Path: utzoo!attcan!uunet!bu.edu!rpi!zaphod.mps.ohio-state.edu!samsung!cs.utexas.edu!yale!cs.yale.edu!mcdermott-drew From: mcdermott-drew@cs.yale.edu (Drew McDermott) Newsgroups: comp.ai.philosophy Subject: Re: Testing Machine Consciousness Message-ID: <26909@cs.yale.edu> Date: 24 Oct 90 17:02:38 GMT References: <26852@cs.yale.edu> <65780@iuvax.cs.indiana.edu> Sender: news@cs.yale.edu Organization: Yale University Computer Science Dept., New Haven, CT 06520-2158 Lines: 140 Nntp-Posting-Host: aden.ai.cs.yale.edu Originator: dvm@aden.CS.Yale.Edu As usual, my previous posting on Why There is no Other-Minds Problem needs further explanation. David Chalmers (dave@cogsci.indiana.edu) objects thus: In article <26852@cs.yale.edu> mcdermott-drew@cs.yale.edu (Drew McDermott) writes: >Okay, [how] about this phrasing: As granted above, >mental phenomena have three aspects, subjective, behavioral, and >implementational. Why couldn't it simply be the case that the first >one is often or always absent? The only agent I'm sure has subjective >experience is me, so what's my evidence that anyone else has? Yes, this is the right way to state the problem. "Mind" as the term is traditionally used has behavioural, functional, and phenomenological aspects. The "other minds" problem is concerned only with the phenomenological (subjective) aspects. >The problem with this objection is that it misconstrues {\it aspects} >as {\it parts}. This construal would make sense for problems like the >``other-planets problem'': We observe one star with planets and ask >what our evidence is that some or any other stars have any. But >subjective experience is not an appendage of mentation in this way. A >mental event can be experienced from several different angles, >including (although the categories are fairly arbitrary), the >subjective, the behavioral, and the implementational. But the >experiences are all {\it of the same event}. The flaw in your argument is right here -- the rest is just wrapping. Well, I'm not going to acknowledge that my argument is a flaw plus wrapping! I will acknowledge that we have a clash of several different intuitions, and that people are hard to budge from those they're used to. What I want to do is reassure those who think that there's something silly about the other-minds problem that their intuitions are basically healthy. It may be that in many cases -- in particular, the cases where subjective phenomena exist -- the subjective, behavioural, and functional aspects are all aspects of the same event. But this does not imply that in all cases, they must be tied together. Certainly, in *my* case the three aspects go together. Does that allow be to deduce that in your case, or in a robot's case, they must also? Of course not. The coincidence of functional organization, behavioural consequences and subjectivity in some cases is quite compatible, a priori, with the coincidence of functional organization, behavioural consequences and non-subjectivity in others. Analogy time: In certain cases, temperature coincides with the motion of gas molecules -- they are literally two aspects of the same thing. In other cases, temperature exists without the motion of gas molecules. Of course, you may be wanting to make a *claim* that in all cases where at least one of them occur, the three aspects of "mind" will be tied together as different aspects of the same thing. This is a non-trivial claim, whose truth-value may or may not be "true". The non-triviality of this claim is the reason why there is an Other-Minds Problem. (I in fact believe that the claim is true -- more specifically, I believe that the phenomenological is supervenient on the functional. But I can't prove it, at least not easily, and so the Other Minds problem can still be raised.) This restates the problem again pretty well, but fails to convince me it's real. It seems to propose that we could account for all the *observable properties* of subjectivity (or consciousness) and still not be sure we had "really" accounted for consciousness. This is what seems to me to be a preposterously high standard. Suppose we build a robot, and it claims to be subjectively aware. We ask it how it tells the difference between red things and green things, and it says they look different. When we press it, it starts to tell us about qualia. It treats its own decisions as free. Inspection of its blueprints shows that it has the same functional organization as the brain. (We can't do this today, of course.) It's at this point that we hit the intuition that we still couldn't know whether it *really* experienced anything. My claim is that this intuition is empty. When we've accounted for everything, there's nothing left to account for. If necessary, we could even put into the robot the intuition that mere observables are not enough to be sure an entity is subjectively aware. (I don't think it's necessary; I think the intuition is due to faulty education, not wiring. But I could be wrong.) Your phrasing right here concedes that the OMP is really a problem. "Extremely likely" is not good enough. It was always extremely likely that you and other people all have subjective experience. The OMP is "How can you know *for sure*?" As far as I'm concerned, the best answer for now is "You can't, but you can be pretty confident on various inductive grounds". What is the standard? Do we know the theory of evolution "for sure"? Most scientists are considerably annoyed by creationists' refusal to grant that the theory of evolution is as certain as anything ever gets in science. They should also be annoyed by philosophers' attempts to uphold a similar refusal here with respect to a hypothetical future theory of mind. Whence this refusal? I think Chalmers's use of the word "phenomenological" above is quite revealing. This word derives from Kant's notion of "phenomenon," or thing-as-it-appears. Kant and his buddies were obsessed by the distinction between appearance and reality. To our minds this obsession seems quaint. We picture the world as populated by a variety of information-processing systems, some conscious, others simpler. All of them introduce errors and approximations into data, and most manage to cope with these distortions most of the time. We can often come up with a quantitative theory about how close a system is to the truth about a situation. But *this* theory of appearance vs. reality is obviously not what Kant and Descartes were worried about. They were thinking of the situation of a mind that could be absolutely certain only of "appearance," and had to reason back from there to reality. If you take this picture seriously, then a mind isn't a mind unless things "appear" to it in this way. And nothing can "appear" to a robot in this sense, because (a) there's no absolute certainty; and (b) it's not even possible or necessary to single out one part of the robot as the "subject" of these "phenomena." *This,* I think, is what make people nervous about the idea that robots can have minds. All we have to do is junk the whole epistemological framework, and the problem goes away. Consciousness becomes just another part of the world, like photosynthesis. It's perfectly consistent to hold a quite materalist theory of mind, where subjective phenomena are identical to certain material events -- but only *certain* material events, such as ones that occur in a given biochemistry. I believe that for various reasons this is highly implausible, but it is a coherent position. Dave Chalmers (dave@cogsci.indiana.edu) Center for Research on Concepts and Cognition, Indiana University. By the way, I agree with this point. I am emphatically not arguing that a computationalist or functionalist position is correct a priori. It may well turn out (although I doubt it) that consciousness is a biochemical property, or a property of enormous and unintelligible neural nets. It just seems to me that *any* materialist theory is going to be open to the "objection" that there's no way to be *sure* that the objects it predicts are conscious *really* have minds, blah, blah. Drew McDermott mcdermott@cs.yale.edu