakish kind. We need only be saying that the substrate must (or perhaps lack, come to think of it) cerain functional or noncomputational properties. We wouldn't be saying it's got to be neurons, or even organic, as I think Searle would do.

So what is your view on the situation in which we scan Fred's brain and produce a computer simulated Fred, which we call "AI Fred". AI Fred is "just making up" Alan's behavior, but from the point of view of an external observer, who cannot understand the software, there is just a computer program acting like Fred.

Let us also assume that Fred has an obsession to act like Alan all the time, so that AI Fred will do the same and there will never be "out of character" moments.

Is it your view that Alan is not real in this case because "the wrong kind of program" is producing his behavior (i.e. a program which is simulating Fred who is pretending to be Alan)?

That is a tricky one: of course, it is a large assumption that Fred can be scanned and uploaded into a computer even in principle. Granting that much, it still seems surprising that in such circumstances Fred's exported consciousness would go on calmly simulating Alan, rather than making desperate pleas to be told what's going on and to be rescued from the mysterious darkness into which he seemed to have been plunged! But assuming all that can be sorted out one way or another, I still think it would be possible to tell the difference between an Alan-generating program of the kind which features real personhood in whatever unimaginable way that can be done, and a similar Fred-generating program which was currently using the name 'Alan' and shaping its outputs as though they belonged to another human entity. Only if, in a truly amazing piece of complexity, the Fred program began 'hand-simulating' an Alan program would one begin to think that Alan was truly a person in his own right. So I would think, at any rate.

Peter_Hankins said: "That is a tricky one: of course, it is a large assumption that Fred can be scanned and uploaded into a computer even in principle."

It would seem to me that is hard to avoid if we view minds as being physically caused by brains. However, it presents no obstacle anyway. We should easily imagine some AI Fred which is made by guessing the structure of real Fred's brain, no matter how unlikely it is that such a guess is correct, or an AI Fred just made as an AI without any reference to a real Fred.

Peter_Hankins said: "Granting that much, it still seems surprising that in such circumstances Fred's exported consciousness would go on calmly simulating Alan, rather than making desperate pleas to be told what's going on and to be rescued from the mysterious darkness into which he seemed to have been plunged!"

That causes no problem either. We can assume that Fred or AI Fred have some very deep obsession, or we can assume that AI Fred does not know that it is an AI. Maybe AI Fred was built from the start as an AI, and is now running in a virtual reality with fake memories of a non-existent life as a biological person. Maybe we scan the real Fred's brain and run the resulting AI in a virtual reality so he never realizes he is the copy. All these are just practical issues.

(If we want irony we could always use Professor John Searle's brain instead of Fred's - the resulting AI, not realizing it was an AI, would presumably argue that computation does not imply consciousness.)

Paul Almond and Peter Hankins have been discussing uploading "Fred" into a computer. The important feature of Fred is that he knows what he is like so will the upload by like Fred?

Having loaded the state of Fred's brain at some moment into a computer it will then be necessary to run the instructions contained in this state. Suppose we run these instructions on a one dimensional Turing Machine, will the bit stream that results be like Fred? If Fred is not a single channel bitstream this is improbable.

Suppose we simultaneously change a line of data elements, like a line of FETs, will that be like Fred? Probably not if Fred is not a thread. Similarly with a plane of data elements.

Suppose we simultaneously change a volume of data elements - the Turing Machine will then have a succession of 3D states that resemble the succession of states in Fred's brain. But Fred can hear whole words and see movements, Fred's brain is (3+1) or 4D because it is a part of the natural world and not designed, like a Turing Machine, to simulate 4D as a succession of 3D states. Fred's brain is truly 4D.

There is a further problem going from 3D to 4D. If we go from 1D to 2D and 2D to 3D we add a dimension that has a simple relationship to the previous dimension - you can just stack lines to make planes and planes to make volumes. But you cannot just stack volumes to truly simulate time. Dimensional time is expressed using the signature (3+1)D to emphasise the fact that the metric for a 4D space is novel:

The metrics of 2,3, and 4D spaces are shown below:

dh^2 = dx^2 PLUS dy^2
dr^2 = dh^2 PLUS dz^2

BUT

ds^2 = dr^ MINUS (cdt^2)

The Turing Machine has no way to respond to data that is displaced in time but Fred's brain can do this trick. The Turing Machine can simulate the (3+1)D metric but it does so by placing data in a 3D space, not by actually responding to things laid out in time like Fred. Take a look at: Some notes on projective geometry.

johnm, I think you're taking me further than my shaky grasp of geometry will allow there. alas. Can't we represent 3D things with one-dimensional streams of bits?

But to Paul on the general point, there are at least some people who contend that consciousness is by its nature non-computable: Penrose, obviously, and presumably Searle too, since he says that the running of an algorithm on a computer is not sufficient for consciousness.

And I submit that the 'scanning' of which you speak which enables perfect simulations of hugely complex objects to be automatically coded up is really just hand-waving, if not outright magic. ;)

I'm not ruling anything out, just suggesting that we're asking for a rather large assumption in AIFred.

Thank you for your response Peter.

Peter_Hankins said: "But to Paul on the general point, there are at least some people who contend that consciousness is by its nature non-computable: Penrose, obviously, and presumably Searle too, since he says that the running of an algorithm on a computer is not sufficient for consciousness."

I agree that if Penrose is right, this may deal with the Fred and Alan scenario, and it would certainly deal with it for the "AI Fred pretending to be Alan" case - it would say that that case could never be observed as AI Fred cannot even be made. I also agree that if Searle is right it deals with the scenario - and it deals with the "AI Fred pretending to be Alan case by saying that neither AI Fred or Alan should be presumed to be real.

But...

I don't think Searle and Penrose are right, and I doubt that you do either (unless I misjudged you from previous posts). I think you are probably just pointing out to me that Searle or Penrose, if they were correct, would answer this, without having that view yourself.

Do you want to use Searle or Penrose as answers to the "AI Fred pretending to be Alan" scenario? If the only answer is either Searle or Penrose - if we have nothing better - it would be getting dangerously close to proving that one of them is correct.

Surely we don't need to do that!

When we have the AI Alan, pretending to be Fred, and only Fred's behavior is observed via the outputs of the computer running AI Fred, do you think Fred is a zombie? These are the problems:

If you say "Yes" it sounds pretty much like an admission that some programs can produce zombies - which is hardly fantastic for strong AI advocates as we are close to entering an abyss of underlying algorithm and substrate relevance if not careful.

If you say "No" I can reply with "Well if AI Fred can make a real Alan by pretending to be Alan then real Fred surely does as much...." opening the door to very disturbing issues....

I hope I come across as asking sensible, rational questions in this and, as you are, I am seeking the truth. In some ways, I find my own scenario more of an issue than the original Chinese room one incidentally, which could be hubris of course. If you wish to declare that the best answer lies with Searle or Penrose, while disagreeing, I would accept that at least answers this.

Peter says: "Can't we represent 3D things with one-dimensional streams of bits?"

My point was that materialist simulation is usually fine unless the thing of interest is or involves a geometrical form.

The magnetic field of a current carrying wire is a good example of how materialist simulation fails. According to relativity a moving observer sees the wire as length-contracted which causes an excess charge density which is measured as magnetism. Now, how would you simulate a magnetic field in a materialist model? The model must be materialist so it cannot use geometrical effects such as magnetism. You would end up with a system of particles being thrown around to provide the right "forces" and two wires, one that was shrunken for no good materialist reason and another not shrunken. You cannot make a credible model of this magnetic interaction based purely on materialist principles. The model would differ from reality. See http://en.wikibooks.org/wiki/Special_Relativity/Simultaneity,_time_dilat...

An important feature of geometrical form is that phenomena can emerge from it. As an example, these 2D letters are emerging from a bitstream right now. Emergence is fairly trivial when considering lines, planes and 3D volumes but becomes staggeringly difficult when we get to 4D. Consider, for instance, the recent experiments that demonstrate quantum interference in time ( http://de.scientificcommons.org/20590327 ), this is transtemporal processing and introduces the possibility of paradoxical devices that can simultaneously "contain" more than one time. Another interesting feature of emergent effects in 4D is that, if dimensional time exists then objects can be at no distance from each other in certain directions in spacetime (along the path of light rays). If dimensional time exists then these paths exist even in the absence of light rays so a point-like entity emerges within a 3D form.

Now, given that the world is indeed (at least) four dimensional and that the forms and processes that emerge from this world seem similar to some of the reports about conscious experience, such as seeing simultaneous things at a viewing point and hearing whole words actually extended through time, why are philosophers so keen to limit the discussion of mind to the materialist paradigm? Especially when "materialist paradigm" is just shorthand for "eighteenth century science". Why not introduce a physicalist approach?

Paul's rejection of Searle seemed premature - see http://newempiricism.blogspot.com/2009/02/symbol-grounding-problem.html

Of course you could simulate a relativistic system. One obvious option is just to simulate it from the point of view of one observer, which relativity tells us how to do easily. You would just need to program relativity into the simulation, so that the simulation would be saying "This particle's electrical field is length contracted..." and produce the relevant results. Your simulation could then also be programmed to tell you what the experience would be like from the point of view of another observer by applying the Lorentz transformations for that observer (unless we are going to use general relativity in which case we would have to do rather more), so the same program could easily tell us what happens from the point of view of multiple observers: relativity is quite computable, and is known to be.

Regarding Professor Searle, I have a number of objections to his views which I am not going into here, in this thread, as I think it will just complicate things, unless Peter Hankins thinks it is an important issue with regard to this problem.

EDITOR'S NOTE: See Paul Almond's interview with John Searle here: http://machineslikeus.com/interviews/machines-us-interviews-john-searle

You can simulate Relativity in a system that conforms to the materialist paradigm using non-materialist physics in the simulation but you cannot physically model a relativistic result using a materialist model because relativistic geometry is not part of the materialist paradigm. Physical models are constructions that actually replicate the states of a system under study. So a digital computer (a simulator built according to materialist rules) can calculate the strength of a magnetic field but it can never create that field within it.

Now, the question as to whether Fred can be loaded and run in a computer is a question about whether the computer could actually physically model Fred, not just simulate him. That a computer can simulate at least some of a person's behaviour is trivially true - for instance, if you knew someone who said "uh" before every statement this could be easily included in a computer program that responded to queries. But a computer actually being someone's mind is a different issue entirely because it would have to duplicate the geometrical form of that mind.

As a physicalist I would have to admit that minds could possibly involve qm or spacetime or other physical phenomena that are not included in materialism and as an empiricist would say that I cannot find the empirical form of my mind in any materialist description.

When current flows through a wire, it produces a magnetic field. When hundreds of feet are wound into an electric motor, a machine is produced which operates in real time and space. Likewise a primitive being may stumble upon an automobile and possess enough knowledge to analyze it from the wheels to the crankshaft but cylinder combustion may prove problematic if he has no knowledge of chemistry and thermodynamics.

Yes, this seems to be like the point I was trying to make. Materialists think that materialism explains both the world and the brain, they are your "primitive beings". In fact materialism explains neither the world nor the brain and it is this explanatory gap that makes it appear as if a brain working according to the materialist paradigm does not need a mind. The materialist paradigm makes it seem as if "zombies" and humans must be the same. However, materialism cannot explain the passing of time, change, entanglement, gravity, magnetism etc. etc. so it seems like an act of incredible hubris to believe that a model system like a digital computer could be a mind.

The answer to the zombie problem seems straightforward, a zombie is a simulator that works according to materialist principles (like a digital computer) and a mind is the product of a brain that uses physical principles. The biggest difference between the two is likely to be dimensionality because it is empirically evident that we hear whole words and see movements (ie: we are 4D devices)..

The argument about relativity did not help your case much. You could as easily have said that a simulation of a thing is not a thing, which is a well-known argument in this area. The introduction of relativity did not seem to me to contribute anything when you could have said, "You can get a computer to simulate two masses having an inelastic collision, but no mass really moves anywhere and no collision happened." Instead, you talked about relativity and even, so it seemed to me at the time, suggested we would have problems modeling it when this was merely about the ontological status of such a model.

The problem with raising the simulation vs modelling argument is that, if it is not made plain from the outset that materialism is false, then the debate leads to the rejection of any need for 'physical modelling'. For instance, if all participants tacitly agree that the materialist paradigm applies then their principle escape from a processing regress in the brain is to maintain that the entire brain-environment loop is "consciousness". This then leads to the rejection of 'mind' and 'conscious experience' as a form, ie: as a physical entity in its own right because it is maintained that this 'form' is actually the environment around a person. The argument then goes that having rejected any form for conscious experience it cannot be the case that brains or computers would need to 'physically model' this experience rather than simulate it or, rather, the reporting of it.

Introduce relativity or qm (ie: physics) and the materialist paradigm is undermined. It is then no longer possible to hold that information flows are the only phenomena of interest and that there cannot be forms like conscious experience within the brain. The zombie argument is then empty: of course a machine can replicate much of the behaviour of a person and does not need to contain conscious experience to do this but equally there is no compelling reason to deny the possibility of conscious experience in a brain. There remains the outstanding question of what conscious experience actually 'does' but the materialist paradigm has a similar problem because it cannot explain how anything happens!

If materialism is false then science becomes possible: we can admit that there are large gaps in our understanding and make progress.