crowston@athena.mit.edu (Kevin Crowston) (12/18/89)
I also read Searle's and the Churchland's articles in Scientific American and I'm not sure I understand Searle's argument. Perhaps someone who does can try to explain once more. Searle seems to be saying that the Turing Test is meaningless as a test of understanding because the Chinese Room can pass it, even though the person in the Chinese Room doesn't understand Chinese. But it seems to me that this argument equates the mind and the brain and thus mislocates the thing doing the understanding. I agree that the man in the room doesn't understand Chinese; but I would argue similarly that a computer, as a collection of wires and silicon, or a brain, as a blob of protoplasm, don't understand anything either. I all three cases, the thing doing the understanding is the program not the hardware. Searle acknowledges this argument (it's counterargument c in his article), but answers it by imagining a situation in which the man in the room memorizes the rules, the inbox, etc. He argues that it can't be the rules that do the understanding, since all there is in the room is the man (who we agree doesn't understand Chinese). The part I don't understand is, what difference does it make how the rules are stored? I don't see why it makes a difference if the man memorizes the rules or reads them off a piece of paper. In the latter case, admittedly, you can point to the rule book; but that doesn't mean the rule book doesn't exist in the former case. It seems to me that Searle's second example is really the same example, in which case the argument (that it's the rules that do the understanding, not the man in the room) remains unanswered. I expect the Scientific American Articles will set off another wave of articles; I look forward to the debate. Kevin Crowston
jfw@bilbo.mc.duke.edu (John Whitehead) (12/19/89)
In article <1989Dec18.014229.18058@athena.mit.edu> crowston@athena.mit.edu (Kevin Crowston) writes: >I also read Searle's and the Churchland's articles in Scientific American >and I'm not sure I understand Searle's argument. Perhaps someone who >does can try to explain once more. For a good analysis of this -- and many other similar thought-challenging papers -- check out _The Mind's I_, edited by Douglas Hofstadter and Daniel Dennett. I haven't seen the Sci Am article, but I imagine it is similiar (if not identical) to the one in this book. - John Whitehead
sarge@metapsy.UUCP (Sarge Gerbode) (12/19/89)
In article <1989Dec18.014229.18058@athena.mit.edu> crowston@athena.mit.edu (Kevin Crowston) writes: >Searle seems to be saying that the Turing Test is meaningless as a test of >understanding because the Chinese Room can pass it, even though the >person in the Chinese Room doesn't understand Chinese. But it seems to >me that this argument equates the mind and the brain and thus mislocates >the thing doing the understanding. I agree that the man in the room doesn't >understand Chinese; but I would argue similarly that a computer, as a >collection of wires and silicon, or a brain, as a blob of protoplasm, >don't understand anything either. In all three cases, the thing doing the >understanding is the program not the hardware. On reflection, I don't think you can dispose of the issue that easily by differentiating between the program and the hardware. The program is a schema that describes the electronic state the hardware should be in when the code file is loaded. In a very real sense, then, the shape of the physical machine has been altered by loading the code file, just as much as if you had flipped switches within the machine (as we used to do with the old panel switches). So after the code is loaded, there is actually a different physical machine there, just as much as if one had gone out and bought a different machine. Just because it isn't "hard" (i.e., you can't kick it, and it's easy to change), doesn't mean it isn't a physical entity. -- Sarge Gerbode -- UUCP: pyramid!thirdi!metapsy!sarge Institute for Research in Metapsychology 431 Burgess Drive; Menlo Park, CA 94025
crowston@athena.mit.edu (Kevin Crowston) (12/19/89)
In article <968@metapsy.UUCP> sarge@metapsy.UUCP (Sarge Gerbode) writes: >On reflection, I don't think you can dispose of the issue that easily >by differentiating between the program and the hardware. The program >is a schema that describes the electronic state the hardware should >be in when the code file is loaded. In a very real sense, then, the >shape of the physical machine has been altered by loading the code >file, just as much as if you had flipped switches within the machine >(as we used to do with the old panel switches). So after the code is >loaded, there is actually a different physical machine there, just as >much as if one had gone out and bought a different machine. But even so, the program still exists in both cases, right? Actually, I think you've made a key point here. Loading the software essentially gives you a different machine. But I think this actually supports my argument. Imagine the effect on the "understanding" done by the Chinese room of replacing the person with someone else (assuming that someone can also follow the rules). Now imagine changing the rulebook. In the first case, the Chinese room will be unaffected; in the second, it might change. I would argue that this is further evidence that it's the program not the hardware that matters. Since it could be anyone in the Chinese Room, it shouldn't matter what that person happens to think. > >Just because it isn't "hard" (i.e., you can't kick it, and it's easy >to change), doesn't mean it isn't a physical entity. Actually, this was my point. Software exists, even though you can't point to it. Kevin Crowston
dave@cogsci.indiana.edu (David Chalmers) (12/19/89)
"Programs" do not think. Cognition is not "symbol-manipulation." The "hardware/software" distinction is unimportant for thinking about minds. However: Systems with an appropriate causal structure think. Programs are a way of formally specifying causal structures. Physical systems which implement a given program *have* that causal structure, physically. (Not formally, physically. Symbols were simply an intermediate device.) Physical systems which implement the appropriate program think. -- Dave Chalmers (dave@cogsci.indiana.edu) Concepts and Cognition, Indiana University. "It is not the least charm of a theory that it is refutable" -- Fred
mbb@cbnewsh.ATT.COM (martin.b.brilliant) (12/20/89)
From article <31821@iuvax.cs.indiana.edu>, by dave@cogsci.indiana.edu (David Chalmers)... Slightly edited to make the bones barer: 1. Systems with an appropriate causal structure think. 2. Programs are a way of formally specifying causal structures. 3. Physical systems implement programs. 4. Physical systems which implement the appropriate program think. I take it that (1) is an acceptable definition. Does anybody think it begs the question? The weakest link here may be (2), the supposition that programs can implement any causal structure whatever, even those that do what we call thinking. The software/hardware duality question is semantically resolved by (3). The conclusion is (4), which seems to assert "strong AI." M. B. Brilliant Marty AT&T-BL HO 3D-520 (201) 949-1858 Holmdel, NJ 07733 att!hounx!marty1 or marty1@hounx.ATT.COM After retirement on 12/30/89 use att!althea!marty or marty@althea.UUCP Disclaimer: Opinions stated herein are mine unless and until my employer explicitly claims them; then I lose all rights to them.
kp@uts.amdahl.com (Ken Presting) (12/20/89)
In article <968@metapsy.UUCP> sarge@metapsy.UUCP (Sarge Gerbode) writes: >On reflection, I don't think you can dispose of the issue that easily >by differentiating between the program and the hardware. The program >is a schema that describes the electronic state the hardware should >be in when the code file is loaded. In a very real sense, then, the >shape of the physical machine has been altered by loading the code >file, just as much as if you had flipped switches within the machine >(as we used to do with the old panel switches). So after the code is >loaded, there is actually a different physical machine there, just as >much as if one had gone out and bought a different machine. This is e very good point, and often overlooked. The physical instantiation of data immensely complicates the concept of "symbol system". When machines were built from gears and axles, it was trivial to distinguish symbols from mechanisms. Symbols are part of a language, are written or spoken, and (most importantly) have no mechanical functions. But communication and computing devices blur the distinction. In these machines, an instance of a symbol (a charge, a current pulse, a switch) has a mechanical role in the operation of the device. The first problem that arises is how to distinguish symbol manipulation systems from other machines. What makes a symbol "explicit"? The clearest case of explicit symbols is printed text in a human language, but we need to resolve hard cases. One hard case is microcode or firmware. The hardest case is probably neural nets. Conclusion: No definition of "symbol manipulation system" which uses the term "explicit" will be of much help (until "explicit" is defined).
lee@uhccux.uhcc.hawaii.edu (Greg Lee) (12/20/89)
From article <6724@cbnewsh.ATT.COM>, by mbb@cbnewsh.ATT.COM (martin.b.brilliant): >From article <31821@iuvax.cs.indiana.edu>, by dave@cogsci.indiana.edu >(David Chalmers)... > >Slightly edited to make the bones barer: > >1. Systems with an appropriate causal structure think. >2. Programs are a way of formally specifying causal structures. >3. Physical systems implement programs. >4. Physical systems which implement the appropriate program think. > >I take it that (1) is an acceptable definition. Does anybody think it >begs the question? ... This and similar discussions have seemed to revolve around an equivocation between theories about how a thing works and how the thing does work. 1-4 invite this equivocation in several ways, consequently they do not serve to clarify. `causal' and `structure' have to do with theory-making -- we attribute cause/effect and structure to something in our efforts to understand it. So 1. in effect says that we can now understand, or will come to understand, how people think by making a theory involving cause and structure. If the former, it's false; if the latter, it does beg the question. If `program' in 3. is read as `theory' and `physical system' read as thing about which the theory is made, which is the best I can make of it, 3. is a generalization of 1. -- we can make (good) theories about things. As applied to human thought, and interpreted as a prediction, it likewise begs the question. Greg, lee@uhccux.uhcc.hawaii.edu
kp@uts.amdahl.com (Ken Presting) (12/20/89)
In article <6724@cbnewsh.ATT.COM> mbb@cbnewsh.ATT.COM (martin.b.brilliant) writes: >From article <31821@iuvax.cs.indiana.edu>, by dave@cogsci.indiana.edu >(David Chalmers)... > > 1. Systems with an appropriate causal structure think. > 2. Programs are a way of formally specifying causal structures. > 3. Physical systems implement programs. > 4. Physical systems which implement the appropriate program think. > >I take it that (1) is an acceptable definition. Does anybody think it >begs the question? I don't think so. Presumably, humans think because of the way we're built, and the mechanical/chemical/electrical structure determines the causal structure of our brains. >The weakest link here may be (2), the supposition that programs can >implement any causal structure whatever, even those that do what we >call thinking. Agreed. The multi-body problem of astrophysics is a clear case of a causal system which cannot be precisely represented by an algorithm. But the argument could succeed with a weaker version of 2, IF we could figure out which causal structures are relevant to thought >The software/hardware duality question is semantically resolved by (3). This is problematic. Harnad's "symbol grounding problem" (and some of Searle's objections, I think) point out the difficulty of claiming that some object "thinks" strictly on the basis of its internal operation, or even on the basis of it's outputs. Harnad would want to know how the symbols found in the output are grounded, while Searle might claim that the machine *simulated* thinking, but did not itself *think*. I agree that the correct resolution of the software/hardware duality can only be resolved by the concept of implementation used in (3). I'm just repeating a familiar (but important) theme.
miken@wheaties.ai.mit.edu (Michael N. Nitabach) (12/20/89)
In article <5767@uhccux.uhcc.hawaii.edu>, lee@uhccux.uhcc.hawaii.edu (Greg Lee) says: >`causal' and >`structure' have to do with theory-making -- we attribute >cause/effect and structure to something in our efforts to >understand it. This view of the fundamental nature of causation derives from a particular metaphysical tradition, beginning with the British Empiricists, e.g. Locke and Hume. This is the view that causation is not an aspect of the world which our mentality can recognize, but rather a schema which our mind imposes on events with appropriate spatiotemporal relations. A conceptually opposite--Realist--stance would be that causation exists as an actual attribute of certain pairs of physical events. Greg's argument in that posting rests on a particular metaphysical assumption, and not on a simple matter of definition or brute fact. Mike Nitabach
ele@cbnewsm.ATT.COM (eugene.l.edmon) (12/20/89)
In article <31821@iuvax.cs.indiana.edu> dave@cogsci.indiana.edu (David Chalmers) writes: >Systems with an appropriate causal structure think. Could you elaborate on this a bit? -- gene edmon ele@cbnewsm.ATT.COM
sm5y+@andrew.cmu.edu (Samuel Antonio Minter) (12/20/89)
1988:11:19:05:13 SFT
Couldn't you use the Chinese room analogy to prove that Humans don't
truly understand either. In this case the matter/energy in the human body
take the role of the man in the room and all his stacks of cards, while
the basic laws of physics take the role of the instuction book. After all
just as the instruction book tells the man what to do thus simulating a room
which understands Chinese, the laws telling how various atoms, electrons,
energy fields, etc. interact with each other "instruct" the matter and energy
of the human body how to simulate intelligent behavior. Maybe even
understanding Chinese! Is there an error in this argument that I'm missing.
If there isn't then it is a more powerful counter argument than the
agument "of course the man dosn't understand, but the whole room does."
1988:11:19:05:19 SFT
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
~ |\ You have just read a message from Abulsme Noibatno Itramne! ~
~ | \ ~
~ | /\ E-Mail@CMU: sm5y+@andrew.cmu.edu <Fixed Width Fonts!> ~
~ |/ \ S-Mail@CMU: Sam Minter First the person ~
~ |\ /| 4730 Centre Ave., #102 next ~
~ | \/ | Pittsburgh, PA 15213 the function!! ~
~ | / | ~
~ |/ | <-----An approximation of the Abulsme symbol ~
~ ~
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~mmcg@bruce.OZ (Mike Mc Gaughey) (12/20/89)
sm5y+@andrew.cmu.edu (Samuel Antonio Minter) [20 Dec 89 05:20:12 GMT]: > 1988:11:19:05:13 SFT > > Couldn't you use the Chinese room analogy to prove that Humans don't > truly understand either. In this case the matter/energy in the human body > take the role of the man in the room and all his stacks of cards, while > the basic laws of physics take the role of the instuction book. After all No - this only proves that the laws of physics don't think (just as the man the room didn't understand). The total system behavior (i.e of a brain) is that of an entity which _does_ understand the concepts represented by the symbols being manipulated. Mike. -- Mike McGaughey ACSNET: mmcg@bruce.cs.monash.oz "You have so many computers, why don't you use them in the search for love?" - Lech Walesa :-)
lee@uhccux.uhcc.hawaii.edu (Greg Lee) (12/21/89)
From article <5610@rice-chex.ai.mit.edu>, by miken@wheaties.ai.mit.edu (Michael N. Nitabach): >In article <5767@uhccux.uhcc.hawaii.edu>, lee@uhccux.uhcc.hawaii.edu >(Greg Lee) says: > >>`causal' and >>`structure' have to do with theory-making -- we attribute >>cause/effect and structure to something in our efforts to >>understand it. > >This view of the fundamental nature of causation derives from a particular >metaphysical tradition, beginning with the British Empiricists, e.g. Locke >and Hume. This is the view that causation is not an aspect of the world >which our mentality can recognize, but rather a schema which our mind imposes >on events with appropriate spatiotemporal relations. ... No, that's not quite the view I expressed. In saying that causation is something we attribute in theory-making, I do not need to go so far as to say "causation is not an aspect of the world". And I don't. It may be, in the case of very good theories, that it is reasonable to confound what the theory says about a thing with the thing itself, or to take the theory to be a discovery rather than an invention. But in the case of not-so-good theories, where there is some doubt as to whether what the theory says is a cause is indeed a cause, confusing the theory with what it describes ought to be avoided. In the present discussion, we are dealing with not-so-good theories. Surely there's no one who is going to try to defend the view that one should never distinguish between a theory and what that theory purports to describe. Greg, lee@uhccux.uhcc.hawaii.edu
dejongh@peirce.cis.ohio-state.edu (Matt Dejongh) (12/21/89)
In article <6724@cbnewsh.ATT.COM> mbb@cbnewsh.ATT.COM (martin.b.brilliant) writes: >From article <31821@iuvax.cs.indiana.edu>, by dave@cogsci.indiana.edu >(David Chalmers)... > >Slightly edited to make the bones barer: > > 1. Systems with an appropriate causal structure think. > >I take it that (1) is an acceptable definition. Does anybody think it >begs the question? I do. What is "an appropriate causal structure?" Give me a definition and an example. matt ---------------------------------------------------------------------------- Matt DeJongh | Laboratory for Artificial Intelligence Research | Department of Computer and Information Sciences dejongh@cis.ohio-state.edu | The Ohio State University, Columbus, Ohio 43210 -=- ---------------------------------------------------------------------------- Matt DeJongh | Laboratory for Artificial Intelligence Research | Department of Computer and Information Sciences dejongh@cis.ohio-state.edu | The Ohio State University, Columbus, Ohio 43210
ladkin@icsib (Peter Ladkin) (12/21/89)
In article <5610@rice-chex.ai.mit.edu>, miken@wheaties (Michael N. Nitabach) writes: >This view of the fundamental nature of causation derives from a particular >metaphysical tradition, beginning with the British Empiricists, e.g. Locke >and Hume. This is the view that causation is not an aspect of the world >which our mentality can recognize, but rather a schema which our mind imposes >on events with appropriate spatiotemporal relations. this is hardly locke's view, and barely that of hume. locke rather strongly held that primary qualities of matter caused secondary qualities. this causation was not a product of anything like a mental schema. and `events with appropriate spatiotemporal relations' were not the only inhabitants of the physical world. you might also count bishop berkeley in with the other two, and for him causation was `in the world'. of course, it got there by being in the intention of a god, for him. all this is well-known and well-researched material. so much for summarising the views of the british empiricists. peter ladkin
sm5y+@andrew.cmu.edu@canremote.uucp (sm5y+@andrew.cmu.edu) (12/21/89)
From: sm5y+@andrew.cmu.edu (Samuel Antonio Minter)
Orga: Carnegie Mellon, Pittsburgh, PA
1988:11:19:05:13 SFT
Couldn't you use the Chinese room analogy to prove that Humans
don't truly understand either. In this case the matter/energy in the
human body take the role of the man in the room and all his stacks of
cards, while the basic laws of physics take the role of the
instuction book. After all just as the instruction book tells the
man what to do thus simulating a room which understands Chinese, the
laws telling how various atoms, electrons, energy fields, etc.
interact with each other "instruct" the matter and energy of the
human body how to simulate intelligent behavior. Maybe even
understanding Chinese! Is there an error in this argument that I'm
missing.
If there isn't then it is a more powerful counter argument than
the agument "of course the man dosn't understand, but the whole room
does."
1988:11:19:05:19 SFT
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
~ |\ You have just read a message from Abulsme Noibatno Itramne! ~
~ | \ ~
~ | /\ E-Mail@CMU: sm5y+@andrew.cmu.edu <Fixed Width Fonts!> ~
~ |/ \ S-Mail@CMU: Sam Minter First the person ~
~ |\ /| 4730 Centre Ave., #102 next ~
~ | \/ | Pittsburgh, PA 15213 the function!! ~
~ | / | ~
~ |/ | <-----An approximation of the Abulsme symbol ~
~ ~
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
---
* Via MaSNet/HST96/HST144/V32 - UN AI
* Via Usenet Newsgroup comp.aidave@cogsci.indiana.edu (David Chalmers) (12/21/89)
gene edmon writes: >...David Chalmers writes: >>Systems with an appropriate causal structure think. > >Could you elaborate on this a bit? Well, seeing as you ask. The basic idea is that "it's not the meat, it's the motion." At the bottom line, the physical substance of a cognitive system is probably irrelevant -- what seems fundamental is the pattern of causal interactions that is instantiated. Reproducing the appropriate causal pattern, according to this view, brings along with it everything that is essential to cognition, leaving behind only the inessential. (Incidentally, I'm by no means arguing against the importance of the biochemical or the neural -- just asserting that they only make a difference insofar as they make a *functional* difference, that is, play a role in the causal dynamics of the model. And such a functional difference, on this view, can be reproduced in another medium.) And yes, of course this is begging the question. I could present arguments for this point of view but no doubt it would lead to great complications. Just let's say that this view ("functionalism", though this word is a dangerous one to sling around with its many meanings) is widely accepted, and I can't see it being unaccepted soon. The main reason I posted was not to argue for this view, but to delineate the correct role of the computer and the program in the study of mind. The other slightly contentious premise is the one that states that computers can capture any causal structure whatsoever. This, I take it, is the true import of the Church-Turing Thesis -- in fact, when I look at a Turing Machine, I see nothing so much as a formalization of the notion of causal system. And this is why, in the philosophy of mind, "computationalism" is often taken to be synonymous with "functionalism". Personally, I am a functionalist first, but accept computationalism because of the plausibility of this premise. Some people will argue against this premise, saying that computers cannot model certain processes which are inherently "analog". I've never seen the slightest evidence for this, and I'm yet to see an example of such a process. (The multi-body problem, by the way, is not a good example -- lack of a closed-form solution does not imply the impossibility of a computational model.) Of course, we may need to model processes at a low, non-superficial level, but this is not a problem. The other option for those who argue against the computational metaphor is to say "yes, but computation doesn't capture causal structure *in the right way*". (For instance, the causation is "symbolic", or it has to be mediated by a central processor.) I've never found much force in these arguments. -- Dave Chalmers (dave@cogsci.indiana.edu) Concepts and Cognition, Indiana University. "It is not the least charm of a theory that it is refutable" -- Fred
kp@uts.amdahl.com (Ken Presting) (12/22/89)
In article <31945@iuvax.cs.indiana.edu> dave@cogsci.indiana.edu (David Chalmers) writes: >The other slightly contentious premise is the one that states that computers >can capture any causal structure whatsoever. This, I take it, is the true >import of the Church-Turing Thesis -- in fact, when I look at a Turing >Machine, I see nothing so much as a formalization of the notion of causal >system. . . . > . . . Some people will argue against this premise, saying that >computers cannot model certain processes which are inherently "analog". I've >never seen the slightest evidence for this, and I'm yet to see an example of >such a process. (The multi-body problem, by the way, is not a good example -- >lack of a closed-form solution does not imply the impossibility of a >computational model.) Of course, we may need to model processes at a low, >non-superficial level, but this is not a problem. What makes the multi-body problem a counter-example is not just the fact that the problem has no closed-form solution, but the chaotic nature of the mechanical system. In a chaotic system, an arbitrarily small change in initial conditions will over time produce an arbitrarily *large* difference in subsequent states. It is true that for any physical system, a numerical solution of the differential equations can generate a prediction of the future state with an error as small as desired. But if a numeric model of a physical system is to run in real time (as needed for the Turing test), or just proportional to real time, then there will be a fixed minimum error in the transitions from state to state. The error may be reduced by using faster processors or faster algorithms, but for a given combination of processor, algorithm, and lag behind real-time, there must be a limit on the number of terms evaluated, and a minimum error. So at the end of the first time slice, there will be a finite difference between the state of the real system and the calculated state of the model. The real system will proceed chaotically (as will the model), to amplify the discrepancy in initial state and each subsequent state until (sooner or later) the real system will be in a state which diverges from the state of the model by any amount (up to the size of the system). That was a rough sketch of a proof that not all causal systems can be modeled by programs. Let me add a plausibility argument, so that the claim will not seem counter-intuitive. What makes the analog causal system different from the algorithm is that each state of the analog system encodes an infinite amount of information. This holds even for electrons bound into the quantum orbitals of an atom. There is a finite number of electrons, and transitions between levels are discrete, but there are infininitely many energy levels (most of them very close together). Of course, a processor has finitely many states, and encodes a finite amount of information. In addition, an analog system need not time-slice its calculations. It can make infinitely many transitions in any interval. A Turing machine would need to have both infinitely many states and run arbitrarily fast to prescisely match an analog system using a numerical algorithm. Now, the lack of precision is insignificant for many analog systems in many applications, where the error is constant or grows slowly. But in a chaotic system, the error in the model can grow very rapidly. If the numerical model cannot be allowed to lag arbitrarily far behind real-time (thus trading time for memory) then the amplification of error will make the model useless. The "solutions in closed form" of rational mechanics gurantee that that a numerical model for an analog system will have a rapidly computable algorithm (often a polynomial). So the fact that the multi-body problem has no solution in closed form is relevant, but that's not the whole story. More important is chaotic amplification of initial differences. This does *not* show that strong AI is impossible with algorithms. There is no way to know whether intelligence requires chaos. But the brain is certainly complicated enough to be as chaotic as fluid flow. Modeling human behavior may well have to deal with this problem. BTW, I'd like to know if anyone has heard this kind of argument before in connection with AI. It must be old hat in simulation circles.
dhw@itivax.iti.org (David H. West) (12/22/89)
In article <ebfl02Ef76Hs01@amdahl.uts.amdahl.com> kp@amdahl.uts.amdahl.com (Ken Presting) writes: |What makes the multi-body problem a counter-example is not just the fact |that the problem has no closed-form solution, but the chaotic nature of |the mechanical system. | |In a chaotic system, an arbitrarily small change in initial conditions |will over time produce an arbitrarily *large* difference in subsequent |states. [...] |That was a rough sketch of a proof that not all causal systems can be |modeled by programs. Let me add a plausibility argument, so |that the claim will not seem counter-intuitive. | |What makes the analog causal system different from the algorithm is that |each state of the analog system encodes an infinite amount of information. Real intelligent systems (e.g. humans) function quite successfully at a finite temperature despite the influence of thermal fluctuations (Brownian motion), which cause finite random perturbations of everything. A finite system embedded in a thermal environment cannot encode an infinite amount of information. This would seem to indicate that your argument has no bearing on what may be necessary for intelligence, at least over a time-scale short enough that the physical embodiment of the intelligence is not disrupted by thermal fluctuations, whether or not these are chaotically amplified. -David West dhw@iti.org
kp@uts.amdahl.com (Ken Presting) (12/22/89)
In article <4689@itivax.iti.org> dhw@itivax.UUCP (David H. West) writes: >Real intelligent systems (e.g. humans) function quite successfully >at a finite temperature despite the influence of thermal >fluctuations (Brownian motion), which cause finite random perturbations >of everything. A finite system embedded in a thermal environment >cannot encode an infinite amount of information. A finite *digital* system is limited in the density of its states by thermal (and other) fluctuations. You are right to point out that the brain is limited in its computational power by this fact. I think that the most productive and interesting approach to cognitive science (if not artificial intelligence) must take advantage of the brain's limitations; CS wants to learn about how the brain computes and the algorithms it follows. But the brain viewed as a computational system is different from the brain viewed as a causal system. The same distinction applies to electronic computers - there are plenty of hardware failure modes to complicate the causal description of the system which are irrelevant to the computational description. If we want to claim that we've accurately modeled the computational power of the brain by demonstrating that our model is faithful to the causal interactions in the brain, we're stuck with modeling the details (including thermal motion) down to whatever level a critic might demand. The word "encode" might be ill-chosen for causal systems. If position & velocity (et al) encode anything, the code would have infinitely many "symbols," one for each state of the system. David continues: >This would seem to indicate that your argument has no bearing on >what may be necessary for intelligence, (...) This is true. Certainly there is no reason to believe that infinite processing power (or infinite anything else) is necessary for intelligence. What I think we can conclude is that a familiar "safety net" argument for AI doesn't quite work (different from the original): (1) The brain is a causal system that thinks (2) We can model causal systems with numerical methods (3) Therefore, we can make a model that thinks . Searle (I believe) would object that we would get a simulation of thinking, not the real thing. I'm objecting that it is physically impossible to make an adequate model. If we could let the model crank as long as it needed before responding (ie relax the real-time constraint) then my argument would not hold. The model would eventually converge close enough to the real system to be indistinguishable. But that won't pass the Turing test.
dove@portia.Stanford.EDU (Dav Amann) (12/22/89)
Here we go. Recently, several posters have debated the concept of thinking machines without explicitly discussing thought. We all know that we think but I doubt that many individuals can articulate what it means to think. Does, for example, thinking imply conciousness? In other words, do things think without being aware that they are thinking? If thought is merely the process of solving problems, then answers become obvious. Solving problems does not imply conciousness. My Mac solves all sorts of problems for me, but it certainly does not spend its off hours contemplating the Eternal Void. (At least, I don't think that it does.) However, I believe that when individuals talk about thought they imply some sort of conciousness, or awareness. When we say a machine that thinks, we mean a machine that understands, reasons, draws conclusions, learns, and is somehow aware. Thus when Searle talks of the Chinese room, he is questioning the awareness of the machine rather than its imitation of reasoning processes. (At least, as far as I can tell.) I believe that the problem of conciousness comes from a choice of metaphysics. Most of us in the Western world are disciples of the ancient Greek metaphysician, Democritus. Democritus, you'll recall, was the philosopher who theorized that all of reality was made of atoms. Understand the atoms and you'll understand reality. Mathematics, physics, logic all went a long way towards cementing this mind set. The whole is equal to the sum of its parts. Understand the parts, and you'll understand the whole. When this viewpoint is applied to a theory of mind, you get a lot of folks saying, "All that's in there is a bunch of neurons firing in certain patterns. All that's happening is the exchange of neurochemicals. Understand those, and you'll understand the mind." Well, perhaps not. Somehow, intuitively speaking, I think that the mind is more than the firing of neurons, though it does seem to encompass the firing of neurons. There's more there, I say, than the exchange of certain neurochemicals. Plato knew more about the human mind than me and he knew much less about the construction of the brain. Perhaps this intuition explains the vigorous defense of Cartesian dualism since Descartes without very much empirical evidence. Lately, however, one of the newer sciences has been breaking away from Democritus. Biology discusses and understands more and more about the individual cell, yet they find it harder and harder to explain the relationship between cells within the pretext of the individual cell. Etymologysts understand a lot about termites but they cannot explain why five termites together will build arches the Romans would be proud of. The whole is more than the sum of its parts. Perhaps the mind is much the same way. Understanding the switches and chemicals inside of the brain may in some way add to the knowledge of our selves, but I don't think that it can ever fully explain our selves and our consciousness. So the question arises, How can we understand ourselves or our consciousness? How can we tell whether a machine thinks? To these questions I profess my ignorance, but I do not think that any method which only looks at the parts of the brain will accomplish that lofty goal. Dav Amann dove@portia.stanford.edu
dmocsny@uceng.UC.EDU (daniel mocsny) (12/22/89)
In article <7853@portia.Stanford.EDU>, dove@portia.Stanford.EDU (Dav Amann) writes: > Recently, several posters have debated the concept of thinking > machines without explicitly discussing thought. We all know that we > think but I doubt that many individuals can articulate what it means > to think. I doubt that *any* individuals can articulate what thinking is, any more than a fish can comprehend that water is wet. **** One fringe benefit of being largely ignorant of both metaphysics and AI is being blissfully unaware of all the ways I am not supposed to think. Exercising this freedom tonight I had the idea (probably not original) that an anology may exist between the questions: "Can machines think?" and "Are viruses alive?" A virus is simply a protein coat around some genetic material. If it is floating in a sterile environment it can't metabolize any available nutrients the way bacteria do. It doesn't reproduce, it doesn't respond to stimuli, it is an inert speck of macromolecules (albeit a highly organized speck). In effect, the virus doesn't possess sufficient causal structure to qualify as "life" in a disorganized environment. However, place the virus in the right environment (i.e., a host) and it mechanistically attaches itself to receptors on host cells by passively responding to intermolecular attractive forces. Its protein coat dissolves mechanistically, and its genetic material enters the host cell and alters its metabolism. As a result, the host cell generates many copies of the original virus, then ruptures. In the environment of the host, the virus sure looks like life. Outside the host it does not. Therefore, whether we choose to consider the virus "alive" depends on our frame of reference. To an alien biologist from another planet who has no knowledge of suitable host organisms, the information in the virus' structure would be so much nonsense. In a sense, the virus requires a host to "ground" the information represented in its internal structure. In other words, a virus is like "life" with a removable context, apart from which it becomes "non-life". Life with an on-off switch, as it were. Now consider a very complex symbol-processing system, say one that could make a good showing in the "Chinese Room", and moreover, one that implements statistical pattern-matching algorithms giving it some "learning" ability. Now, when interacting with chinese speakers, the SPS could give a fairly convincing imitation of "thinking". But is it thinking? The Chinese Room argument is supposed to demonstrate conclusively that the SPS does not think. Assume Searle is correct, and the SPS does not think. In what way, then, does its failure to think constrain its possible behaviors? Let us view the relationship between a symbol-processing system and human minds as analogous to the relationship between a (benign) virus and its host. Apart from the host, the virus is nothing, but while interacting with the host it satisfies enough of the requirements of "life" to be essentially alive. The virus effectively abstracts, or mirrors, some of the essential causal structure of the host, so that in combination with the host it can display highly complex, almost purposive, behaviors. So too the SPS abstracts or mirrors some of the essential causal structure of the human mind that created it. It does not abstract enough of that structure to "stand on its own", i.e., if we merely print out its list of machine instructions we do not see anything vaguely resembling "thinking", and neither can the SPS even imitate thinking in the wrong environment. And yet, while interacting with the appropriate "hosts", the SPS is capable of arbitrarily complex behaviors, perhaps indistinguishable (in an arbitrarily complex SPS) from the behavior of the minds that created it. So perhaps a useful way to view symbol processing systems is not as "thinking systems", but rather, "mind viruses". (My apologies to Dr. Rapaport if he has already published a series of papers exploring this very notion!) :-) > Etymologysts understand a lot about termites but they cannot explain > why five termites together will build arches the Romans would be > proud of. The whole is more than the sum of its parts. If we view "mind" as an emergent property, or epiphenomenon, of "brain", then doesn't that mean we have no way to point to any tangible structure in the brain and say "this produces mind" or "that is mind"? (Because an emergent property, by definition, has no tidy basis in the structure of any part; it only emerges when all the parts get together. This may, of course, be only an artifact of our conceptual deficiency.) Or in other words, perhaps we have no way even in principle to elicit the causal mechanisms that give rise to mind? The flip side to that argument is, of course, that we have no way to arbitrarily restrict the underlying causal mechanisms that could give rise to "mind" as an epiphenomenon. Or even which causal aspects of the brain give rise to "mind". Why, then, couldn't "mind" just as well be an epiphenomenon of (sufficiently complex) "program"? I.e., if we can't say just how the brain gives rise to mind, how can we be so sure programs can't do it to? I don't see how the Chinese Room addresses this at all. Dan Mocsny dmocsny@uceng.uc.edu
dhw@itivax.iti.org (David H. West) (12/22/89)
In article <3185@uceng.UC.EDU> dmocsny@uceng.UC.EDU (daniel mocsny) writes: >One fringe benefit of being largely ignorant of both metaphysics and >AI is being blissfully unaware of all the ways I am not supposed to >think. Exercising this freedom tonight I had the idea (probably not >original) that an anology may exist between the questions: "Can >machines think?" and "Are viruses alive?" I rather the like the quote attributed to Dijkstra (can anyone provide a hard citation for this?): "The question of whether a machine can think is like the question of whether a submarine can swim". >So perhaps a useful way to view symbol processing systems is not as >"thinking systems", but rather, "mind viruses". (My apologies to Dr. >Rapaport if he has already published a series of papers exploring this >very notion!) :-) Richard Dawkins has. His word for it was "meme", a conflation (I take it) of "memory" and "gene". -David West dhw@iti.org
jgk@osc.COM (Joe Keane) (12/23/89)
In article <013Y02gH77ra01@amdahl.uts.amdahl.com> kp@amdahl.uts.amdahl.com (Ken Presting) writes: >If we want to claim that we've accurately modeled the >computational power of the brain by demonstrating that our model is >faithful to the causal interactions in the brain, we're stuck with >modeling the details (including thermal motion) down to whatever level a >critic might demand. This is not true. The concept of an exact simulation may be theoretically interesting, but has nothing to do with AI. The important question is, what level of simulation is necessary to pass the Turing test. I think it is necesary to model different parts of the brain, but not to worry about the exact distribution of potassium ions. If you believe that it must be exact down to the incoming cosmic rays, i don't know what to say. >I'm objecting that it is physically >impossible to make an adequate model. If we could let the model crank >as long as it needed before responding (ie relax the real-time constraint) >then my argument would not hold. The model would eventually converge >close enough to the real system to be indistinguishable. But that won't >pass the Turing test. If i'm not mistaken, you're arguing that it takes an arbitrarily large amount of computation to do an acceptable simulation. I don't believe this at all. Let's perform a thought experiment. I'll make a copy of the universe (that's why it's a thought experiment) and then remove an electron in your clone's brain. This should not cause a severe change in your clone, but since the brain is chaotic, this may cause him to do something differently than the original you. However, despite the fact that you and your clone are now different, there is no way to tell which is the `real' you.
dave@cogsci.indiana.edu (David Chalmers) (12/23/89)
Ken Presting writes: >What makes the multi-body problem a counter-example is not just the fact >that the problem has no closed-form solution, but the chaotic nature of >the mechanical system. Chaos is only a problem if we need to model the behaviour of a particular system over a particular period of time exactly -- i.e., not just capture how it might go, but how it *does* go. This isn't what we're trying to do in cognitive science, so it's not a problem. We can model the system to a finite level of precision, and be confident that what we're missing is only random "noise." So while we won't capture the exact behaviour of System X at 3 p.m. on 12/22/89, we'll generate equally plausible behaviour -- in other words, how the system *might* have gone, if a few unimportant random parameters had been different. This leads to a point which is a central tenet of functionalism -- you don't need to capture a system's causal dynamics exactly, but only at a certain level of abstraction. Which level of abstraction? Well, this is usually specified teleologically, depending on what you're trying to capture. Usually, it's a level of abstraction that captures plausible input/output relationships. Anything below this, we can consider either implementational detail, or noise. Just what this level of abstraction is, of course, is a matter of some debate. The most traditional functionalists, including the practitioners of "symbolic" AI, believe that you may go to a very high level of abstraction before missing anything important. The move these days seems to be towards a much less abstract modelling of causal dynamics, in the belief that what goes on at a low level (e.g. the neural level) makes a fundamental difference. (This view is sometimes associated with the name "eliminative materialism", but it's really just another variety of functionalism. Even at the neural level, what we're trying to capture are causal patterns, not substance.) >What makes the analog causal system different from the algorithm is that >each state of the analog system encodes an infinite amount of information. Arguable. My favourite "definition" of information is due to Bateson, I think (no endorsement of Bateson's other views implied): "Information is a difference that makes a difference." An infinite number of bits may be required to descibe the state of a system, but in any real-world system, all of these after a certain point will not make any difference at all, except as random parameter settings. (The beauty of Bateson's definition is that the final "difference" depends on our purposes. If we wanted a precise simulation of the universe, these bits would indeed be "information". If we want a cognitive model, they're not.) Incidentally, you can concoct hypothetical analog systems which contain an infinite amount of information, even in this sense -- by coding up Chaitin's Omega for instance (and thus being able to solve the Halting Problem, and be better than any algorithm). In the real world, quantum mechanics makes all of this irrelevant, destroying all information beyond N bits or so. Happy Solstice. -- Dave Chalmers (dave@cogsci.indiana.edu) Concepts and Cognition, Indiana University. "It is not the least charm of a theory that it is refutable"
mbb@cbnewsh.ATT.COM (martin.b.brilliant) (12/24/89)
In article <32029@iuvax.cs.indiana.edu> dave@cogsci.indiana.edu (David Chalmers) writes: >Chaos is only a problem if we need to model the behaviour of a particular >system over a particular period of time exactly -- i.e., not just capture >how it might go, but how it *does* go. This isn't what we're trying to do >in cognitive science, so it's not a problem. We can model the system to >a finite level of precision, and be confident that what we're missing is >only random "noise." ..... I second that. The goal of AI is not to model a particular mind, but to create a mind. One thing we know from experience about minds - which is reinforced by the argment based on chaos - is that two minds never think exactly alike. That would almost prove that if we modeled a given mind exactly, we would NOT have created a mind, because a REAL mind never duplicates another mind. To prove we have created a mind, we have to have one that does not exactly model another. The chaos argument proves that if we create a mind, we will automatically meet that requirement. How fortunate! M. B. Brilliant Marty AT&T-BL HO 3D-520 (201) 949-1858 Holmdel, NJ 07733 att!hounx!marty1 or marty1@hounx.ATT.COM After retirement on 12/30/89 use att!althea!marty or marty@althea.UUCP Disclaimer: Opinions stated herein are mine unless and until my employer explicitly claims them; then I lose all rights to them.
sarge@metapsy.UUCP (Sarge Gerbode) (12/26/89)
In article <1989Dec19.061822.27585@athena.mit.edu> crowston@athena.mit.edu (Kevin Crowston) writes: >>[After object] code is loaded, there is actually a different >>physical machine there, just as much as if one had gone out and >>bought a different machine. >But even so, the program still exists in both cases, right? Good question. What *is* a "program", anyway? The ascii source characters, taken as an aggregate? The machine-language code, as a sequence of octal or hex characters? The magnetic patterns on the disc? The electronic patterns in RAM when the programis loaded? Or is it, as I suspect, the detailed *concept* the programmer had in mind when he wrote the source code? Perhaps the program (or, if you will, the overall algorithm) is a *possibility* that can be actualized (implemented) in a variety of ways. This possibility exists in the mind of a conscious being as the concept called "the program". Without the concept, you would not have a "program" but a mere pattern of electronic whatevers. -- Sarge Gerbode -- UUCP: pyramid!thirdi!metapsy!sarge Institute for Research in Metapsychology 431 Burgess Drive; Menlo Park, CA 94025
sarge@metapsy.UUCP (Sarge Gerbode) (12/26/89)
In article <24Yy02PR76bt01@amdahl.uts.amdahl.com> kp@amdahl.uts.amdahl.com (Ken Presting) writes: >In article <968@metapsy.UUCP>sarge@metapsy.UUCP (Sarge Gerbode) writes: >>On reflection, I don't think you can dispose of the issue that easily >>by differentiating between the program and the hardware. The program >>is a schema that describes the electronic state the hardware should >>be in when the code file is loaded. In a very real sense, then, the >>shape of the physical machine has been altered by loading the code >>file, just as much as if you had flipped switches within the machine >>(as we used to do with the old panel switches). So after the code is >>loaded, there is actually a different physical machine there, just as >>much as if one had gone out and bought a different machine. >This is a very good point, and often overlooked. The physical >instantiation of data immensely complicates the concept of "symbol >system". >When machines were built from gears and axles, it was trivial to >distinguish symbols from mechanisms. Symbols are part of a language, >are written or spoken, and (most importantly) have no mechanical >functions. But communication and computing devices blur the >distinction. In these machines, an instance of a symbol (a charge, a >current pulse, a switch) has a mechanical role in the operation of >the device. I may have a somewhat radical viewpoint on this, but to me a symbol is defined as such by the intention of the conscious being using it. A symbol is a perceivable or detectable entity that is used to direct attention to a particular reality or potential reality. Charges, current pulses, etc., are rightly regraded as symbols only to the extent that they are intended (ultimately) to be comprehended by some sort of conscious entity as indicating certain realities (or potential relaities). In the absence of such intentions, they are not symbols but mere charges, current pulses, etc. Of course, things can be decoded without being *intended* to be so decoded. Scientists are continually decoding (understanding) elements of the physical universe. But these elements are (rightly) not thought of as symbols because (unless one thinks of the universe as a communication from God) they are not intended to be decoded in a particular way. -- Sarge Gerbode -- UUCP: pyramid!thirdi!metapsy!sarge Institute for Research in Metapsychology 431 Burgess Drive; Menlo Park, CA 94025
bsimon@stsci.EDU (Bernie Simon) (12/28/89)
I would like to make a few point that seem clear to me, but apparently aren't clear to others in this discussion. 1) All physical objects are not machines. For example, stones, clouds, flowers, butterflies, and people are not machines. This should be obvious, but some people use the word machine to include all physical objects. Not only is this contrary to ordinary usage, it obscures an important distinction between what is an artifact and what is not. 2) Not all machines are computers. Lamps, screwdrivers, and cars are not computers. 3) There are some activities which can be performed by physical objects and machines which cannot be peformed by computers. Birds can fly and airplanes can fly, but computers cannot fly. Of course, a computer can control an airplane, but this misses the distinction I am trying to make. The distinction is that all computers, as computers, are equivalent to Turing machines. If the computers performs some other activity during its operation than executing a program (for example, flying) it is because the machine which contains the computer is capable of the activity (as airplanes are capable of flying). 4) The simulation of a physical activity by a computer cannot be identified with the physical activity. A computer running a flight simulation program is not flying. 5) Hence, while it may be possible to build a machine that thinks, it does not follow that it will be possible to build a computer that thinks, as not all physical activities can be performed by computers. 6) While there are good reasons to believe that thinking is a physical activity, there are no good reasons for believing that thinking is the execution of a computer program. Nothing revealed either through introspection or the examination of the anatomy of the brain leads to the conclusion that the brain is operating as a computer. If someone claims that it is, the burden of proof is on that person to justify that claim. Such proof must be base on analysis of the brain's structure and not on logical, mathematical, or philosophical grounds. Since even the physical basis of memory is poorly understood at present, any claim that the brain is a computer is at best an unproven hypothesis. Bernie Simon
andrew@dtg.nsc.com (Lord Snooty @ The Giant Poisoned Electric Head ) (12/28/89)
. 1) All physical objects are not machines. . 2) Not all machines are computers. What is a machine? It could be said that: A stone is a machine; slow crystallisation processes within. The internal elementary particle dynamics too. This is machinery. What is a computer? It could be said that: Lamps could be computers if composed of Finite State Automata on the molecular level. The program ensures that output intensity remains approximately constant, and that the structural form remains relatively invariant during illumination. This is computing. . 3) There are some activities which can be performed by physical objects . and machines which cannot be peformed by computers. . The distinction is that all computers, as computers, are . equivalent to Turing machines. What if the physical environment were used to compute with, instead of electrical energy? - think abacus. Flying then, for example, might be an emergent and NECESSARY property of such a computer. . 4) The simulation of a physical activity by a computer cannot be . identified with the physical activity. Irrelevant in light of rebuttal of 3) . 5) Hence, while it may be possible to build a machine that thinks, it . does not follow that it will be possible to build a computer that . thinks, as not all physical activities can be performed by computers. There is no known constraint on the physical activities which can be performed by computers, including those of the brain (I naturally exclude violations of the known physical laws). . 6) While there are good reasons to believe that thinking is a physical . activity, there are no good reasons for believing that thinking is the . execution of a computer program. Nothing revealed either through . introspection or the examination of the anatomy of the brain leads to . the conclusion that the brain is operating as a computer. If someone . claims that it is, the burden of proof is on that person to justify that . claim. Such proof must be base on analysis of the brain's structure and . not on logical, mathematical, or philosophical grounds. Since even the . physical basis of memory is poorly understood at present, any claim that . the brain is a computer is at best an unproven hypothesis. Repeat - what is a computer? -- ........................................................................... Andrew Palfreyman a wet bird never flies at night time sucks andrew@dtg.nsc.com there are always two sides to a broken window
mbb@cbnewsh.ATT.COM (martin.b.brilliant) (12/28/89)
In article <1037@ra.stsci.edu> bsimon@stsci.EDU (Bernie Simon) writes:
!I would like to make a few point that seem clear to me, but apparently
!aren't clear to others in this discussion.
Good thing to do.
!1) All physical objects are not machines. For example, stones, clouds,
!flowers, butterflies, and people are not machines...
Meaning that a machine has to be an artifact. OK, people sometimes
call people machines to emphasize that people and machines are governed
by the same physics. But saying that people are machines really begs
the question "Can machines think," doesn't it?
!2) Not all machines are computers. Lamps, screwdrivers, and cars are not
!computers.
OK. But all computers are machines. And a machine can contain a computer.
!3) There are some activities which can be performed by physical objects
!and machines which cannot be peformed by computers. Birds can fly and
!airplanes can fly, but computers cannot fly...
!4) The simulation of a physical activity by a computer cannot be
!identified with the physical activity. A computer running a flight
!simulation program is not flying.
True.
!5) Hence, while it may be possible to build a machine that thinks, it
!does not follow that it will be possible to build a computer that
!thinks, as not all physical activities can be performed by computers.
We seem to have got off the track. The question was not whether
computers can think, but whether machines can think. If you put a
computer into a machine that can accept sensory input and create
motor output, it might be able to do what we call thinking.
!6) While there are good reasons to believe that thinking is a physical
!activity, there are no good reasons for believing that thinking is the
!execution of a computer program....
I wouldn't believe that for a minute. I don't know exactly what
thinking is, but it is probably something a computer can't do alone,
but a machine with a computer in it might be able to do.
!.... Nothing revealed either through
!introspection or the examination of the anatomy of the brain leads to
!the conclusion that the brain is operating as a computer....
Is that a requirement for machines to think? Consider a machine with
sensory inputs and motor outputs. It needs a controller. Do you have
to have an actual brain inside, or will it be sufficient to have a
computer that simulates the brain?
Flying. We talked about flying. A computer can't fly. But if you
build a machine with eyes, and wings, and feet, and it needs a
controller, a machine that simulates a brain will be just as effective
as a genuine brain.
M. B. Brilliant Marty
AT&T-BL HO 3D-520 (201) 949-1858
Holmdel, NJ 07733 att!hounx!marty1 or marty1@hounx.ATT.COM
After retirement on 12/30/89 use att!althea!marty or marty@althea.UUCP
Disclaimer: Opinions stated herein are mine unless and until my employer
explicitly claims them; then I lose all rights to them.dg1v+@andrew.cmu.edu (David Greene) (12/28/89)
Excerpts from netnews.comp.ai: 27-Dec-89 Re: Can Machines Think? martin.b.brilliant@cbnew (2932) > !In article <1037@ra.stsci.edu> bsimon@stsci.EDU (Bernie Simon) writes: > !5) Hence, while it may be possible to build a machine that thinks, it > !does not follow that it will be possible to build a computer that > !thinks, as not all physical activities can be performed by computers. > We seem to have got off the track. The question was not whether > computers can think, but whether machines can think. If you put a > computer into a machine that can accept sensory input and create > motor output, it might be able to do what we call thinking. I would welcome some clarification... Let's assume there is some agreement on what constitutes "what we call thinking" -- a big assumption. Is it the case that machines alone can think? Or is it that a machine requires a computer (as a necessary but not sufficient condition) to think? (and that a computer alone is insufficient) If it is only the machine+computer combination that is capable, what is it about the combination? Is it the ability to control its sensory inputs and outputs (the machine part) or some other distinction? -David -------------------------------------------------------------------- David Perry Greene || ARPA: dg1v@andrew.cmu.edu, dpg@isl1.ri.cmu.edu Carnegie Mellon Univ. || BITNET: dg1v%andrew@vb.cc.cmu.edu Pittsburgh, PA 15213 || UUCP: !harvard!andrew.cmu.edu!dg1v --------------------------------------------------------------------
kp@uts.amdahl.com (Ken Presting) (12/29/89)
Here is the original argument under discussion: 1. Systems with an appropriate causal structure think. 2. Programs are a way of formally specifying causal structures. 3. Physical systems implement programs. 4. Physical systems which implement the appropriate program think. I have been arguing that this argument is unsound because (2) is false. By no means do I dispute the conclusion, though of course others would. David Chalmers writes: >Chaos is only a problem if we need to model the behaviour of a particular >system over a particular period of time exactly -- i.e., not just capture >how it might go, but how it *does* go. This isn't what we're trying to do >in cognitive science, so it's not a problem. We can model the system to >a finite level of precision, and be confident that what we're missing is >only random "noise." So while we won't capture the exact behaviour of System X >at 3 p.m. on 12/22/89, we'll generate equally plausible behaviour -- in other >words, how the system *might* have gone, if a few unimportant random >parameters had been different. M. B. Brilliant writes: >I second that. The goal of AI is not to model a particular mind, but >to create a mind. These objections seem to grant at least a part of my point - some of the characteristics of some causal systems cannot be specified by programs. I agree that an AI need not model any particular person at a particular time. But since the error in a numerical model is cumulative over time slices, it's not just the behavior of the system at a given time that won't match, but also the general shape of the trajectories though the state space of the system. If a numerical model of the brain is claimed to be accurate except for "noise", and therefore claimed to be conscious, then it must be shown that what is called "noise" is irrelevant to consciousness (or thinking). Fluctuations that seem to be "noise" may have significant consequences in a chaotic system. David Chalmers continues: >Incidentally, you can concoct hypothetical analog systems which contain >an infinite amount of information, even in this sense -- by coding up >Chaitin's Omega for instance (and thus being able to solve the Halting >Problem, and be better than any algorithm). In the real world, quantum >mechanics makes all of this irrelevant, destroying all information beyond >N bits or so. Quantum mechanics can't destroy any information - it just make the information statistical. Note that in the wave formulation of QM, the probability waves are continuous, and propagate and interfere deterministically. No probability information is ever lost, but retrieving the probabilistic information can be time consuming. The physical system need not "retrieve" the probabilistic information; it can react directly. John Nagle writes: > 2. Recent work has resulted in an effective way to solve N-body > problems to an arbitrary level of precision and with high > speed. See "The Rapid Evaluation of Potential Fields in > Particle Systems", by L.F. Greengard, MIT Press, 1988. > ISBN 0-262-07110-X. > > Systems with over a million bodies are now being solved using > these techniques. It's not enough to do fast and accurate calculations; the calculations must remain fast no matter how accurate the simulation has to be. Every computer must have a finite word size, so when accuracy levels require multiple words to represent values in a state vector, the model will slow down in proportion to the number of words used. This effect is independent of the efficiency of the basic algorithm.
dhw@itivax.iti.org (David H. West) (12/29/89)
In article <f6xk02b078qO01@amdahl.uts.amdahl.com> kp@amdahl.uts.amdahl.com (Ken Presting) writes: |Here is the original argument under discussion: | | 1. Systems with an appropriate causal structure think. | 2. Programs are a way of formally specifying causal structures. | 3. Physical systems implement programs. | 4. Physical systems which implement the appropriate program think. | |I have been arguing that this argument is unsound because (2) is false. |By no means do I dispute the conclusion, though of course others would. [quotes from David Chalmers and Marty Brilliant omitted] |These objections seem to grant at least a part of my point - some of the |characteristics of some causal systems cannot be specified by programs. Since the essence of your point is the lack of infinite precision, you could just as well say that the characteristics of some systems cannot be perceived. But you haven't given any reasons to suppose that this prohibits intelligence or consciousness, only omniscience. |then it must be shown that what is called "noise" is irrelevant to |consciousness (or thinking). Fluctuations that seem to be "noise" may |have significant consequences in a chaotic system. If I'm trying to choose between nearly-equally-preferred alternatives, fluctuations may tip the balance, but IMO the "thought" aspect here lies in the ability to evaluate utility reasonably well, not in the ability to evaluate it perfectly. Internal and external fluctuations also affect my ability to carry out my intentions, but that doesn't [in itself!] make me unintelligent or non-conscious, just not omnipotent. |It's not enough to do fast and accurate calculations; the calculations |must remain fast no matter how accurate the simulation has to be. Every This seems to imply that speed can make the difference between thought and non-thought, but none of your points 1-4 mention speed. -David West dhw@iti.org
hwajin@wrs.wrs.com (Hwa Jin Bae) (12/30/89)
In article <7853@portia.Stanford.EDU> dove@portia.Stanford.EDU (Dav Amann) writes: [...] >Perhaps this intuition explains the vigorous defense of Cartesian >dualism since Descartes without very much empirical evidence. >Lately, however, one of the newer sciences has been breaking away from >Democritus. Biology discusses and understands more and more about the >individual cell, yet they find it harder and harder to explain the >relationship between cells within the pretext of the individual cell. >Etymologysts understand a lot about termites but they cannot explain >why five termites together will build arches the Romans would be >proud of. The whole is more than the sum of its parts. This theme is further detailed in Fritjof Capra's _The Turning Point_, which states that Cartesian-Newtonian framework is not sufficient for a complete understanding of human and physical problems. [I'm sure you all have been noticing recent abundance on this particular subject in popular literature.] His solution seems to be to incorporate a holistic and ecological aspect into the Cartesian-Newtonian framework, producing a new "multidisciplinary" methodology to appropach problems. Not only that, he seems to be proposing that various different but mutually consistent concepts may be used to describe different aspects and levels of reality, without the need to reduce the phenomena of any level to those of other. Interesting. hwajin
dave@cogsci.indiana.edu (David Chalmers) (12/30/89)
In article <1037@ra.stsci.edu> bsimon@stsci.EDU (Bernie Simon) writes: >I would like to make a few point that seem clear to me, but apparently >aren't clear to others in this discussion. Hmmm, do I take it from the references line that you mean me? >1) All physical objects are not machines. >2) Not all machines are computers. >3) There are some activities which can be performed by physical objects >and machines which cannot be performed by computers. 1) Arguable but not relevant. 2) Of course. 3) Of course. >Birds can fly and airplanes can fly, but computers cannot fly. [...] >4) The simulation of a physical activity by a computer cannot be >identified with the physical activity. A computer running a flight >simulation program is not flying. >5) Hence, while it may be possible to build a machine that thinks, it >does not follow that it will be possible to build a computer that >thinks, as not all physical activities can be performed by computers. If you recall, the *premise* of the current discussion was that thinking is thinking in virtue of its abstract causal structure, and not in virtue of physical details of implementation. If you want to argue with this premise -- functionalism -- then fine. The point was not to defend it, but to defend a view of the relation between computation and cognition which is less simple-minded than "the mind is a computer". Of course, I also believe that functionalism is true. The functionalist believes that thinking is fundamentally different to flying (and heating, swimming, and nose-blowing). The essence of flying certainly *cannot* be captured in an abstract causal structure. This is because there are substantive *physical* criteria for flying. An object, *by definition*, is not flying unless it is (very roughly) engaged in ongoing motion without any connection to the ground. Nothing abstract about this -- it's a solid, physical criterion. If you capture the causal patterns without the correct physical realization, then it's not flying, period. Similarly for nose-blowing and the rest. Thinking, on the other hand, has no such solid criteria in its definition. The only definitive criterion for thought is "having such and such a subjective experience" -- which is far away from physical details (and a criterion which is understood notoriously badly). Of course, this doesn't *prove* that thinking is not nevertheless inseparable from physical details -- a correct theory of mind *might* just require that for these experiences, you can't get away with anything but pointy-headed neurons. But at the very least, physical details are out of the *definition*, and there is thus a principled difference between thinking and flying. Which makes the jump to functionalism much more plausible. Maybe "thinking" is more like "adding" than like "flying". Most arguments against functionalism are in terms of "funny instantiations" -- as in "but *this* has the right causal dynamics, and surely *this* doesn't think". Generally Chinese objects seem to be favoured for these arguments -- whether Rooms, Gyms or Nations. Some people find these intuitively compelling. As for me, I find the arguments sufficiently unconvincing that my "faith" is not only affirmed but strengthened. >6) While there are good reasons to believe that thinking is a physical >activity, there are no good reasons for believing that thinking is the >execution of a computer program. Nothing revealed either through >introspection or the examination of the anatomy of the brain leads to >the conclusion that the brain is operating as a computer. If someone >claims that it is, the burden of proof is on that person to justify that >claim. Such proof must be base on analysis of the brain's structure and >not on logical, mathematical, or philosophical grounds. Since even the >physical basis of memory is poorly understood at present, any claim that >the brain is a computer is at best an unproven hypothesis. I agree. Did you read my first note? The whole point is that you can accept the computational metaphor for mind *without* believing somewhat extreme statements like "the brain is a computer", "the mind is a program", "cognition is just symbol-manipulation" and so on. The role of computer programs is that they are very useful formal specifications of causal dynamics (which happen to use symbols as an intermediate device). Implementations of computer programs, on the other hand, possess *physically* the given causal dynamics. So if you accept (1) functionalism, and (2) that computer programs can capture any causal dynamics, then you accept that implementations of the right computer programs think. -- Dave Chalmers (dave@cogsci.indiana.edu) Concepts and Cognition, Indiana University. "It is not the least charm of a theory that it is refutable"
bwk@mbunix.mitre.org (Kort) (12/30/89)
In article <1037@ra.stsci.edu> bsimon@stsci.EDU (Bernie Simon) writes: > 6) While there are good reasons to believe that thinking is a physical > activity, there are no good reasons for believing that thinking is the > execution of a computer program. Nothing revealed either through > introspection or the examination of the anatomy of the brain leads to > the conclusion that the brain is operating as a computer. If someone > claims that it is, the burden of proof is on that person to justify that > claim. Such proof must be base on analysis of the brain's structure and > not on logical, mathematical, or philosophical grounds. Since even the > physical basis of memory is poorly understood at present, any claim that > the brain is a computer is at best an unproven hypothesis. The brain is a collection of about 400 anatomically identifiable neural networks, interconnected by trunk circuits called nerve bundles, and connected to the outside world by sensory organs (eyes, ears, nose, tactile sensors) and effectors (muscles, vocal cords). Neural networks are programmable computational devices, capable of categorizing stimuli into cases, and capable of instantiating any computable function (some more easily than others). Artificial neural networks are used today for classifying applicants for credit or insurance. They have also been used to read ASCII text and drive a speech synthesizer, thereby demonstrating one aspect of language processing. As to memory, you might want to explore recent research on the Hebb's synapse. --Barry Kort
bwk@mbunix.mitre.org (Kort) (12/30/89)
In article <f6xk02b078qO01@amdahl.uts.amdahl.com> kp@amdahl.uts.amdahl.com (Ken Presting) writes: > I agree that an AI need not model any particular person at a particular > time. But since the error in a numerical model is cumulative over time > slices, it's not just the behavior of the system at a given time that > won't match, but also the general shape of the trajectories though the > state space of the system. If a numerical model of the brain is claimed > to be accurate except for "noise", and therefore claimed to be conscious, > then it must be shown that what is called "noise" is irrelevant to > consciousness (or thinking). Fluctuations that seem to be "noise" may > have significant consequences in a chaotic system. Noise is to thinking as genetic mutations are to evolution. Most noise is counterproductive, but occasionally the noise leads to a cognitive breakthrough. That's called serendipity. --Barry Kort
cam@aipna.ed.ac.uk (Chris Malcolm) (12/30/89)
In article <973@metapsy.UUCP> sarge@metapsy.UUCP (Sarge Gerbode) writes: >In article <1989Dec19.061822.27585@athena.mit.edu> crowston@athena.mit.edu >(Kevin Crowston) writes: >>>[After object] code is loaded, there is actually a different >>>physical machine there, just as much as if one had gone out and >>>bought a different machine. >>But even so, the program still exists in both cases, right? >Good question. What *is* a "program", anyway? > ... >is it, as I suspect, the detailed *concept* the programmer had in >mind when he wrote the source code? Computer programs, like knitting patterns, sometimes arise by serendipitous accident. "Gee, that looked good - wonder if I can do it again?" There are also those cases where one computer program invents another. In these cases there need never have been any mind which had a detailed concept, or even intention, behind the source code. Even in fully deliberate programs, it is sometimes the case that the programmer fixes a bug by accident without understanding it - sometimes the only way, for obscure bugs, just to tinker until it goes away. But I would not like to say that programs which have never been understood are therefore not programs, any more than I would like to say that if God does not understand how I'm built then I'm not really a person. -- Chris Malcolm cam@uk.ac.ed.aipna 031 667 1011 x2550 Department of Artificial Intelligence, Edinburgh University 5 Forrest Hill, Edinburgh, EH1 2QL, UK
gall@yunexus.UUCP (Norm Gall) (12/31/89)
bwk@mbunix.mitre.org (Kort) writes: | Noise is to thinking as genetic mutations are to evolution. Most | noise is counterproductive, but occasionally the noise leads to a | cognitive breakthrough. That's called serendipity. Don't you think you are playing fast and loose with these concepts? What you say noise is when you equate it with genetic mutation is not the same as what a radio operator knows it to be, what a philosopher knows it to be, and what the mother of a teenager in Windsor, ON knows it to be. I'm not saying that you haven't defined your concepts well enough (ai scientists have more that adequately defined it, for their purposes). My question is "What licenses you to shift the meaning of any particular term?" nrg -- York University | "Philosophers who make the general claim that a Department of Philosophy | rule simply 'reduces to' its formulations Toronto, Ontario, Canada | are using Occam's razor to cut the throat _________________________| of common sense.' - R. Harris
miron@fornax.UUCP (Miron Cuperman ) (12/31/89)
In article <f6xk02b078qO01@amdahl.uts.amdahl.com> kp@amdahl.uts.amdahl.com (Ken Presting) writes: > David Chalmers writes: >>So while we won't capture the exact behaviour of System X >>at 3 p.m. on 12/22/89, we'll generate equally plausible behaviour -- in other >>words, how the system *might* have gone, if a few unimportant random >>parameters had been different. > >These objections seem to grant at least a part of my point - some of the >characteristics of some causal systems cannot be specified by programs. >I agree that an AI need not model any particular person at a particular >time. But since the error in a numerical model is cumulative over time >slices, it's not just the behavior of the system at a given time that >won't match, but also the general shape of the trajectories though the >state space of the system. My thoughts: Let us say you see a leaf falling. Since you are a chaotic system your trajectory through space-time may be completely different than it would be if you did not see that leaf. But did that leaf make you non human? Did it 'kill' you because it changed your future so drastically? I don't think so. Let us say that we model someone on a computer but we do not capture everything. Because of the imperfections of the model the resulting system will diverge. (Also because the inputs to this system and to the original are different.) Isn't that equivalent to the falling leaf incident? (assuming the model is close enough so it does not cause a breakdown in the basic things that make a human -- whatever those are.) I don't agree that some characteristics cannot be specified by programs. They can be specified up to any precision we would like. I don't think the human brain is so complex that it has an infinite number of *important* parameters (that without them you will fail the 'thinking test'). Actually I don't think there are so many $10M will not capture today (if we knew how to model them). You also wrote that chaotic systems are specificaly hard to model. A computer is a chaotic system. It is very easy to model a computer. Therefore it may be possible to model other chaotic systems. You have to justify your claim better. Summary: Inaccurate modeling may have an effect similar to 'normal' events. Since the inputs will be different anyway, the inaccuracy may not matter. Miron Cuperman miron@cs.sfu.ca
jpp@tygra.UUCP (John Palmer) (12/31/89)
}In article <85217@linus.UUCP> bwk@mbunix.mitre.org (Barry Kort) writes: }In article <1037@ra.stsci.edu} bsimon@stsci.EDU (Bernie Simon) writes: } } } 6) While there are good reasons to believe that thinking is a physical } } activity, there are no good reasons for believing that thinking is the } } execution of a computer program. Nothing revealed either through } } introspection or the examination of the anatomy of the brain leads to } } the conclusion that the brain is operating as a computer. If someone } } claims that it is, the burden of proof is on that person to justify that } } claim. Such proof must be base on analysis of the brain's structure and } } not on logical, mathematical, or philosophical grounds. Since even the } } physical basis of memory is poorly understood at present, any claim that } } the brain is a computer is at best an unproven hypothesis. } }The brain is a collection of about 400 anatomically identifiable }neural networks, interconnected by trunk circuits called nerve bundles, }and connected to the outside world by sensory organs (eyes, ears, nose, }tactile sensors) and effectors (muscles, vocal cords). Neural networks }are programmable computational devices, capable of categorizing stimuli }into cases, and capable of instantiating any computable function (some }more easily than others). Artificial neural networks are used today }for classifying applicants for credit or insurance. They have also }been used to read ASCII text and drive a speech synthesizer, thereby }demonstrating one aspect of language processing. As to memory, you }might want to explore recent research on the Hebb's synapse. } }--Barry Kort But the brain is not structurally programmable. The tradeoff principle states that no system can have structural programmability, evolutionary adaptability and efficiency at the same time. Digital computers are programmable, but lack efficiency (I may post more on this later) and evolutionary adaptability. The brain (ie: humans) has evolutionary adaptability and (relative) efficiency. Biological neurons are much more complex than their weak cousins (artificial neurons) and contain internal dynamics which play a very important role in their function. Things like second messenger systems and protein/substrate interactions are important. Internal dynamics rely heavily on the laws of physics and we cannot determine what "function" a neuron "computes" unless we do a physics experiment first. Computer engineers work very hard to mask off the effects of the laws of physics (ie: by eliminating the effects of background noise) in order to produce a device which is structurally programmable. Biological neurons, on the other hand, RELY on the laws of physics to do their work. The basic computing element of biological systems, the protein, operates by recognizing a substrate. This is accomplished by Brownian Motion and depends on weak bonds (VanderWaals interactions, etc). Thus, there is a structure/function relationship which is essential. Artificial neural nets will still be unable to solve hard problems (patttern recognition, REAL language processing, etc) because they are implemented in silicon (usually as a virtual machine on top of a standard digital computer) and are therefore inherently inefficient. In theory (Church-Turing Thesis) it is possible for such problems to be solved by digital computers, but most of the hard problems are intractable. We are very quickly reaching the limits of speed of silicon devices. The only hope of solving these hard problems is by developing devices which take advantage of the laws of physics and that have a very strong structure/function relationship. Of course, these devices will not be structurally programmable, but will have to be developed by an evolutionary process. My point: We are not going to solve the hard problems of AI by simply developing programs for our digital computers. We have to develope hardware that has a strong structure/function relationship. Sorry if this posting seems a little incoherent. Its 5am and I just woke up. I'll post more on this later. Most of these ideas are to be attributed to Dr. Michael Conrad, Wayne State University, Detroit, MI. -- = CAT-TALK Conferencing Network, Prototype Computer Conferencing System = - 1-800-446-4698, 300/1200/2400 baud, 8/N/1. New users use 'new' - = as a login id. E-Mail Address: ...!uunet!samsung!sharkey!tygra!jpp = - <<<Redistribution to GEnie PROHIBITED!!!>>>> -
sarge@metapsy.UUCP (Sarge Gerbode) (01/02/90)
In article <1779@aipna.ed.ac.uk> cam@aipna.ed.ac.uk (Chris Malcolm) writes: >In article <973@metapsy.UUCP>sarge@metapsy.UUCP (Sarge Gerbode) writes: >>What *is* a "program", anyway? >>... >>is it, as I suspect, the detailed *concept* the programmer had in >>mind when he wrote the source code? >Computer programs, like knitting patterns, sometimes arise by >serendipitous accident. "Gee, that looked good - wonder if I can do it >again?" >There are also those cases where one computer program invents >another. In these cases there need never have been any mind which had a >detailed concept, or even intention, behind the source code. You haven't really defined "program", yet. Do you mean the ascii code? I can see how that could arise from a random source, but doesn't it take a conscious being to look at the source code and label it as a "program"? I suppose it would be easy to design a program that would randomly generate syntactically correct ascii C code that would compile and run without run-time errors (probably has been done). But would you really call such a random product a program? And if so, what's so interesting about programs as such? >Even in fully deliberate programs, it is sometimes the case that the >programmer fixes a bug by accident without understanding it - >sometimes the only way, for obscure bugs, just to tinker until it >goes away. But I would not like to say that programs which have never >been understood are therefore not programs, any more than I would >like to say that if God does not understand how I'm built then I'm >not really a person. Good point. But even if a program were accidentally generated randomly (like the fabled monkeys accidentally producing a Shakespeare play), would it not require a conscious being to *label* such a production a "program", in order for it to be one? I'm not sure about this point. I suppose there might be an argument for saying that the enterprise of science is to discover the programs that exist in Nature, so that we can understand, predict, and control Nature. In particular, the DNA system could be (has been) described as a program. I'm not sure if this usage is legitimate, or if we are engaging in a bit of anthropomorphizing, here. Or "theomorphizing", if we find ourselves thinking as if the universe was somehow programmed by some sort of intelligent Being and we are discovering what that program is. -- Sarge Gerbode -- UUCP: pyramid!thirdi!metapsy!sarge Institute for Research in Metapsychology 431 Burgess Drive; Menlo Park, CA 94025
dhw@itivax.iti.org (David H. West) (01/03/90)
In article <191@fornax.UUCP> miron@cs.sfu.ca (Miron Cuperman) writes: >You also wrote that chaotic systems are specificaly hard to model. A >computer is a chaotic system. How so?
byoder@smcnet.UUCP (Brian Yoder) (01/03/90)
In article <979@metapsy.UUCP>, sarge@metapsy.UUCP (Sarge Gerbode) writes: > In article <24Yy02PR76bt01@amdahl.uts.amdahl.com> kp@amdahl.uts.amdahl.com > (Ken Presting) writes: > >In article <968@metapsy.UUCP>sarge@metapsy.UUCP (Sarge Gerbode) writes: > > I may have a somewhat radical viewpoint on this, but to me a symbol > is defined as such by the intention of the conscious being using it. > A symbol is a perceivable or detectable entity that is used to direct > attention to a particular reality or potential reality. > > Charges, current pulses, etc., are rightly regraded as symbols only to > the extent that they are intended (ultimately) to be comprehended by > some sort of conscious entity as indicating certain realities (or > potential relaities). In the absence of such intentions, they are not > symbols but mere charges, current pulses, etc. Consider the real implementation of most programs though. THey are written in a high-level language like C, Pascal, FORTRAN, or COBOL. That's what the programmer knew about. The Compiler turns those symbols into symbols that no human (usually) ever looks at or understands. The end user sees neither of these, he sees the user interface and understands what the {{program is doing from yet another perspective. What is the intelligence that understands the machine language symbols? One more step higher in complexity is to consider systems with complex memories that load memory as they go (virtual memory kinds of systems) which have a different physical configuration each time they are executed. One more step takes us to self-modifying languages like LISP which can execute and build statements in their own language. No human ever sees these intermediate symbols, but those constructs are processed and are reflected in the behavior of the program. Finally, we have really dynamic systems like neural networks that aren't so much "loaded with a program" as "taught" what to do. They like us, don't have a static program controling the behavior outputs. In a sense our brains become "different machines" from minute to minute as we learn and act. (Some might say that large portions of the population remain changeless through video stimulation, but this effect has not yet been proven :-) Are all of these working with "symbols"? If not which are? Is it only humans that can identify a symbol? What if all of the records about punched cards were destroyed while card readers still existed, would the little holes in card decks still be symbols? After the readers were destroyed? Brian Yoder > -- > Sarge Gerbode -- UUCP: pyramid!thirdi!metapsy!sarge > Institute for Research in Metapsychology > 431 Burgess Drive; Menlo Park, CA 94025 -- -<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>- | Brian Yoder | answers *byoder(); | | uunet!ucla-cs!smcnet!byoder | He takes no arguments and returns the answers | -<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-
byoder@smcnet.UUCP (Brian Yoder) (01/03/90)
In article <6902@cbnewsh.ATT.COM>, mbb@cbnewsh.ATT.COM (martin.b.brilliant) writes: > In article <1037@ra.stsci.edu> bsimon@stsci.EDU (Bernie Simon) writes: [Parts 1-4 deleted for brevity] > !5) Hence, while it may be possible to build a machine that thinks, it > !does not follow that it will be possible to build a computer that > !thinks, as not all physical activities can be performed by computers. I think that this is being a little bit too restrictive. It is pretty clear (at least to those of us who believe that humans can think ;-) that brains think. However without the "machine" through which it operates it couldn't do much about changing the world or discovering facts about it. To be fair, our potentially intelligent computer would have to have some kind of "body" with senses and output devices (hands, wheels, or at least a video display). > !6) While there are good reasons to believe that thinking is a physical > !activity, there are no good reasons for believing that thinking is the > !execution of a computer program.... > > I wouldn't believe that for a minute. I don't know exactly what > thinking is, but it is probably something a computer can't do alone, > but a machine with a computer in it might be able to do. What would be missing is something for the computer/machine to think about and a way for it to let us know that it thought something. There's not much to think about without any input. As for what thinking is, the definition ought to include interpretation of information, the deduction of new information, and decisions about courses of action. Isn't that something both brains and programs both do pretty well? > !.... Nothing revealed either through > !introspection or the examination of the anatomy of the brain leads to > !the conclusion that the brain is operating as a computer.... Maybe we should look at it the other way around, we could have a computer acting as a brain in this machine/computer. If it interpreted sensory data selected actions, and orchestrated their implementation (say, by flapping wings) isn't that accomplishing the same end as a brain would? Brian Yoder -- -<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>- | Brian Yoder | answers *byoder(); | | uunet!ucla-cs!smcnet!byoder | He takes no arguments and returns the answers | -<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-
bwk@mbunix.mitre.org (Kort) (01/03/90)
In article <6126@yunexus.UUCP> gall@yunexus.UUCP writes: > In article <85218@linus.UUCP> bwk@mbunix.mitre.org (Barry Kort) writes: > > Noise is to thinking as genetic mutations are to evolution. Most > > noise is counterproductive, but occasionally the noise leads to a > > cognitive breakthrough. That's called serendipity. > Don't you think you are playing fast and loose with these concepts? > What you say noise is when you equate it with genetic mutation is not > the same as what a radio operator knows it to be, what a philosopher > knows it to be, and what the mother of a teenager in Windsor, ON > knows it to be. I'm using noise as a metaphor and analogizing it to random perturbations in an otherwise deterministic system. I find that analogies and metaphors are useful tools in creative thinking, helping to direct the mind toward deeper understanding of complex processes. > I'm not saying that you haven't defined your concepts well enough (AI > scientists have more that adequately defined it, for their purposes). > My question is "What licenses you to shift the meaning of any > particular term?" My birthright licenses me to use my brain and mind to seek knowledge and understanding, and to communicate interesting and intriguing ideas with like-minded philosophers. I hope you share the same birthright. I would hate to see you voluntarily repress your own opportunity to participate in the exploration of interesting ideas. --Barry Kort
flink@mimsy.umd.edu (Paul V Torek) (01/04/90)
kp@amdahl.uts.amdahl.com (Ken Presting) writes: >If a numerical model of the brain is claimed to be accurate except for >"noise", and therefore claimed to be conscious, then it must be shown >that what is called "noise" is irrelevant to consciousness (or thinking). Are you suggesting that (a) Some types of conscious thought might go wrong were it not for the "noise", or (b) Although a "noiseless" system might pass the Turing Test, "noise" might be necessary for consciousness to exist at all? (Or something else?) Most of the rest of your article suggests (a), but (b) strikes me as a more interesting thesis. I can't think of any argument against (b). -- "There ain't no sanity clause" --Marx Paul Torek flink@mimsy.umd.edu
gilham@csl.sri.com (Fred Gilham) (01/04/90)
Brian Yoder writes: | Consider the real implementation of most programs though. THey are written | in a high-level language like C, Pascal, FORTRAN, or COBOL. That's what the | programmer knew about. The Compiler turns those symbols into symbols | that no human (usually) ever looks at or understands. The end user sees | neither of these, he sees the user interface and understands what the | {{program is doing from yet another perspective. What is the intelligence | that understands the machine language symbols? I'm pretty sure that you are using the word `symbol' here in different ways. In the case of a programmer writing in some programming language, I would say that symbols (at various levels of abstraction) are being used. However, when the program is compiled, the symbols disappear. To say that the compiler turns the symbols into other symbols is, I believe, to speak metaphorically. The point is that symbols only exist when there is someone to give them a meaning. I envision the process in this way: meaning (in the mind) | computerized |====>symbol==>(some physical pattern)==>syntactic transformation | | meaning<==symbol<==(some physical pattern)<=======| (back in the mind) It seems to me that the computer starts and ends with the physical patterns. Everything else happens in our heads. The fact that the transformations themselves can be described symbolically tends to fool people into thinking that the computer is actually using and manipulating symbols, or even manipulating meaning. This has been described as a ``hermeneutical hall of mirrors'', where we project onto the computer our own thought processes. The computer manipulates the patterns in ways that are meaningful to us; therefore the computer must be doing something involving meaning. But it isn't, any more than the Eliza program actually understood the people that talked to it, even though THEY thought it did. -Fred Gilham gilham@csl.sri.com
kp@uts.amdahl.com (Ken Presting) (01/04/90)
In article <21606@mimsy.umd.edu> flink@mimsy.umd.edu (Paul V Torek) writes: >kp@amdahl.uts.amdahl.com (Ken Presting) writes: >>If a numerical model of the brain is claimed to be accurate except for >>"noise", and therefore claimed to be conscious, then it must be shown >>that what is called "noise" is irrelevant to consciousness (or thinking). > >Are you suggesting that >(a) Some types of conscious thought might go wrong were it not for the > "noise", or >(b) Although a "noiseless" system might pass the Turing Test, "noise" > might be necessary for consciousness to exist at all? >(Or something else?) > >Most of the rest of your article suggests (a), but (b) strikes me as a >more interesting thesis. I can't think of any argument against (b). I have in mind the "something else". The point about noise in chaotic systems arises as an objection to the argument that if all other attempts at AI fail, at least we can numerically model the phyics of the brain. For this argument to work, we need to be sure that we really *can* make an accurate model. Chaotic systems can mechanically amplify small discrepancies in initial state, such as noise. Numerical models trade speed for precision, so if a model is to have the arbitrary precision needed to eliminate all discrepancies, the model would run well behind real time, and fail the Turing test. I think (b) is reversed. Random brain events are probably important in human behavior, thus affecting the Turing test. But at least the sort of thinking that is used to evaluate decision functions or logical arguments seems to depend little on randomness. Creative thinking - inventing proofs, constructing metaphors - could very well profit from random influences.
markh@csd4.csd.uwm.edu (Mark William Hopkins) (01/05/90)
In article <35@tygra.UUCP> jpp@tygra.UUCP (John Palmer) writes:
* My point: We are not going to solve the hard problems of AI by
* simply developing programs for our digital computers. We have to
* develope hardware that has a strong structure/function relationship.
Don't let biological precedent mislead you into this conclusion. Nobody ever
said that nature has the best of what is possible.
If digital systems and quasi-analogical systems such as neural nets have
complimentary strangths, then combining a digital computer and neural net into
one integrated system (where each tackles those tasks it can best handle) will
undoubtedly create a system capable of more than either is by itself ... and
probably capable of much more than biological systems ever were.
I can already see places where neural nets can be used in conjunction with
a classical problem-solving AI program (e.g. to "learn" evaluation functions)
... and these are just simplistic applications.miron@fornax.UUCP (Miron Cuperman ) (01/05/90)
In article <4711@itivax.iti.org> dhw@itivax.UUCP (David H. West) writes: >In article <191@fornax.UUCP> miron@cs.sfu.ca (Miron Cuperman) writes: >>You also wrote that chaotic systems are specificaly hard to model. A >>computer is a chaotic system. > >How so? Ok. I was unclear. The point is that many digital systems diverge in output when just one info bit is changed. My concept of chaos may be fuzzy. Actually, I would like to see some references on chaotic systems. If I get enough, I may even post a list. Miron Cuperman miron@cs.sfu.ca
Nagle@cup.portal.com (John - Nagle) (01/05/90)
We had this discussion last year. Everybody in the field has heard it altogether too many times. Could we get the philosophy out of comp.ai, please? John Nagle
hougen@umn-cs.CS.UMN.EDU (Dean Hougen) (01/06/90)
In article <25621@cup.portal.com> Nagle@cup.portal.com (John - Nagle) writes: > > We had this discussion last year. Everybody in the field has >heard it altogether too many times. Could we get the philosophy >out of comp.ai, please? I was here for last year's discussion, and it was obvious that not everybody had heard the arguments enough times, or at least had not paid attention. Quite a few people for example, talked about Searle's Chinese Room arguement as if it had something to do with translation (it doesn't). Should we get the philosophy out of comp.ai? Definately, and the CogSci, and the math, and the EE, and the ... CompSci. ;) Alot of people see overlap between ai and phil. If you don't, simply add a subject line or two to your kill file until the talk dies down. Or call for the creation of a new group to handle this discussion. Until then the calls for help and calls for papers and calls for participants will be mixed with some real discussion. :) Dean Hougen -- "The world move on a womans hips. She start to walk and she shake it up. - Talking Heads
zocy641@ut-emx.UUCP (01/08/90)
Fred writes a program on his clone to sieve for primes. He then does a hex-dump of the compiled machine code, and puts the printout in a bottle. The bottle washes ashore at the feet of Sam. Sam has never seen a clone, in fact he has had no contact with any modern folks. Still he belives the symbols are some sort of message. Through great effort he finally extracts the algorithm from the code. Is there not then a realtion between the symbols the two men express the algorithm with? Where then where the symbols hiding in the bottle? BTW INHO I will be convenced to let the AIs vote in Usenet elections, when they can proform visual parsing. I.e. I send them some pictures in (say) Postscript, and they reply as to what they see. When I am content with the response, I will welcome them aboard. As a test case to calibrate the results, let's try the test on the current Usenet lusers. Send your entries to: Henry J. Cobb hcobb@walt.cc.utexas.edu
jpp@tygra.UUCP (John Palmer) (01/08/90)
In article <23040@ut-emx.UUCP} hcobb@walt.cc.utexas.edu (Henry J. Cobb) writes: } } Fred writes a program on his clone to sieve for primes. He then does }a hex-dump of the compiled machine code, and puts the printout in a bottle. } } The bottle washes ashore at the feet of Sam. Sam has never seen a }clone, in fact he has had no contact with any modern folks. Still he }belives the symbols are some sort of message. Through great effort he finally }extracts the algorithm from the code. Is there not then a realtion between }the symbols the two men express the algorithm with? } } Where then where the symbols hiding in the bottle? } It would not be possible for "Sam" to extract the algorithm. -- = CAT-TALK Conferencing Network, Prototype Computer Conferencing System = - 1-800-446-4698, 300/1200/2400 baud, 8/N/1. New users use 'new' - = as a login id. E-Mail Address: ...!uunet!samsung!sharkey!tygra!jpp = - <<<Redistribution to GEnie PROHIBITED!!!>>>> -
gilbert@cs.glasgow.ac.uk (Gilbert Cockton) (01/09/90)
In article <25621@cup.portal.com> Nagle@cup.portal.com (John - Nagle) writes: > > We had this discussion last year. Everybody in the field has >heard it altogether too many times. Could we get the philosophy >out of comp.ai, please? Look, we wait all year for the science, and if we don't see any we're entitled to some philosophy at the end of the year as a reward for our patience. I vote the discussion continues until an interesting result in AI is announced :-) As an alternative to the Sci Am discussion, post articles on the following: "The death of positivism in the study of Man rules out Truth in AI" No credit will be given for postings which do not use real AI papers as examples. Candidates must write on one side of the text editor. -- Gilbert Cockton, Department of Computing Science, The University, Glasgow gilbert@uk.ac.glasgow.cs <europe>!ukc!glasgow!gilbert
byoder@smcnet.UUCP (Brian Yoder) (01/09/90)
In article <GILHAM.90Jan3110834@cassius.csl.sri.com>, gilham@csl.sri.com (Fred Gilham) writes: > Brian Yoder writes: > | Consider the real implementation of most programs though. THey are written > | in a high-level language like C, Pascal, FORTRAN, or COBOL. That's what the > | programmer knew about. The Compiler turns those symbols into symbols > | that no human (usually) ever looks at or understands. The end user sees > | neither of these, he sees the user interface and understands what the > | {{program is doing from yet another perspective. What is the intelligence > | that understands the machine language symbols? > > I'm pretty sure that you are using the word `symbol' here in different > ways. In the case of a programmer writing in some programming > language, I would say that symbols (at various levels of abstraction) > are being used. However, when the program is compiled, the symbols > disappear. To say that the compiler turns the symbols into other > symbols is, I believe, to speak metaphorically. The point is that > symbols only exist when there is someone to give them a meaning. > I envision the process in this way: > meaning (in the mind) > | computerized > |====>symbol==>(some physical pattern)==>syntactic transformation > | > | > meaning<==symbol<==(some physical pattern)<=======| > (back in > the mind) > It seems to me that the computer starts and ends with the physical > patterns. Everything else happens in our heads. > The fact that the transformations themselves can be described > symbolically tends to fool people into thinking that the computer is > actually using and manipulating symbols, or even manipulating meaning. > This has been described as a ``hermeneutical hall of mirrors'', where > we project onto the computer our own thought processes. The computer > manipulates the patterns in ways that are meaningful to us; therefore > the computer must be doing something involving meaning. But it isn't, > any more than the Eliza program actually understood the people that > talked to it, even though THEY thought it did. The point I was trying to make was that the information loaded in the memory of the computer IS a set of symbols. If anyone bothered to look in there with a debugging tool they'd see the symbols in there (the machine language) even though it was never in anyone's head before. Would you maintain that in this example they pattern in memory does not consist of symbols, then it does after it has been probed by the debugger? That seems a bit odd. Do they go back into being non-symbols when the debugger is removed? Are the words on this screen symbols when you stop looking at them? When you forget them? When they disappear from the screen? I say that they are carriers of information and exist in whatever medium they are expressed in. Thus, symbols exist all the time (though perhaps they cannot be translated into certain forms with the available equipment: Brain containing Paper containing Disk containing Symbols =====>Symbols ====>Symbols ===+ | Memory containing translated symbols (Object Code) | Brain Containing Screen containing Computations | Symbols <=====Symbols <====Express Symbols==+ If we had a book written in chinese and all people able to read chinese suddenly dropped dead wouldn't the things in the book still be symbols? Would they not still express information? I don't think a symbol needs to be read for it to be a symbol any more than a boat needs to float before it's a boat. What do you think? Brian Yoder -- -<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>- | Brian Yoder | answers *byoder(); | | uunet!ucla-cs!smcnet!byoder | He takes no arguments and returns the answers | -<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-<>-
gilham@csl.sri.com (Fred Gilham) (01/10/90)
Brian Yoder writes: =============== Brain containing Paper containing Disk containing Symbols =====>Symbols ====>Symbols ===+ | Memory containing translated symbols (Object Code) | Brain Containing Screen containing Computations | Symbols <=====Symbols <====Express Symbols==+ =============== I reply: Brain containing Paper containing Disk containing Symbols =====>marks ====> magnetic domains==+ | Memory containing electric charges | Brain Containing Screen containing Computations | Symbols <=====Light patterns <====(transformation==+ from physical states to other physical states) Using a debugger doesn't really change this. It is only when the light patterns get translated into symbols in our heads that the symbols exist. We use physical states to represent symbols. When you use a debugger, you impose a transformation from physical states of the computer's memory to other physical states (light patterns) on the screen or whatever. These light patterns have no intrinsic meaning, only that we impose on them. If you assert otherwise, I don't see how you can escape the conclusion that the physical patterns have meaning for everyone, since they themselves embody meaning. But this cannot be true. For example, when I read some mathematical text, I may see some fancy squiggle. My first task is to find out what the author means by this fancy squiggle. If the meaning were implicit in the fancy squiggle, I wouldn't have this problem. -Fred Gilham gilham@csl.sri.com
jb3o+@andrew.cmu.edu (Jon Allen Boone) (01/10/90)
gilham@csl.sri.com (Fred Gilham) writes: > I reply: > > Brain containing Paper containing Disk containing > Symbols =====>marks ====> magnetic domains==+ > | > Memory containing > electric charges > | > Brain Containing Screen containing Computations | > Symbols <=====Light patterns <====(transformation==+ > from physical states > to other physical > states) > Using a debugger doesn't really change this. It is only when the > light patterns get translated into symbols in our heads that the > symbols exist. We use physical states to represent symbols. When you > use a debugger, you impose a transformation from physical states of > the computer's memory to other physical states (light patterns) on the > screen or whatever. These light patterns have no intrinsic meaning, > only that we impose on them. If you assert otherwise, I don't see how > you can escape the conclusion that the physical patterns have meaning > for everyone, since they themselves embody meaning. But this cannot > be true. For example, when I read some mathematical text, I may see > some fancy squiggle. My first task is to find out what the author > means by this fancy squiggle. If the meaning were implicit in the > fancy squiggle, I wouldn't have this problem. Well, for the computer, the physical states don't mean anything until we symbol-processors force it to interpret them one way or another.... thus to the computer, they are symbols. In other words, if we didn't build computers the way we do, then the binary state of a particular section of memory wouldn't *have* to mean what it does to the computer. Thus, i claim that the computer uses symbols too, albeit different ones than we use. To me it is as valid as you claiming that a fancy squiggle is a symbol, since you have to go find out what it means. The computer is microcoded to figure out what its symbols are supposed to mean - it doesn't happen that way by default. > -Fred Gilham gilham@csl.sri.com - iain "trip to the left please...."
flink@mimsy.umd.edu (Paul V Torek) (01/11/90)
I asked: pt>Are you suggesting that pt>(a) Some types of conscious thought might go wrong were it not for the pt> "noise", or pt>(b) Although a "noiseless" system might pass the Turing Test, "noise" pt> might be necessary for consciousness to exist at all? pt>(Or something else?) kp@amdahl.uts.amdahl.com (Ken Presting) writes: >I have in mind the "something else". > >The point about noise in chaotic systems arises as an objection to the >argument that if all other attempts at AI fail, at least we can >numerically model the phyics of the brain. For this argument to work, >we need to be sure that we really *can* make an accurate model. Chaotic >systems can mechanically amplify small discrepancies in initial state But that doesn't matter unless (a) is true. As many people pointed out in reply to you, the fact that an AI system doesn't duplicate the thought processes of any *particular* person, is no problem for strong AI. The fact that my thought processes are different from yours doesn't necessarily mean I'm wrong (or that I'm not really thinking) -- it just means I'm different. Now suppose that (a) *were* true -- the "noiseless" system goes wrong, because (say) it can't think creatively, because "noise" is necessary to do so. Now *that* would be a problem. >I think (b) is reversed. Random brain events are probably important in >human behavior, thus affecting the Turing test. But at least the sort >of thinking that is used to evaluate decision functions or logical >arguments seems to depend little on randomness. I agree that any particular person's behavior probably depends on random events in her brain, but I doubt that this would affect the Turing test -- a noiseless system would not respond "wrongly", just differently. That's my hunch. But let's let that pass. Your last sentence says that a noiseless system would probably pass those aspects of the Turing Test which involve such tasks as you mention. I agree, but what (b) was suggesting was that the Turing Test might not be an adequate test of whether a "thinker" is conscious. (And if you define "thought" such that it must be conscious, then non-conscious things can't think.) -- "There ain't no sanity clause" --Marx Paul Torek flink@mimsy.umd.edu
kp@uts.amdahl.com (Ken Presting) (01/11/90)
In article <21745@mimsy.umd.edu> flink@mimsy.umd.edu (Paul V Torek) writes: >I asked: >pt>Are you suggesting that >pt>(a) Some types of conscious thought might go wrong were it not for the >pt> "noise", or > >kp@amdahl.uts.amdahl.com (Ken Presting) writes: > ..... For this {"backstop"} argument to work, >>we need to be sure that we really *can* make an accurate model. Chaotic >>systems can mechanically amplify small discrepancies in initial state > >But that doesn't matter unless (a) is true. As many people pointed out >in reply to you, the fact that an AI system doesn't duplicate the thought >processes of any *particular* person, is no problem for strong AI. The >fact that my thought processes are different from yours doesn't >necessarily mean I'm wrong (or that I'm not really thinking) -- it just >means I'm different. The divergence of the model from the real system means that for any person, at any time, the model would diverge significantly from that person's states (assuming that the brain is significantly chaotic, for the purpose of discussion). So it's not just that the numerical model can't simulate me or you, it can't simulate *anybody*, *ever*. So if we want to claim that the simulation is close enough to brain function to be simulated thought, then we have to show that the chaotic aspects of brain function are inessential to thought. BTW, my thanks to you and to the others in comp.ai who are participating in the philosophical discussions here. You folks have helped me to clarify my ideas by your constructive and thoughtful comments. This group is a good example of the solid intellectual value of Usenet.
hankin@sauron.osf.org (Scott Hankin) (01/11/90)
kp@uts.amdahl.com (Ken Presting) writes: >The divergence of the model from the real system means that for any >person, at any time, the model would diverge significantly from that >person's states (assuming that the brain is significantly chaotic, for the >purpose of discussion). So it's not just that the numerical model can't >simulate me or you, it can't simulate *anybody*, *ever*. So if we want to >claim that the simulation is close enough to brain function to be >simulated thought, then we have to show that the chaotic aspects of brain >function are inessential to thought. However, if the brain is sufficiently chaotic, we would have to assume that a "perfect duplicate" (such as might come out of a matter duplicator) would immediately diverge from the original. I fail to see how that matters. Would the duplicate therefore not be thinking? Would he/she not be the same as the original? I suspect the answer to the first question is no, for the duplicate would still function in the same manner as the original, who, we can only assume, thinks. I also suspect that the answer to the second is no. The duplicate would cease being the same as the original at the point of duplication. They would be substantially the same, to be sure, but start to differ almost immediately. I don't feel that the issue is the simulation of any given personality, but rather whether a simulation could have thought processes as close to yours as yours are to mine. - Scott ------------------------------ Scott Hankin (hankin@osf.org) Open Software Foundation
bls@cs.purdue.EDU (Brian L. Stuart) (01/12/90)
In article <85YE02kP7duL01@amdahl.uts.amdahl.com> kp@amdahl.uts.amdahl.com (Ken Presting) writes: >The divergence of the model from the real system means that for any >person, at any time, the model would diverge significantly from that >person's states (assuming that the brain is significantly chaotic, for the >purpose of discussion). So it's not just that the numerical model can't >simulate me or you, it can't simulate *anybody*, *ever*. So if we want to >claim that the simulation is close enough to brain function to be >simulated thought, then we have to show that the chaotic aspects of brain >function are inessential to thought. > This is not what's really at issue here. To simulate (or possess) intelligence is not to simulate one that possesses intelligence. We don't need to accurately simulate anyone. The questions that are significant here are: first, are the chaotic properties of the brain necessary for intelligence? If so, then what characteristics of the brain's attractor are necessary? If these characteristics are also sufficient, then there is no reason that any system possessing the same characteristics in its attractor will not also be intelligent. If the attractor characteristics are not sufficient, then we have the problem of finding out what else is necessary. In general, just because small changes in the input to chaotic systems can lead to qualatitivly different behavior does not mean that that behavior is unconstrained. It is still constrained by the system's attractor. Simulating an existing intelligence is a red herring; natural intelligent systems don't simulate others, so artificial ones likewise need not. >BTW, my thanks to you and to the others in comp.ai who are participating >in the philosophical discussions here. You folks have helped me to >clarify my ideas by your constructive and thoughtful comments. This group >is a good example of the solid intellectual value of Usenet. Ditto. Brian L. Stuart Department of Computer Science Purdue University
ercn67@castle.ed.ac.uk (M Holmes) (01/12/90)
Just a thought while hairs are being split on the difference between thinking computers, thinking machines, and thinking hybrids of computers and machines. It seems to have been suggested that computers would need to be able to manipulate the environment (basically have senses and have hands) in order to do what we call thinking. I'm not sure I'd agree but I think it's irrelevant anyway, for the following reasons. As a thought experiment (which is all thinking computers/machines are at the present time) suppose that we simulate a world within a computer system. Then we build an artificial intelligence embedded withing this simulation and allow "it" a simulated ability to sense and manipulate the simulated environment. This would seem to fulfill the criteria for a hybrid computer/machine which can sense and manipulate the "real" world. It would however simply be a program in a computer system. The point being that both sense and manipulation are simply a form of information processing which is what computers do anyway. It could be argued that this would just be "simulated thinking" but it isn't clear that this would be any different from the real thing. -- A Friend of Fernando Poo
ssingh@watserv1.waterloo.edu ($anjay "lock-on" $ingh - Indy Studies) (01/15/90)
In article <35@tygra.UUCP> jpp@tygra.UUCP (John Palmer) writes: > >Artificial neural nets will still be unable to solve hard problems >(patttern recognition, REAL language processing, etc) because they >are implemented in silicon (usually as a virtual machine on top of >a standard digital computer) and are therefore inherently inefficient. >In theory (Church-Turing Thesis) it is possible for such problems to >be solved by digital computers, but most of the hard problems are >intractable. We are very quickly reaching the limits of >speed of silicon devices. The inherent inefficiency you attribute to digital computers may be due in part to the Von Neumann bottleneck (See Hillis, W. Daniel, The Connection Machine, MIT Press 1985). The strong version of the Church-Turing Thesis, described in Hofstadter (Metamagical Themas, Bantam Books, 1985) implies that a digital computer can, given enough time, solve ANY and ALL problems. This was the intention of having such a general architecture for computers; ie, if everything is done in the memory, nothing need be physically changed. But when you try to program something as intensive as image or language processing on such a general architecture, things quickly bog down because the serial architecture of the computer, while capable of carrying out the computations necessary, gets stuck in the bottleneck between processor and memory. Image or language processing are problems that lend themselves well to a parallel architecture because they can be broken down and solved over many processors, providing a far greater information throughput than is possible with a purely serial design. Serial machines are good for simulation, because they are so open-ended, but as actual implementations of intelligence, they are somewhat lean. >The only hope of solving these hard problems is by developing devices >which take advantage of the laws of physics and that have a very >strong structure/function relationship. > >My point: We are not going to solve the hard problems of AI by >simply developing programs for our digital computers. We have to >develope hardware that has a strong structure/function relationship. This is why neural nets are the preferred mode of exploration today in large parts of AI research. There does indeed exists a strong structure/ function relation between the NN's parallel design, and the parallel nature of the problems they are being built to solve. -- $anjay "lock-on" $ingh ssingh@watserv1.waterloo.edu "A modern-day warrior, mean mean stride, today's Tom Sawyer, mean mean pride."
es@sinix.UUCP (Dr. Sanio) (02/06/90)
In article <4050@jarthur.Claremont.EDU> jwilkins@jarthur.Claremont.EDU (Jeff Wilkinson) writes: >so why don't we make one that does "continiously aquire, test, and >generalize"? why don't we make one with reasoning imperfect and chaotic >enough to simulate human behavior? why not make a mammoth machine, a >dynamic system so complex that it boggles the immagination, a self >organizing system of such sclae that it no longer "computes", but instead >manipulates fuzzy, vague, HUMAN-type thoughts symbolicly? would this not be >thought in a machine? > The problem is, IMHO, that we are still not sure what we are looking for. As far as I know, nobody has given, up to now, a valid answer on the question "what is intelligence (thought)". We could identify some aspects - such as pattern recognition, symbol creation and manipulation, some aspects of logic, and vaguely try to combine them to a generalized model of thought. In fact, we can simulate those isolated aspects more or less well in algorith- mic machines. Even trying to combine the to a system with the capabilities of a rodent exceeds our skills and the capabilities of our machines. Personally, I don't share the point that human mind is something mysterious which cannot be modelled or reproduced at all for all times. But I doubt that we are much closer to that goal (which can be questioned for its useful- ness, btw, but thats a different topic) than the medieval alchemists when they modeled a human body from clay and treated it by some substances, elec- tricity (some experimented with static electricity!) etc in order to give them the spirit of life. If the human brain is comparable to our computers, it must be regarded (under my opinion and state of knowledge) as a machine which recursively and steadi- ly reprograms code, data and even hardware (to stay inside the metaphor). (Even simple self-modifying code is not even appreciated in the area of programming, as you probably know, too - we're fairly unable to do it in a sufficiently reliable way, so for us, it's a widely banned technique) Recently, we have neither understood the hardware (the coercion of the neurons, the way of storing information etc. - we have some basic knowledge and lots of speculation about that) nor have we - beyond some initial steps - decoded the firmware (the genetic information laid down in a single cell - that's why the discussion about inheritance of intelligence is freguently breaking out again in this group and sci.psychology). About the software, we're completely specula- ting. Like the alchemists, we have some intuition about what's going on won by introspection and observation, but no valid knowledge about what "intelligence" is nor how a brain works. IMHO, the goal to build an "intelligent" machine is - at least recently - pure megalomania. > -=jefsoph/jeff wilkinson/wilky=- > -=jwilkins@jarthur.claremont.edu=- regards, es
ian@oravax.UUCP (Ian Sutherland) (02/07/90)
In article <907@athen.sinix.UUCP> es@athen.UUCP (Dr. Sanio) writes: >In article <4050@jarthur.Claremont.EDU> jwilkins@jarthur.Claremont.EDU (Jeff Wilkinson) writes: >>so why don't we make one that does "continiously aquire, test, and >>generalize"? why don't we make one with reasoning imperfect and chaotic >>enough to simulate human behavior? why not make a mammoth machine, a >>dynamic system so complex that it boggles the immagination, a self >>organizing system of such sclae that it no longer "computes", but instead >>manipulates fuzzy, vague, HUMAN-type thoughts symbolicly? would this not be >>thought in a machine? Maybe so, but why in the world would we want to build such a machine? For people who don't have enough flesh-and-blood friends? If I were building a machine to help me with a task which needed to function like a human (e.g. a robot to perform or supervise a sophisticated task in a dangerous environment), I'd want it to have LESS of the kinds of chaotic, fuzzy vagueness described above than a human. >The problem is, IMHO, that we are still not sure what we are looking for. >As far as I know, nobody has given, up to now, a valid answer on the >question "what is intelligence (thought)". Indeed. I don't see why such a definition is necessary, or even helpful. It seems to me that most of the useful work that gets done in the area of AI happens when people stop trying to make a machine that "thinks", whatever that means, and adopt a more concrete goal, like trying to make a machine to do medical diagnoses. I think the pursuit of "intelligence" in the field of AI is very counterproductive. >But I doubt >that we are much closer to that goal [...] >than the medieval alchemists when >they modeled a human body from clay and treated it by some substances, elec- >tricity (some experimented with static electricity!) etc in order to give >them the spirit of life. The likening of AI to alchemy has got to be one of the most apt metaphors I've ever heard ... -- Ian Sutherland ian%oravax.uucp@cu-arpa.cs.cornell.edu Sans Peur
taplin@thor.acc.stolaf.edu (Brad Taplin) (02/07/90)
In article <1326@oravax.UUCP> ian@oravax.odyssey.UUCP (Ian Sutherland) writes: >>>>mamoth, fuzzy, almost "humane" computer... >..........but why in the world would we want to build such a machine? >For people who don't have enough flesh-and-blood friends? If I were >building a machine to help me with a task which needed to function >like a human (e.g. a robot to perform or supervise a sophisticated >task in a dangerous environment), I'd want it to have LESS of the >kinds of chaotic, fuzzy vagueness described above than a human. Depends on the application. My first inspiration for getting into AI was the movie "2001". There's an application in which I would prefer the "face", if not the acual algorithms, to appear more "humane". This "friend in a box" could be found and activated at will to play the perfect sounding board. Her/his memory and attidude could be specifically designed to encourage and help the user forget it's just a machine doing all the "multi-tasking" I imagine such a system doing for me. Perhaps we'd be wise to endow our "friendly computer" with two linked minds, one for information processing and retrieval, and one for the pleasant, clear, almost artful (if you believe a computer can produce art) presentation of everything. The difference between such machines and today's computers could be as important (or un- if you like) as that between a stark office and a warmly decorated one. Some won't care, but most of us would find it much easier to work in the more comfy environment. Sorry if I've offended your spartan nature. >>...medeival monks and clay models......... >> (some experimented with static electricity!) etc in order to give >> them the spirit of life. Who was that again? 1500s some monk wasn't he? Buried models, seeded with sperm or something, forty days in horse dung. The products had a name? >The likening of AI to alchemy has got to be one of the most apt >metaphors I've ever heard ... ditto... but what fun to know we're onto something, even if our current understanding is riddled with wild presumptions. -- ######################################################################## "...I've gotten two thousand, fourteen times smarter since then..." B.R.T c/o Jan Aho St.Olaf Northfield MN 55057 taplin@thor.acc.stolaf.edu ########################################################################
ian@oravax.UUCP (Ian Sutherland) (02/08/90)
In article <11185@thor.acc.stolaf.edu> taplin@thor.stolaf.edu (Brad Taplin) writes: >In article <1326@oravax.UUCP> ian@oravax.odyssey.UUCP (Ian Sutherland) writes: > >>>>>mamoth, fuzzy, almost "humane" computer... >Depends on the application. Indeed. If you really DON'T have enough flesh and blood friends, and you prefer your friends to be mammoth, fuzzy and vague ;-) you might want such a machine. The kind of application the original poster was talking about was one in which you wanted the machine to be INTELLIGENT. For applications such as this, I claim you don't want to mimic the attributes of humans which cloud their thinking. -- Ian Sutherland ian%oravax.uucp@cu-arpa.cs.cornell.edu Sans Peur
andrew@dtg.nsc.com (Lord Snooty @ The Giant Poisoned Electric Head ) (02/08/90)
If we substitute "universe" for "chinese room", and "the laws of physics" for "the rules of chinese language", are we to conclude that the cosmos is a conscious entity? -- ........................................................................... Andrew Palfreyman andrew@dtg.nsc.com Albania before April!
hougen@umn-cs.cs.umn.edu (Dean Hougen) (02/08/90)
In article <621@berlioz.nsc.com> andrew@dtg.nsc.com (Lord Snooty @ The Giant Poisoned Electric Head ) writes: > >If we substitute "universe" for "chinese room", and "the laws of physics" >for "the rules of chinese language", are we to conclude that the cosmos >is a conscious entity? >Andrew Palfreyman andrew@dtg.nsc.com Albania before April! Who is it talking to? Dean Hougen -- "Summoning his cosmic powers, And glowing slightly from his toes, The psychic emination grows." - Pink Floyd
ian@oravax.UUCP (Ian Sutherland) (02/08/90)
In article <621@berlioz.nsc.com> andrew@dtg.nsc.com (Lord Snooty @ The Giant Poisoned Electric Head ) writes:
-
-If we substitute "universe" for "chinese room", and "the laws of physics"
-for "the rules of chinese language", are we to conclude that the cosmos
-is a conscious entity?
Why is this absurd (as your "Summary" line suggests)?
--
Ian Sutherland ian%oravax.uucp@cu-arpa.cs.cornell.edu
Sans Peurrandy@ms.uky.edu (Randy Appleton) (02/08/90)
I just read the Jan Scientific American, the one with Searle and so on. Here is the one burning question I have. I think a satisfactory answer to this will convince me that Searle is right, and strong-AI is wrong. But until then, I find Searle's argument to be imprecise gobbly-gook. What exactly IS the difference between "understanding" and "the formal manipulation of syntatic symbols"? He uses those two phrases quite alot, and I think it is this difference that is his main argument. BUT HE NEVER SAYS WHAT IT IS! ARG! Well, thanks Randy
taplin@thor.acc.stolaf.edu (Brad Taplin) (02/08/90)
In article <1328@oravax.UUCP> ian@oravax.odyssey.UUCP (Ian Sutherland) writes: >In article <11185@thor.acc.stolaf.edu> taplin@thor.stolaf.edu (Brad Taplin) writes: >>In article <1326@oravax.UUCP> ian@oravax.odyssey.UUCP (Ian Sutherland) writes: >>>>>>mamoth, fuzzy, almost "humane" computer... >>Depends on the application. >Indeed. If you really DON'T have enough flesh and blood friends, and >you prefer your friends to be mammoth, fuzzy and vague ;-) you might >want such a machine. The kind of application the original poster was >talking about was one in which you wanted the machine to be >INTELLIGENT... SPOILER: Several screenfuls of some crazed AI terrorist's ideas! Thanks for the reassuring ;-). I do have trouble finding trustworthy friends so I would want such a machine :-Y. But about INTELLIGENCE... Understood. I counter your cliam by suggesting that SOME (albeit few) of those "fuzzy" characteristics can actually aid "thinking", speed problem solving. Your straight algorithms might lead directly to complex solutions, but I can imagine situations in which a quick, reliable guess beats a somewhat slower but perfect response. Ever read "The Art of Motorcycle Maintenance"? Persig suggests that when his excessively single-minded pursuits of "Truth" in the mountains got bogged down he'd clear his head and let his attention drift. The critical eye remained open (wrote Persig) but the focus became whatever struck the imagination. I've tried such "lateral thinking" and found it often led to useful and accurate ideas my previous trains of thought might have never reached. If indeed the possibilities are as complex and ever-changing in real-world scenarios as I imagine them to be, then could not practical AI benefit from the "alchemy" Persig suggests? Imagine I've designed a computer to take over the responsibilities of an air traffic control tower. Some paniced pilot radios that they need an emergency landing NOW, yet the "independent" mind in my computer knows damn well that recalculating precisely which planes go where on a crowded New York Friday will probably take way too long. My computer then divies up its tasks into three: The first watches and prioritizes everything, the second (top priority now) starts making "fuzzy" educated guesses, the third calculates under a more methodical system the best possible solution. Now, if the pilot needs an answer before program3 is done, program1 (judge, manager, communicator) takes the best program2 has yet to offer and offers it on, along with a rough estimate on its chances of working, to all planes, vehicles, and people involved. While prog2 spools up new ideas of ever-greater complexity prog1 keeps both the thoughtful and quick progs informed. Once prog3 has decided the best possible solution the pilot is informed, and if s/he thinks that ultimate decision is still workable prog1 helps everyone execute the prog3 plan. If not, then everything starts afresh and prog2 has a spool of untested ideas waiting to be considered. One might argue that all three algorithms in this crisis control situation should be seen as arrow-straight, but I'm still under the impression that the quick-thinking prog2 must work most efficiently by being "softer", not just simpler, than prog3. Well-placed random variables could result in a very workable solution, even if it ain't the best, in this painfully strict timeframe. Tell me if (and why) I'm barking up a felled tree. -- ######################################################################## "...I've gotten two thousand fourteen times smarter since then..." -MCP Brad Taplin, alum, magna sin laude, afloat? taplin@thor.acc.stolaf.edu ########################################################################
andrew@dtg.nsc.com (Lord Snooty @ The Giant Poisoned Electric Head ) (02/09/90)
In article <1329@oravax.UUCP>, ian@oravax.UUCP (Ian Sutherland) writes: > I wrote: > -If we substitute "universe" for "chinese room", and "the laws of physics" > -for "the rules of chinese language", are we to conclude that the cosmos > -is a conscious entity? > Why is this absurd (as your "Summary" line suggests)? Sans Peur Because, as Dean points out, there is no external referent to "universe". However, if we now consider a microcosm which obeys the laws of physics (and we are not proponents of Mach) perhaps the absurdity goes away? -- ........................................................................... Andrew Palfreyman andrew@dtg.nsc.com Albania before April!
asanders@adobe.COM (02/09/90)
>... are we to conclude that the cosmos is a conscious entity? >>Who is it talking to? Itself, perhaps?
kp@uts.amdahl.com (Ken Presting) (02/12/90)
(This article is very long. I hope it's useful.) In article <14069@s.ms.uky.edu> randy@ms.uky.edu (Randy Appleton) writes: >I just read the Jan Scientific American, the one with Searle and so on. >Here is the one burning question I have. I think a satisfactory answer to this >will convince me that Searle is right, and strong-AI is wrong. But until >then, I find Searle's argument to be imprecise gobbly-gook. > >What exactly IS the difference between "understanding" and "the formal >manipulation of syntatic symbols"? He uses those two phrases quite alot, >and I think it is this difference that is his main argument. BUT HE NEVER >SAYS WHAT IT IS! ARG! I think this is the $64K question. Here is part of the answer: The simplest way to define "understanding" is "knowledge of meanings". That is, you understand English because given (most) any English expression, you know what it means. This much is common sense. As the definition becomes more technical and precise, it also becomes more controversial (because the technicalities may not match common sense). We can analyze the Chinese Room well enough with just this definition. Now, let me say a few words about knowledge. There is much less agreement about the definition of knowledge than about the definition of semantics. The most popular definition is "true beliefs, accompanied by reasons". The best example is mathematical knowledge - we may well believe a conjecture such as the 4-color map theorem (which is true), but until we find the proof, we can't call it knowledge. Another important distinction is between knowledge and know-how. Knowledge properly describes the relation between a person, his beliefs, and reality (ie the truth of the beliefs). Know-how depends only on abilities. You can "know how" to hit a baseball, without knowing any physics or physiology. Finally, let's look at the case where Searle memorizes the rules and passes the Turing test without the books. Searle is correct to say that he still does not know Chinese. Anyone who knows both English and Chinese must be able to translate from one into the other, but Searle cannot. What he has learned by memorizing the rules is how to respond to Chinese questions. So he has some Chinese know-how, but no knowledge of Chinese. So I think the Chinese Room example has a real point. If asked in Chinese about the meaning of a Chinese phrase, Searle would no doubt be able to respond correctly. This might suggest that he does in fact "understand" Chinese. But notice that if his questioner should ask Searle his name, or the time of day, or the color of his tie, he would *not* be able to answer correctly. This is because Searle's rules are limited to procedures for manipulating Chinese symbols, and do not include procedures for looking at his watch or his tie. By learning the rules, Searle knows that the correct response to the Chinese question, "What is a tie?" is the Chinese answer "A strip of cloth worn around the neck." But he does not know that the Chinese phrase "your tie" denotes the strip of cloth around his own neck. That is why he can correctly claim that by learning the rules he does not learn Chinese. Searle is correct that the rules contain no information about Chinese semantics. But he is wrong about *why* that information is absent. He thinks that programs have no semantics, which is an obvious mistake. It is not because the Chinese responses are programmed that they have no semantics. Rather, the Turing test itself is too easy. Turing did not insist that the conversation in the imitation game include references to events outside the dialogue. The Turing test (as most people think of it) can be passed by a program that uses no semantic information. Searle's argument revolves around the claim that information of a certain type - semantic information - cannot be learned by memorizing rules. Let's look more closely at what Searle can learn by memorizing rules. He would not learn the semantics of Chinese, but he might learn the syntax of Chinese. If he were asked in Chinese whether some expression were grammatically correct, he would apply the rules and produce the correct answer. If he were asked in English about the same Chinese phrase, he would _examine_ the rules, and perhaps find no rule which applies to the expression. Searle could infer that the expression is ungrammatical, on the assumption that the rules cover all valid Chinese expressions. If the rules cover ungrammatical expressions as well, there would probably be a small set of resonses to the effect of "I don't understand", and an examination of the rules would exhibit a large group of expressions for which the "I don't understand" symbol was the prescribed response. Depending on the sophistication of the rules, inferring the syntax of Chinese might be easy or hard, but by definition the rules contain all the information necessary to infer a complete specification of Chinese syntax. Since information content is invariant under inference, by learning the rules that enable him to pass the Chinese Turing Test, Searle _would_ learn Chinese syntax, and could apply that knowledge in English conversations (once he has performed the necessary inferences, no trivial task). Now suppose that Searle is provided with rules which not only allow him to pass the standard Turing test, but also enable him to answer Chinese questions about the color of his tie, and all the other everyday queries he might encounter living in China. When he is given the Chinese question "What color is your tie?" the rules will no doubt direct him to look at his tie, note its color, and select a Chinese symbol appropriate to that color. Clearly Searle is on his way to learning the semantics of the Chinese color vocabulary. The path from here to complete knowledge of Chinese semantics is difficult. Language-learning problems related to this have been studied by philosophers under the name "radical translation" or "radical interpretation". Armed with the rules for manipulating the symbols and the procedures for assigning symbols to observable qualities, Searle would be well prepared for the radical translation process. So if we add the appropriate proviso to the Turing test, requiring that the system not only respond coherently in kind to Chinese questions, but also display native competence in Chinese descriptions of its physical environment, then by learning the same rules Searle _would_ learn Chinese. Or at least, he would have enough information to figure out Chinese. And that knowledge of Chinese would be part of Searle's own knowledge, not a part of some "second personality". At this point, I think I've dismembered Searle's original example, but I should anticipate a probable objection: (make that several objections) Objection 1: Searle is a smart guy, speaks a couple of languages, knows about radical translation, and in general is already a thinking thing before he memorizes the rules for Chinese. Not so for a computer running the same program. The strong AI idea is that just by loading the rules into the machine, the machine will understand Chinese, that is, know the meaning of Chinese expressions. But a computer has none of the pre-existing talents that can be attributed to Searle. So what if the program contains all the information about Chinese syntax and semantics? The computer can't perform a radical translation into a language it already speaks, because it doesn't speak any language at all - and don't say it speaks machine language, there aren't even any declarative sentences in machine language. Plus the computer would have to be programmed to perform a radical translation, and off you go into an infinite regress. What Searle has before the radical translation is just more know-how about Chinese syntax and semantics, so when the rules are programmed into a computer, all you'll get is a mechanized rulebook, not a thinking thing. Reply: Mechanized, yes; rulebook, no. If you can find any symbols inside a computer, you're looking at it through a hermeneutic hall of mirrors. Objection 2: It doesn't matter that programming languages have semantics. What you need to do is get semantics into the *data* - the output of the machine. Reply: It's the implementation that forces semantics onto the data. Nobody claims that a program that's not running can think about anything. Objection 3: And what about feelings/emotions/sensations/qualia/consciousness? Reply: You define 'em, I'll argue about 'em. (Actually I have some definitions of my own for these concepts, but if I told you, that would start an even bigger argument) Objection 4: Ah, but what about the reasons for beliefs? Searle has good reasons to believe his answers to questions about Chinese syntax and semantics. The computer has no choice but to answer as it is programmed. Pressed to explain his answers, Searle could cite the expertise of the rule-writers and his own success in applying the rules. Searle has real experience of success with the rules, and real experience of the author's reliability. The computer has no such background, and therefore has no knowledge. Reply: Okay, so the computer has only opinions. I thought you wanted a thinking machine. Now you want Athena, sprung fully-formed from Zeus's brow. How is it that *you* know what English words mean? No - I mean *before* you learned about linguistics. Objection 5: Ever heard of the frame problem? To suppose that a set of rules could specify native competence in syntactic performance is one thing. But such semantic performances as forming perceptual judgements and reporting them are quite another matter. You might as well build an android, and you might have to. Reply: The answer to the frame problem is to use a smaller frame. 24 x 80 is about right. *********************************************************************** I'd better stop wisecracking before I get into trouble. So far, I've talked about understanding, but not discussed "formal symbol manipulation" at all. That is (perhaps surprisingly) MUCH more difficult. Common sense notions of understanding and knowledge are good enough to show what's happening in the Chinese Room, but we will need very precise concepts of formal symbol, symbol token, semantics, operation, program, implementation, and interpretation, before we can coherently discuss symbol manipulation. (The problem is getting your _manos_ on an _objectus_abstractus_) All the objections here depend on the difference between people and computers. The Chinese Room is easy because it deals only with a person's knowledge and abilites. I won't be able to say much about the objections above until I've made some points about computers, but I promise I'll get to them (supposing anybody cares). I didn't want to leave the impression that I was unaware of the issues. I'll be thinking furiously and typing spasmodically for a day or two. In the meantime, I'd be delighted to get any feedback whatsoever on this article. I think it's pretty slick. Ken Presting
sn13+@andrew.cmu.edu (S. Narasimhan) (02/12/90)
> Excerpts from netnews.comp.ai: 8-Feb-90 Re: Can Machines Think? Randy > Appleton@ms.uky.ed (549) > I just read the Jan Scientific American, the one with Searle and so on. > Here is the one burning question I have. I think a satisfactory answer to this > will convince me that Searle is right, and strong-AI is wrong. But until > then, I find Searle's argument to be imprecise gobbly-gook. > What exactly IS the difference between "understanding" and "the formal > manipulation of syntatic symbols"? He uses those two phrases quite alot, > and I think it is this difference that is his main argument. BUT HE NEVER > SAYS WHAT IT IS! ARG! > Well, thanks > Randy Surprisingly, none of the postings on this subject ever dealt with this question directly. However, I believe that a clear distinction exists between symbol manipulation and "understanding". I would say a system "understands" iff , given a certain event in some representation , the system can retrieve from its memory (this can be called the "case-base") a previous event which is "related " to the current event in some way . The case-base is a collection of previous experiences either manually input or acquired thru learning. The above mentioned relation between events can be quite subjective, just like responses to a certain question can be quite subjective. It is possible to design a test , which I call the case-retrieval test, to determine whether a system understands a certain event. Should the system be "intelligent" to "understand" things ? This depends on what we mean by "intelligence". However, it is possible to deal with the question of whether a system "understands" without defining "intelligence". The system ,however, should be able to reason by what is called as the "analogical reasoning" or in general "case-based reasoning". On the other hand, a system which can only manipulate symbols or rather, which can give "good" responses to input symbols need not "understand "at all. On the basis of the above, I agree with Searle that the person in the chinese room need not *necessarily* understand the question. However, I don't agree with him when he says " It is impossible to build an 'understanding' system which manipulates only symbols." Note that he also doesn't define what semantics is . I believe that even semantics is basically a group of symbols. (If interested see my Feb.7 posting in comp.ai titled "Semantics are symbols"). You might wonder on what basis I say that case retrieval is "understanding". I'll give an example. Do you understand "x" ? If no, can you say why you don't understand "x" ? Is it because it is meaningless? But, why is it "meaningless" ? Do you "understand" the following group of symbols : "John walked yesterday." Do you "understand" this : " xyzgf#$ ran yesterday ". Do you "understand" this : " John walked $$3ewr" Do you "understand" this : " John ee##2323 yesterday." Do you "understand" this : " $%#$@ #@@@@ FGGFd " I think you will notice the difference in the degree of your "understanding" the above sentences. For the first one , you were able to retrieve a "complete" case from your memory with respect to object, action ,time etc. However, this is'nt the extreme case. Suppose you had a friend whose name was John and you actually saw him walking yesterday then the above sentence might have retrieved that case. I call this exact matching as "knowing" and the particular case as "knowledge" cf. rules in an expert system. Coming back to the above examples, you'll notice that you understand them lesser and lesser ie., retrieve more inexact cases until you reach the last one where you may not able to retrieve any case at all. You'd say that you don't understand the last sentence completely, but do understand the others partially. Interestingly, note that if you "know" something then you don't require to understand it. For example do you "understand" that 2 X 2 = 4 ? Narasimhan.
weyand@csli.Stanford.EDU (Chris Weyand) (02/12/90)
kp@uts.amdahl.com (Ken Presting) writes:
::Finally, let's look at the case where Searle memorizes the rules and
::passes the Turing test without the books. Searle is correct to say that
::he still does not know Chinese. Anyone who knows both English and Chinese
::must be able to translate from one into the other, but Searle cannot.
::What he has learned by memorizing the rules is how to respond to Chinese
::questions. So he has some Chinese know-how, but no knowledge of Chinese.
::So I think the Chinese Room example has a real point.
::If asked in Chinese about the meaning of a Chinese phrase, Searle would
::no doubt be able to respond correctly. This might suggest that he does
::in fact "understand" Chinese. But notice that if his questioner should
::ask Searle his name, or the time of day, or the color of his tie, he
::would *not* be able to answer correctly. This is because Searle's rules
::are limited to procedures for manipulating Chinese symbols, and do not
::include procedures for looking at his watch or his tie. By learning the
::rules, Searle knows that the correct response to the Chinese question,
::"What is a tie?" is the Chinese answer "A strip of cloth worn around the
::neck." But he does not know that the Chinese phrase "your tie" denotes
::the strip of cloth around his own neck. That is why he can correctly
::claim that by learning the rules he does not learn Chinese.
No! Searle's assumption is that the room answers "all" questions. This
includes "what is your name?" If the questioner asked Searle for *his* name
Searle would reply "Searle" (unless he asked in Chinese). If he asked in
Chinese Searle may manipulate the book in which case the CR would respond with
a name. Remember there are obviously two agents here;
Searle and the Chinese speaker. Clearly a system couldn't pass
the Turing Test if it couldn't answer questions that would (weakly) imply self-
awareness. But Searle's assumption is that the CR does pass the TT!
Searle rather than marvel at such a machine that could pass the TT would
simply scoff and say "yea but it's still just a simulation"
::Searle is correct that the rules contain no information about Chinese
::semantics. But he is wrong about *why* that information is absent. He
::thinks that programs have no semantics, which is an obvious mistake.
::It is not because the Chinese responses are programmed that they have
::no semantics. Rather, the Turing test itself is too easy. Turing did
::not insist that the conversation in the imitation game include references
::to events outside the dialogue. The Turing test (as most people think of
::it) can be passed by a program that uses no semantic information.
Absolutely not! The Turing Test if anything is too hard. Turing even
acknowledge that himself. Turing didn't insist anything in particular
about what the interregator should ask. He simply said that rather than
ask the question "could a machine think" we should ask whether it can fool
us into believing it is a person. Clearly for us to believe an agent was
a person we would have to ask it all kinds of questions that referred to
various events. We'd ask how they felt at the moment, if they like to read,
to tell us a romantic story, to explain what it means to be conscious,
whether or not it had free will, why? etc.
::Searle's argument revolves around the claim that information of a certain
::type - semantic information - cannot be learned by memorizing rules.
::Let's look more closely at what Searle can learn by memorizing rules.
::He would not learn the semantics of Chinese, but he might learn the syntax
::of Chinese. If he were asked in Chinese whether some expression were
::grammatically correct, he would apply the rules and produce the correct
::answer. If he were asked in English about the same Chinese phrase, he
::would _examine_ the rules, and perhaps find no rule which applies to the
::expression. Searle could infer that the expression is ungrammatical, on
::the assumption that the rules cover all valid Chinese expressions. If the
::rules cover ungrammatical expressions as well, there would probably be
::a small set of resonses to the effect of "I don't understand", and an
::examination of the rules would exhibit a large group of expressions for
::which the "I don't understand" symbol was the prescribed response.
::Depending on the sophistication of the rules, inferring the syntax of
::Chinese might be easy or hard, but by definition the rules contain all
::the information necessary to infer a complete specification of Chinese
::syntax. Since information content is invariant under inference, by
::learning the rules that enable him to pass the Chinese Turing Test, Searle
::_would_ learn Chinese syntax, and could apply that knowledge in English
::conversations (once he has performed the necessary inferences, no trivial
::task).
Clearly Searle has internalized more than the rules of Chinese syntax.
The CR can carry on a conversation well enough to pass the Turing Test.
This obviuosly takes more intelligence than a natural language parsing
system and look-up table. In effect Searle has internalized an entire
brain/mind! A ridculous thought even in principle. Searle has grossly
misled the reader who buys into this argument that since all of the system
is in him and since he doesn't understand there is no understanding.
::Now suppose that Searle is provided with rules which not only allow him
::to pass the standard Turing test, but also enable him to answer Chinese
::questions about the color of his tie, and all the other everyday queries
::he might encounter living in China. When he is given the Chinese question
::"What color is your tie?" the rules will no doubt direct him to look at
::his tie, note its color, and select a Chinese symbol appropriate to that
::color. Clearly Searle is on his way to learning the semantics of the
::Chinese color vocabulary. The path from here to complete knowledge of
::Chinese semantics is difficult. Language-learning problems related to
::this have been studied by philosophers under the name "radical translation"
::or "radical interpretation". Armed with the rules for manipulating the
::symbols and the procedures for assigning symbols to observable qualities,
::Searle would be well prepared for the radical translation process.
I think you are confused about the CR situation. Searle is only manipulating
the signs and symbols of the book. The book with Searle manipulating it
is another agent that happens to speak Chinese. If the interregators asked
for the color of the agent's tie the CR would certainly not respond with the
color of Searle's tie. The questions are not aimed towards Searle.
This is all part of Searle's sophistry. He wants us to identify with him
the manipulator (the CPU) and not with the Chinese Speaker embodied within the
book. Obviuosly Searle doesn't understand a word of Chinese, Searle doesn't
have to convince me of that.
In Dennet and Hofstadter's "The Mind's I" Searle's article is included
with comments from Hofstadter and Dennet. There rebuttal to Searle is
very good and if you haven't read it I would refer you to it. I think you'll
find it very interesting.
::So if we add the appropriate proviso to the Turing test, requiring that
::the system not only respond coherently in kind to Chinese questions,
::but also display native competence in Chinese descriptions of its physical
::environment, then by learning the same rules Searle _would_ learn Chinese.
::Or at least, he would have enough information to figure out Chinese. And
::that knowledge of Chinese would be part of Searle's own knowledge, not a
::part of some "second personality". At this point, I think I've dismembered
::Searle's original example, but I should anticipate a probable objection:
::(make that several objections)
We don't need to add any proviso. We can ask whatever questions we want.
There were never any constraints on the questions; hence the power of the
test.
It's possible that Searle could learn Chinese if he had some way of relating
the symbols coming in to the room with the world. But more likely Searle
would sit in the room for the rest of his life without ever knowing what he
was doing. Searle is simply a symbol manipulator.
::I'll be thinking furiously and typing spasmodically for a day or two.
::In the meantime, I'd be delighted to get any feedback whatsoever on this
::article. I think it's pretty slick.
::Ken Presting
Very interesting ideas I'll be reading.
--Chris Weyand
--weyand@csli.Stanford.Eduted.kihm@canremote.uucp (TED KIHM) (02/12/90)
ru>What exactly IS the difference between "understanding" and "the formal ru>manipulation of syntatic symbols"? ru>BUT HE NEVER SAYS WHAT IT IS! ARG! It's that little bit of the ineffable that makes us different from machines! Perfectly legitimate argument. In any case, despite his rambling on, Searle really only claims one point. Searle is stoutly refuting the proposition that ALL computer programs are intelligent. Now that we've been enlightened to the fact that "Hello World" does not constitute an intelligent entity, lets get on with it! --- ~ DeLuxe 1z11a18 #2979 If I had finished this Tagline, ~ QNet 2.04: NorthAmeriNet: Sound Advice BBS ~ Gladstone ~ MO
kp@uts.amdahl.com (Ken Presting) (02/13/90)
In article <12214@csli.Stanford.EDU> weyand@csli.Stanford.EDU (Chris Weyand) writes: >kp@uts.amdahl.com (Ken Presting) writes: >::Finally, let's look at the case where Searle memorizes the rules and >::passes the Turing test without the books. > >::. . . notice that if his questioner should >::ask Searle his name, or the time of day, or the color of his tie, he >::would *not* be able to answer correctly. > >No! Searle's assumption is that the room answers "all" questions. This >includes "what is your name?" If the questioner asked Searle for *his* name >Searle would reply "Searle" (unless he asked in Chinese). If he asked in >Chinese Searle may manipulate the book in which case the CR would respond with >a name. Remember there are obviously two agents here; >Searle and the Chinese speaker. Clearly a system couldn't pass >the Turing Test if it couldn't answer questions that would (weakly) imply self- >awareness. But Searle's assumption is that the CR does pass the TT! It is not obvious to Searle that there are two agents! I would like to avoid discussing whether he is right about that, because I think we can make real progress if we stick to some simpler issues first. Searle has carefully separated the CR example from the rest of his argument about AI, and I want to follow him in that. The CR does a great job of splitting out one human capacity from the rest of thinking. Searle wants to focus on learning meanings, which is fine by me. So let's consider only what Searle (the guy who speaks English) learns from the rulebooks. >::Searle is correct that the rules contain no information about Chinese >::semantics. But he is wrong about *why* that information is absent. He >::thinks that programs have no semantics, which is an obvious mistake. >::It is not because the Chinese responses are programmed that they have >::no semantics. Rather, the Turing test itself is too easy. Turing did >::not insist that the conversation in the imitation game include references >::to events outside the dialogue. The Turing test (as most people think of >::it) can be passed by a program that uses no semantic information. > >Absolutely not! The Turing Test if anything is too hard. Turing even >acknowledge that himself. Turing didn't insist anything in particular >about what the interregator should ask. He simply said that rather than >ask the question "could a machine think" we should ask whether it can fool >us into believing it is a person. Clearly for us to believe an agent was >a person we would have to ask it all kinds of questions that referred to >various events. We'd ask how they felt at the moment, if they like to read, >to tell us a romantic story, to explain what it means to be conscious, >whether or not it had free will, why? etc. The questions you suggest are perfect examples of what I mean about the usual idea of the Turing test being to easy. But you do have a good point about the Turing test being too hard. I agree that in some respects it is too hard. When you suggest questions such as "Do you like to read", you allow for fixed responses. Now, none of us programmers would be particularly impressed just because somebody programmed a computer to tell the time. But check this out: No "formal symbol manipulator" can tell the time. A clock, even a clock chip, is not a formal symbol, and reading a clock is not a formal manipulation. (I swiped this time-of-day example from somebody here on the net, but I've forgotten who) >::He would not learn the semantics of Chinese, but he might learn the syntax >::of Chinese. > >Clearly Searle has internalized more than the rules of Chinese syntax. >The CR can carry on a conversation well enough to pass the Turing Test. >This obviuosly takes more intelligence than a natural language parsing >system and look-up table. . . . True, but slow down! Searle has not admitted that he would learn *anything* by memorizing the rules. At this stage, we are only talking about knowledge of language. We'll add other knowledge to the argument later. >::Now suppose that Searle is provided with rules which not only allow him >::to pass the standard Turing test, but also enable him to answer Chinese >::questions about the color of his tie, and all the other everyday queries >::he might encounter living in China. When he is given the Chinese question >::"What color is your tie?" the rules will no doubt direct him to look at >::his tie, note its color, and select a Chinese symbol appropriate to that >::color. Clearly Searle is on his way to learning the semantics of the >::Chinese color vocabulary. The path from here to complete knowledge of >::Chinese semantics is difficult. Language-learning problems related to >::this have been studied by philosophers under the name "radical translation" >::or "radical interpretation". Armed with the rules for manipulating the >::symbols and the procedures for assigning symbols to observable qualities, >::Searle would be well prepared for the radical translation process. > >I think you are confused about the CR situation. Searle is only manipulating >the signs and symbols of the book. The book with Searle manipulating it >is another agent that happens to speak Chinese. If the interregators asked >for the color of the agent's tie the CR would certainly not respond with the >color of Searle's tie. The questions are not aimed towards Searle. >This is all part of Searle's sophistry. He wants us to identify with him >the manipulator (the CPU) and not with the Chinese Speaker embodied within the >book. Obviuosly Searle doesn't understand a word of Chinese, Searle doesn't >have to convince me of that. I'm trying to convince you that Searle *does* understand Chinese! Or at least, from the right kind of rulebooks, he could figure it out. This means that if sombody holds up the Chinese character for blue, and says "What does this mean" in English, Searle will be able to say "Blue", and be able to explain his reasons for thinking that the character means blue. Given the two-agents-in-one-body point of view, I can see how you would find some ambiguity in the question "What color is your tie?". So turn it around. Suppose the interrogator asks "What color is MY tie?" (in Chinese, of course). Plug that into the paragraph above, and you should see my point. >In Dennet and Hofstadter's "The Mind's I" Searle's article is included >with comments from Hofstadter and Dennet. There rebuttal to Searle is >very good and if you haven't read it I would refer you to it. I think you'll >find it very interesting. I have read it, thank you. I think Dennet does a little better than Hofstadter; Searle's terminology is more familiar to philosophers. >::So if we add the appropriate proviso to the Turing test, requiring that >::the system not only respond coherently in kind to Chinese questions, >::but also display native competence in Chinese descriptions of its physical >::environment, then by learning the same rules Searle _would_ learn Chinese. >::Or at least, he would have enough information to figure out Chinese. And >::that knowledge of Chinese would be part of Searle's own knowledge, not a >::part of some "second personality". At this point, I think I've dismembered >::Searle's original example, but I should anticipate a probable objection: >::(make that several objections) > >We don't need to add any proviso. We can ask whatever questions we want. >There were never any constraints on the questions; hence the power of the >test. >It's possible that Searle could learn Chinese if he had some way of relating >the symbols coming in to the room with the world. But more likely Searle >would sit in the room for the rest of his life without ever knowing what he >was doing. Searle is simply a symbol manipulator. The proviso is just that the right kind of questions do get asked. If Searle can tell the time, he's not just a symbol manipulator. >Very interesting ideas I'll be reading. >--Chris Weyand >--weyand@csli.Stanford.Edu Thanks for taking the time to comment.
lee@uhccux.uhcc.hawaii.edu (Greg Lee) (02/13/90)
From article <898D02hl87rd01@amdahl.uts.amdahl.com>, by kp@uts.amdahl.com (Ken Presting): " ... But notice that if his questioner should " ask Searle his name, or the time of day, or the color of his tie, he " would *not* be able to answer correctly. ... Yes, he would: name?: (in Chinese) Hao Wang. time?: (in Chinese) I'm not wearing my watch. tie color?: (in Chinese) Green. Greg, lee@uhccux.uhcc.hawaii.edu
kp@uts.amdahl.com (Ken Presting) (02/13/90)
In article <6573@uhccux.uhcc.hawaii.edu> lee@uhccux.uhcc.hawaii.edu (Greg Lee) writes: >From article <898D02hl87rd01@amdahl.uts.amdahl.com>, by kp@uts.amdahl.com (Ken Presting): >> ... But notice that if his questioner should >> ask Searle his name, or the time of day, or the color of his tie, he >> would *not* be able to answer correctly. ... > >Yes, he would: > > name?: (in Chinese) Hao Wang. > time?: (in Chinese) I'm not wearing my watch. > tie color?: (in Chinese) Green. I'm not sure what you have in mind. If the answers are false, Chinese interrogators will know Searle is faking. Searle may not know the difference between "Hao Wang" and "John Searle" (printed in Chinese characters), but the audience would. When I said "answer correctly" I meant "make a statement which is true", not just "make a statement that is meaningful Chinese, relevant to the topic of the question".
radford@ai.toronto.edu (Radford Neal) (02/13/90)
In article <d74702yL871701@amdahl.uts.amdahl.com> kp@amdahl.uts.amdahl.com (Ken Presting) writes: >Searle has carefully separated the CR example from the rest of his >argument about AI, and I want to follow him in that. The CR does a great >job of splitting out one human capacity from the rest of thinking. Searle >wants to focus on learning meanings... I think this is one of the places where Searle goes seriously wrong. "Meanings" have no meaning outside the context of consciousness in general. To illustrate, let's consider the question of whether an air traffic control computer understands the meaning of the word "airplane". Certainly, we wouldn't say it understood "airplane" if it, say, issued instructions to pilots that would make sense only on the assumption that airplanes can fly under water. But, asks the skeptic, even if it does a wonderfull job of air traffic control, does it really understand the word "airplane"? The answer is: Who cares? If turning over air traffic control to the computer reduces the number of accidents, I (and I presume everyone else) am all in favour of doing so. Debating whether the computer understands the word "airplane" is something best left to those incapable of doing anything useful with their time. Now consider a computer that is said to understand the words "love", "fear", and "courage". It is clear to me that any entity that truely understands these words has the moral status of a "person". Conversely, I would not consider any entity to have such moral status if it didn't understand, to at least some degree, at least some such concepts. [ I will ignore here the problem of entities that are, perhaps, only embryonic or degenerate persons, such as babies and the severly demented. ] Given this, it is perverse to discuss in isolation the question of whether the computer really understands "love", "fear", or "courage". The answer hinges on the whole question of whether the computer is a person, a question which we will answer in accord with our empathic sense. I will believe the computer is a person, and understands those important words, if and only if I recognize in it the essential attributes that make my own life valuable. Unlike the question of whether the air traffic control computer understands "airplane", this question has real implications - a computer that is a person has the moral rights and responsibilities of a person, with all that implies for our actions. I don't think this question can be answered by sort of debate that accompanies the Chinese Room Problem. Radford Neal
daryl@oravax.UUCP (Steven Daryl McCullough) (02/13/90)
In article <e58H02l087KM01@amdahl.uts.amdahl.com>, kp@uts.amdahl.com (Ken Presting) writes: > In article <6573@uhccux.uhcc.hawaii.edu> lee@uhccux.uhcc.hawaii.edu (Greg Lee) writes: > >From article <898D02hl87rd01@amdahl.uts.amdahl.com>, by kp@uts.amdahl.com (Ken Presting): > >> ... But notice that if his questioner should > >> ask Searle his name, or the time of day, or the color of his tie, he > >> would *not* be able to answer correctly. ... > > > >Yes, he would: > > > > name?: (in Chinese) Hao Wang. > > time?: (in Chinese) I'm not wearing my watch. > > tie color?: (in Chinese) Green. > > I'm not sure what you have in mind. > > If the answers are false, Chinese interrogators will know Searle is faking. > Searle may not know the difference between "Hao Wang" and "John Searle" > (printed in Chinese characters), but the audience would. When I said > "answer correctly" I meant "make a statement which is true", not just > "make a statement that is meaningful Chinese, relevant to the topic of > the question". In the original Turing Test, it was required that the interrogator only be able to question the "contestant" via a teletype system, not in "person". The reason for this stipulation is that the goal of artificial intelligence is to reproduce a human mind, *not* a human body. It isn't fair, then, to look at the contestant and say "Hey, I can tell you are a computer by your keyboard!" Likewise, I think it is not fair in the Chinese Room to test the veracity of answers like "What color tie are you wearing?". If Searle keeps inside the Chinese Room, then the interrogator wouldn't be able to know that the answer is false. Answering *correctly* is not required for the Turing Test, only answering convincingly. Someone in this newsgroup (I don't remember who) brought up the issue that if computer program succeeded in passing the Turing Test, it would have to do so through lying; it would have to claim to be a human being, to have headaches occasionally, to wear green ties, etc. I don't think the fact that these claims are false should in any way be held against the computer program; it could very well have the *mind* of a human being with stomach aches, etc., and so could be answering truthfully as far as it knows. A human being can be similarly mistaken about the state of his or her own body; for example, the "phantom limb" experience of amputees, or the "phantom odors" experienced when one's brain is stimulated by an electrode. Daryl McCullough, Odyssey Research Associates oravax.uucp!daryl@cu-arpa.cs.cornell.edu
kp@uts.amdahl.com (Ken Presting) (02/14/90)
In article <90Feb12.205915est.10612@ephemeral.ai.toronto.edu> radford@ai.toronto.edu (Radford Neal) writes: >In article <d74702yL871701@amdahl.uts.amdahl.com> kp@amdahl.uts.amdahl.com (Ken Presting) writes: > >>Searle has carefully separated the CR example from the rest of his >>argument about AI, and I want to follow him in that. The CR does a great >>job of splitting out one human capacity from the rest of thinking. Searle >>wants to focus on learning meanings... > >I think this is one of the places where Searle goes seriously wrong. >"Meanings" have no meaning outside the context of consciousness in general. > >To illustrate, let's consider the question of whether an air traffic control >computer understands the meaning of the word "airplane". . . . > >The answer is: Who cares? . . . Agreed. Of course, an air traffic control computer does not have to be conscious to work well. >Now consider a computer that is said to understand the words "love", "fear", >and "courage". It is clear to me that any entity that truely understands >these words has the moral status of a "person". Conversely, I would not >consider any entity to have such moral status if it didn't understand, to >at least some degree, at least some such concepts. I think you are on the right track here, but I would put more emphasis on concepts such as "promise", "truth", "due process", and a lot of others. In general, the question "What qualities require that we grant human rights to an organism" is high on my list of important philosphical issues related to AI. >Given this, it is perverse to discuss in isolation the question of whether >the computer really understands "love", "fear", or "courage". The answer >hinges on the whole question of whether the computer is a person, a question >which we will answer in accord with our empathic sense. I will believe the ~~~~~~~~~~~~~~ >computer is a person, and understands those important words, if and only if >I recognize in it the essential attributes that make my own life valuable. I disagree strongly on this point. I don't believe there is any such "sense", although I would grant that there are emotions which do a similar job. These emotions are very important, but I think we are going to need rational grounds to make judgments such as who or what has human rights. I wouldn't trust my, your, or anybody else's emotions on these questions. >Unlike the question of whether the air traffic control computer understands >"airplane", this question has real implications - a computer that is a >person has the moral rights and responsibilities of a person, with all that >implies for our actions. I don't think this question can be answered by >sort of debate that accompanies the Chinese Room Problem. Well, here we are discussing that very question in regard to the Chinese Room! I'm glad you brought it up. Here is why I think the Chinese Room is a valuable contribution to AI. Turing's test isolated conversation from all the other forms of human behavior, and allowed AI research to concentrate its attention. It did not *force* researchers to concentrate on language, of course - that would have been stupid. Searle's Chinese Room allows us to focus even more clearly on one aspect of human language use. Again, it does not force anybody to do anything, and is not intended to do so. Humans not only can pronounce words and respond to words, they can also understand. Perhaps you will agree that those who state problems also contribute to a project, though not always as significantly as those who solve the problems. You have pointed out that the ability to understand certain words may be necessary if moral standing is to be granted to computers. Searle has argued that it is impossible for computers to understand any words at all. I have argued that programs can contain enough information to allow for understanding, which does not solve Searle's problem. But is does show that Searle has not proved the problem unsolvable.
kp@uts.amdahl.com (Ken Presting) (02/14/90)
In article <1336@oravax.UUCP> daryl@oravax.UUCP (Steven Daryl McCullough) writes: >In the original Turing Test, it was required that the interrogator >only be able to question the "contestant" via a teletype system, not >in "person". The reason for this stipulation is that the goal of >artificial intelligence is to reproduce a human mind, *not* a human >body. . . . > . . . Answering *correctly* is not required for the Turing Test, >only answering convincingly. I agree completely that convincing answers are all that is required. My point is that it is trivially simple to get a "pure symbol system" to generate unconvincing answers. This is great for Strong AI, because it shows that computers are anything but pure symbol systems. Suppose the interrogator asks, in Chinese, "What day is it?" or "What month is it?" It is common to become confused occaisionally about the date, or day of the week. But a rulebook like Searle's, or a computer which was so lazily programmed that it did not examine the system clock, would *never* get it straight. So an interrogator would start to get suspicious. Now consider a teletype-oriented question. Suppose the interrogator types as fast as he can the question "How long did this question take to type?" Then suppose he types the same question again, very slowly. A human on the other teletype could tell the difference immediately. SO COULD A REAL COMPUTER. But Searle, manipulating symbols, wouldn't have a chance. What this shows is that Searle's Axiom 1 is false. Programs *do* have semantics. It does *not* show that the program understands what it is saying or doing, but that is something I will get to later. >Someone in this newsgroup (I don't remember who) brought up the issue >that if computer program succeeded in passing the Turing Test, it >would have to do so through lying; it would have to claim to be a >human being, to have headaches occasionally, to wear green ties, etc. That was me! >I don't think the fact that these claims are false should in any way >be held against the computer program; it could very well have the >*mind* of a human being with stomach aches, etc., and so could be >answering truthfully as far as it knows. A human being can be >similarly mistaken about the state of his or her own body; for >example, the "phantom limb" experience of amputees, or the "phantom >odors" experienced when one's brain is stimulated by an electrode. The problem is not that the computer lies. There is only a problem if the computer does not know the truth. To put it better, there is a big problem if the program does not know *any* of the truth. Ken Presting
daryl@oravax.UUCP (Steven Daryl McCullough) (02/14/90)
<6573@uhccux.uhcc.hawaii.edu> <2dSM02LL88qx01@amdahl.uts.amdahl.com> In article <2dSM02LL88qx01@amdahl.uts.amdahl.com>, kp@uts.amdahl.com (Ken Presting) writes: > >I don't think the fact that these claims are false should in any way > >be held against the computer program; it could very well have the > >*mind* of a human being with stomach aches, etc., and so could be > >answering truthfully as far as it knows. A human being can be > >similarly mistaken about the state of his or her own body; for > >example, the "phantom limb" experience of amputees, or the "phantom > >odors" experienced when one's brain is stimulated by an electrode. > > The problem is not that the computer lies. There is only a problem if > the computer does not know the truth. To put it better, there is a big > problem if the program does not know *any* of the truth. > > Ken Presting I'm not sure if we are in disagreement or not. I don't usually consider it to be part of intelligence to *know* what is true and what is not. Knowing what is true (insofar as this is possible) depends on the sophistication and reliability of one's information-gathering equipment, which for a nonhandicapped human being includes eyes, ears, etc. In my opinion, the only criterion intelligence is the ability to correctly reach conclusions based on the information one has. The fact that a person or computer program has no access to a watch or a calendar to determine the time of day does not indicate a lack of intelligence, in my opinion. Let me call a computer program "virtually intelligent" if it can converse perfectly intelligently about information that it receives through conversation alone, but has no other source of new information (that is, it may have memories, but it has no way of learning what time it is, or whether it is raining, or any other fact about the real world unless that fact is revealed through conversation). It seems to me that it would be a relatively small task to modify a "virtually intelligent" program to make it "truly intelligent"; it would only require hooking up timers and TV cameras, etc. Do you agree? Daryl McCullough