On Jan 16, 12:15 pm, Jason Resch <[email protected]> wrote: > Craig, > > Do you have an opinion regarding the possibility of Strong AI, and the > other questions I posed in my earlier post? >
Sorry Jason, I didn't see your comment earlier. On Jan 15, 2:45 am, Jason Resch <[email protected]> wrote: > On Sat, Jan 14, 2012 at 9:39 PM, Craig Weinberg <[email protected]>wrote: > > >wrote: > > > > > Thought I'd throw this out there. If computationalism argues that > > > > zombies can't exist, > > > > I think the two ideas "zombies are impossible" and computationalism are > > > independent. Where you might say they are related is that a disbelief in > > > zombies yields a strong argument for computationalism. > > > I don't think that it's possible to say that any two ideas 'are' > > independent from each other. > > Okay. Perhaps 'independent' was not an ideal term, but computationalism is > at least not dependent on an argument against zombies, as far as I am aware. What computationlism does depend on though is the same view of consciousness that zombies would disqualify. > > > All ideas can be related through semantic > > association, however distant. As far as your point though, of course I > > see the opposite relation - while admitting even the possibility of > > zombies suggests computationalism is founded on illusion., but a > > disbelief in zombies gives no more support for computationalism than > > it does for materialism or panpsychism. > > If one accepts that zombies are impossible, then to reject computationalism > requires also rejecting the possibility of Strong AI > (https://secure.wikimedia.org/wikipedia/en/wiki/Strong_AI). What I'm saying is that if one accepts that zombies are impossible, then to accept computationalism requires accepting that *all* AI is strong already. > > > > > > > > > > > > > > > therefore anything that we cannot distinguish > > > > from a conscious person must be conscious, that also means that it is > > > > impossible to create something that acts like a person which is not a > > > > person. Zombies are not Turing emulable. > > > > I think there is a subtle difference in meaning between "it is impossible > > > to create something that acts like a person which is not a person" and > > > saying "Zombies are not Turing emulable". It is important to remember > > that > > > the non-possibility of zombies doesn't imply a particular person or thing > > > cannot be emulated, rather it means there is a particular consequence of > > > certain Turing emulations which is unavoidable, namely the > > > consciousness/mind/person. > > > That's true, in the sense that emulable can only refer to a specific > > natural and real process being emulated rather than a fictional one. > > You have a valid point that the word emulable isn't the best term, but > > it's a red herring since the point I was making is that it would not > > be possible to avoid creating sentience in any sufficiently > > sophisticated cartoon, sculpture, or graphic representation of a > > person. Call it emulation, simulation, synthesis, whatever, the result > > is the same. > > I think you and I have different mental models for what is entailed by > "emulation, simulation, synthesis". Cartoons, sculptures, recordings, > projections, and so on, don't necessarily compute anything (or at least, > what they might depict as being computed can have little or no relation to > what is actually computed by said cartoon, sculpture, recording, > projection... For actual computation you need counterfactuals conditions. > A cartoon depicting an AND gate is not required to behave as a genuine AND > gate would, and flashing a few frames depicting what such an AND gate might > do is not equivalent to the logical decision of an AND gate. I understand what you think I mean, but you're strawmanning my point. An AND gate is a generalizable concept. We know that. It's logic can be enacted in many (but not every) different physical forms. If we built the Lego AND mechanism seen here: http://goldfish.ikaruga.co.uk/andnor.html# and attached each side to a an effector which plays a cartoon of a semiconductor AND gate, then you would have a cartoon which is simulates an AND gate. The cartoon would be two separate cartoons in reality, and the logic between them would be entirely inferred by the audience, but this apparatus could be interpreted by the audience as a functional simulation. The audience can jump to the conclusion that the cartoon is a semiconductor AND gate. This is all that Strong AI will ever be. Computationalism assumes that consciousness is a generalizable concept, but we don't know that is true. My view is that it is not true, since we know that computation itself is not even generalizable to all physical forms. You can't build a computer without any solid materials. You can't build it out of uncontrollable living organisms. There are physical constraints even on what can function as a simple AND gate. It has no existence in a vacuum or a liquid or gas. Just as basic logic functions are impossible under those ordinary physically disorganized conditions, it may be the case that awareness can only develop by itself under the opposite conditions. It needs a variety of solids, liquids, and gases - very specific ones. It's not Legos. It's alive. This means that consciousness may not be a concept at all - not generalizable in any way. Consciousness is the opposite, it is a specific enactment of particular events and materials. A brain can only show us that a person is a live, but not who that person is. The who cannot be simulated because it is an unrepeatable event in the cosmos. A computer is not a single event. It is parts which have been assembled together. It did not replicate itself from a single living cell. > > > You can't make a machine that acts like a person without > > it becoming a person automatically. That clearly is ridiculous to me. > > What do you think about Strong AI, do you think it is possible? The whole concept is a category error. It's like saying do you think it's possible to have human colored paint. It is possible to have technology that seems to us like Strong AI, just as a mannequin can seem like a person to us momentarily. The better the simulation, the longer it will take for more people to doubt it's authenticity, but there will always be ways to tell the difference (you might need a trained guinea pig or a voice stress analyzer to do it, but eventually you could probably tell). > If so, if > the program that creates a strong AI were implemented on various > computational substrates, silicon, carbon nanotubes, pen and paper, pipes > and water, do you think any of them would yield a mind that is conscious? No. By definition, consciousness has to come from the substrate itself. If the substrate is conscious, then the program can be conscious, but the more something is conscious, the less possible it is that it can be programmed. > If yes, do you think the content of that AI's consciousness would differ > depending on the substrate? No, it's the ability to accept the program that would differ depending on the substrate. The sensorimotive awareness of any substrate is already different from any other. We play a song on a computer but the computer does not experience the song, nor do the speakers in your headphones, or even your cochlea. They do probably experience vibration, and maybe the cochlea experiences 'sound' in a zoological sense, but the song level interpretation is private to anthropological level experience. You can't put an mp3 directly into your ear or your brain. There is no AI independent of substrate. I can draw a straight line or walk a straight line, but there is no universal straight line experience. Straight and linear are sensorimotive qualities carried by particular channels of sense. > And finally, if you believe at least some > substrates would be conscious, are there any cases where the AI would > respond or behave differently on one substrate or the other (in terms of > the Strong AI program's output) when given equivalent input? I can wear a suit and tie and stand in a department store. A mannequin can do the same thing. AI is the suit and tie. Does the suit make the mannequin look more like me when I'm wearing the same suit? Sure. Does it make any difference to the mannequin? No. Does it make any difference to me? Yes, my experience of the mannequin depends on how good of a mannequin it is and how directly I look at it and for how long. > > > > > > > > If we run the zombie argument backwards then, at what substitution > > > > level of zombiehood does a (completely possible) simulated person > > > > become an (non-Turing emulable) unconscious puppet? How bad of a > > > > simulation does it have to be before becoming an impossible zombie? > > > > > This to me reveals an absurdity of arithmetic realism. Pinocchio the > > > > boy is possible to simulate mechanically, but Pinocchio the puppet is > > > > impossible. Doesn't that strike anyone else as an obvious deal breaker? > > > > Not every Turing emulable process is necessarily conscious. > > > Why not? What makes them unconscious? > > In my guess, it would be a lack of sophistication. For example, one > program might simply consist of a for loop iterating from 1 to 10. Is this > program conscious? I don't know, but it almost certainly isn't conscious > in the way you or I are. If that were the case then sophistication alone would be consciousness. It's not though. Our consciousness is certainly sophisticated but a beach full of sand is sophisticated too. Would a program that makes a copy of itself every 10 iterations be any more conscious than one that doesn't copy itself? Without some kind of capacity for sense and motive within the loops from the start, there isn't anything that knows there is any looping going on. We have to realize that there is no such thing as a 'loop' in general, anymore than there is a such thing as a touchdown in general. When we talk about a for loop we are talking about a common sense neurological modeling which relates to certain organizations of physical objects and the computational manipulation thereof. There is no looping for vapor or in a vacuum. > > > You can't draw the line in one > > direction but not the other. If you say that anything that seems to > > act alive well enough must be alive, then you also have to say that > > anything that does not seem conscious may just be poorly programmed. > > When you talk about changing substitution levels, you are talking about > different programs. Some levels may be so high-level that the important > and necessary aspects are eliminated and replaced with functions which > fundamentally alter the experience of the simulated mind. Whether or not > this would be noticed depends on the sophistication of the Turing test. > Examination of outward appearance may not even be sufficient. I think Ned > Block had an argument against that you could have a giant state table that > is infinite in size and for any possible question it had the stored > output. Such a program might pass a Turing test, but internally it is > performing only a very trivial computation. If we inspected the code of > this program we could say it has no understanding of individual words, no > complex thought processes, etc. However, most zombies are defined to be > functionally (if not physically) identical rather than merely capable of > passing a some limited test based on external appearances. Zombiehood has nothing to do with external appearances, other than that they are presumed to be the same as a non-zombie. What makes a zombie a zombie is that it lacks interiority. It doesn't matter if it is possible to test it or not, if we call it a zombie, that means that it is a given that it does not have conscious interior experience. All programs are zombies, and all consciousness is more than a program. Craig -- You received this message because you are subscribed to the Google Groups "Everything List" group. To post to this group, send email to [email protected]. To unsubscribe from this group, send email to [email protected]. For more options, visit this group at http://groups.google.com/group/everything-list?hl=en.
