harnad@elbereth.rutgers.edu (Stevan Harnad) (12/11/89)
This is a multiple reply, to multiple errors: cam@aipna.ed.ac.uk (Chris Malcolm) of Dept of AI, Edinburgh University, UK, wrote: > my game is assembly robotics... The assembly agent is designed not only > to succeed in its tasks, but to present a suitable virtual world to the > planner, there is an extra constraint on the task modularisation. That > constraint is sometimes referred to as the symbol grounding problem. As the mint that coined the term, I think I speak with a little semantic authority when I say that that's decidedly *not* the symbol grounding problem but rather a symptom of it! Virtual worlds are not real worlds, and what goes on inside a simulation is just meaningless symbol crunching. The only way to ground symbols is in the real world. Let me add the prediction that whereas a "virtual world" may allow one to ground a toy robot in the real world, it will never lead to what -- for a psychobiologist, at any rate -- is the real goal: A robot that passes the Total Turing Test. The reason is again the symbol grounding problem: Virtual worlds cannot embody all the contingencies of the real world, they can only capture as many of them symbolically as we can anticipate. In ordinary AI this was known as the "frame" problem -- but of course that's just another manifestation of the symbol grounding problem. mike@cs.arizona.edu (Mike Coffin) of U of Arizona CS Dept, Tucson, wrote: > an artificial intelligence living... in a (sub-)simulation on a Cray-9 > would have no choice but to accept the simulated flight to Istanbul to > the AI conference as ``reality.'' Here is the other side of the coin, what I have called the "hermeneutic hall of mirrors" created by projecting our interpretations onto meaningless symbols. As long as you allow yourself to interpret ungrounded symbols you'll keep coming up with "virtual reality." The only trouble is, what we're after is real reality (and that distinction is being lost in the wash). There's nobody home in a symbol cruncher, and it's not because they're on a virtual flight to Istanbul! This is what I call "Cog Sci Fi." yamauchi@cs.rochester.edu (Brian Yamauchi) of University of Rochester Computer Science Department wrote: > My complaint about most AI programs is not the worlds are simulated, > but that the simulated worlds often are very unlike any type of > perceptual reality sensed by organic creatures. It's a matter of > semantics to argue whether this is "intelligence"... > It seems that one interesting approach to AI would be to use the > virtual reality systems which have recently been developed as an > environment for artificial creatures. Then they would be living in a > simulated world, but one that was sophisticated enough to provide a > convincing illusion for *human* perceptions. Illusion is indeed the right word! Simulated worlds are no more "like" a reality than books are: They are merely *interpretable* by *us* as being about a world. The illusion is purely a consequence of being trapped in the hermeneutic hall of mirrors. And that's not just "semantics," it's just syntax... -- Stevan Harnad INTERNET: harnad@confidence.princeton.edu harnad@princeton.edu srh@flash.bellcore.com harnad@elbereth.rutgers.edu harnad@princeton.uucp BITNET: harnad@pucc.bitnet CSNET: harnad%princeton.edu@relay.cs.net (609)-921-7771
yamauchi@cs.rochester.edu (Brian Yamauchi) (12/11/89)
In article <Dec.10.11.48.06.1989.2717@elbereth.rutgers.edu> harnad@elbereth.rutgers.edu (Stevan Harnad) writes: > >yamauchi@cs.rochester.edu (Brian Yamauchi) of >University of Rochester Computer Science Department wrote: > >> My complaint about most AI programs is not the worlds are simulated, >> but that the simulated worlds often are very unlike any type of >> perceptual reality sensed by organic creatures. It's a matter of >> semantics to argue whether this is "intelligence"... >> It seems that one interesting approach to AI would be to use the >> virtual reality systems which have recently been developed as an >> environment for artificial creatures. Then they would be living in a >> simulated world, but one that was sophisticated enough to provide a >> convincing illusion for *human* perceptions. > >Illusion is indeed the right word! Simulated worlds are no more >"like" a reality than books are: They are merely *interpretable* >by *us* as being about a world. There is a *big* difference between a book and a virtual reality (or a movie, for that matter). When you read a book you are interpreting linguistic symbols, when you watch a movie you are processing raw sensory perceptions. Suppose virtual reality technology develops to the point where it is impossible for a human to tell an illusion from reality (this is already the case for still computer graphic images of some objects). In this case, the imaging patterns hitting the person's retina will be the same regardless of whether he is viewing the real world or a simulated one. Now, suppose that we can develop a program which can react to these images in the same way that a human can. Does it make any difference whether the inputs to the program come from the simulator or a pair of cameras observing the real world? Now, if you are arguing that it will be impossible in *practice* to build a simulator which has the complexity of the real-world, in terms of interactivity and modeling complex physical laws, then you may have a point. >The illusion is purely a >consequence of being trapped in the hermeneutic hall of mirrors. Actually, the illusion is the result of having very sophisticated graphics software... _______________________________________________________________________________ Brian Yamauchi University of Rochester yamauchi@cs.rochester.edu Computer Science Department _______________________________________________________________________________
blenko-tom@CS.YALE.EDU (Tom Blenko) (12/11/89)
In article <Dec.10.11.48.06.1989.2717@elbereth.rutgers.edu> harnad@elbereth.rutgers.edu (Stevan Harnad) writes: |... |Here is the other side of the coin, what I have called the "hermeneutic |hall of mirrors" created by projecting our interpretations onto |meaningless symbols. As long as you allow yourself to interpret |ungrounded symbols you'll keep coming up with "virtual reality." |The only trouble is, what we're after is real reality (and that |distinction is being lost in the wash). There's nobody home in |a symbol cruncher, and it's not because they're on a virtual |flight to Istanbul! This is what I call "Cog Sci Fi." You appear to be arguing both with the assumption that there simply is no escape from this situation, and the related proposal that no escape is necessary. If you wish to argue that there is any such thing as a symbol grounding problem, I think you have to address both of these views (which I understand to be widely accepted). The argument that there is no escape goes like this: even if all information from the environment were in principal available to a putative intelligent entity (it makes no difference whether it is artificial or not), there are necessarily limitations on what information the entity could extract. So the fact that some information may not be available in principal is at most a different facet of a existing fundamental limitation. I believe that recognition of this property was first expressed in Herb Simon's Principle of Bounded Rationality. Why isn't escape necessary? All sorts of entities (corporations, the roach species, you and I, etc.) make imperfect use of incomplete information in order to survive and reproduce. Certainly the validity of these entities as predictors of their own, real-world futures plays a role in their survival -- but there are a host of other important strategies they use that are very far divorced from anything we would term "intelligence" (e.g., genetic recombination, reproductive strategies, role specialization). So some "virtual realities" are good enough, and (in the case or roaches) may be strikingly simpler than "real reality". So the conclusion is that "real reality" is neither attainable nor necessary for any phenomenon termed "intelligence" to be realized. Tom
jwi@cbnewsj.ATT.COM (Jim Winer @ AT&T, Middletown, NJ) (12/11/89)
> >>Brian Yamauchi wrote: > >> > >> My complaint about most AI programs is not the worlds are simulated, > >> but that the simulated worlds often are very unlike any type of > >> perceptual reality sensed by organic creatures. It's a matter of > >> semantics to argue whether this is "intelligence"... > >> It seems that one interesting approach to AI would be to use the > >> virtual reality systems which have recently been developed as an > >> environment for artificial creatures. Then they would be living in a > >> simulated world, but one that was sophisticated enough to provide a > >> convincing illusion for *human* perceptions. > >Stevan Harnad writes: > > > >Illusion is indeed the right word! Simulated worlds are no more > >"like" a reality than books are: They are merely *interpretable* > >by *us* as being about a world. > Brian Yamauchi writes: > > There is a *big* difference between a book and a virtual reality (or a > movie, for that matter). When you read a book you are interpreting > linguistic symbols, when you watch a movie you are processing raw > sensory perceptions. When you read a book, you are processing raw sensory perceptions and interpreting them as literary symbols which recall associated memories or previous perceptions and emotional states. Whe you watch a movie you are processing raw sensory perceptions and interpreting them as visual symbols which recall associated memories or previous perceptions and emotional states. To an artificial creature without the necessary referent previous perceptions and emotional states of a human, the interpretation of human reality is likely to be impossible. To a human without the necessary referent previous perceptions and emotional states of a creature living in an artificial reality, the interpretation of artificial reality is likely to be equally impossible. In short, we probably couldn't communicate -- the problem isn't words, the problem is what would we have to say to an intelligent tree? > >The illusion is purely a > >consequence of being trapped in the hermeneutic hall of mirrors. > Actually, the illusion is the result of having very sophisticated > graphics software... Actually, the ilusion is the result of falsely thinking that an individual has the referents to communicate outside its particular system, or the ability to even perceive outside its system. Jim Winer -- Post, don't email, I usually can't reply. ----------------------------------------------------------------- opinions not necessarily | "And remember, rebooting your brain and do not represent | can be tricky." -- Chris Miller any other sane person | especially not employer. |
jwi@cbnewsj.ATT.COM (Jim Winer @ AT&T, Middletown, NJ) (12/11/89)
> Tom Blenko writes: > > So the conclusion is that "real reality" is neither attainable nor > necessary for any phenomenon termed "intelligence" to be realized. Perhaps, but "real reality" or at least a "common consensual reality" may very well be necessary for meaningful communication with any phenomenon termed "intelligence." Jim Winer -- Post, don't email, I usually can't reply. ----------------------------------------------------------------- opinions not necessarily | "And remember, rebooting your brain and do not represent | can be tricky." -- Chris Miller any other sane person | especially not employer. |
harnad@phoenix.Princeton.EDU (Stevan Harnad) (12/12/89)
yamauchi@cs.rochester.edu (Brian Yamauchi) University of Rochester Computer Science Department > Suppose virtual reality technology develops to the point where it is > impossible for a human to tell an illusion from reality (this is > already the case for still computer graphic images of some objects)... > Now, suppose that we can develop a program which can react to these > images in the same way that a human can... Now, if you are arguing that > it will be impossible in *practice* to build a simulator which has the > complexity of the real-world, in terms of interactivity and modeling of > complex physical laws, then you may have a point. First of all, the last point WAS my point: The problem of designing a robot that will pass the Total Turing Test (TTT) is a tiny subset of the problem of simulating the world the robot is in, not vice versa. (Another way to put it is that an analog object or state of affairs is infinitely more compact than any possible symbolic description of it. To approximate it closely enough, the description quickly becomes ludicrously large.) Second, the point of the TTT is for the ROBOT to pass it, in the world and for us, not for its WORLD to pass it for us. Finally, it's irrelevant what graphics we hook onto one symbol cruncher in providing inputs to another symbol cruncher. (That's like having two computers play chess against one another: it may as well all be one computer.) Symbol crunchers don't see, not even if you hook transducers onto them. It's just playing on the illusion from OUR point of view to bother having a graphics interface. So it's still just the hermeneutic circle, whether we're projecting our interpretations on symbolic text or on symbol-governed graphics. -- Stevan Harnad Department of Psychology Princeton University harnad@confidence.princeton.edu srh@flash.bellcore.com harnad@elbereth.rutgers.edu harnad@pucc.bitnet (609)-921-7771
jiii@visdc.UUCP (John E Van Deusen III) (12/12/89)
In article <Dec.10.11.48.06.1989.2717@elbereth.rutgers.edu> harnad@elbereth.rutgers.edu (Stevan Harnad) writes: > > ... Virtual worlds are not real worlds, and what goes on inside a > simulation is just meaningless symbol crunching. The only way to > ground symbols is in the real world. Real world? You have no way of proving that there is a "real world", that there is only one, or that we are not living inside a simulation. Intelligence is intelligence, no matter if it is accepting patterns in {a,b}* or operating a fork lift on planet earth. It is only a matter of degree. > ... a "virtual world" ... will never lead to what -- for a > psychobiologist, at any rate -- is the real goal: A robot that passes > the Total Turing Test. Does a "Total" Turing Test differ from a Turing Test in that psychobiologists will, when confronted with a robot that passes the test to the satisfaction of everyone else, insist upon running the tests out to infinity? It has been proven that equivalence testing programs with that level of generality can not exist. > ... Virtual worlds cannot embody all the contingencies of the real > world, they can only capture as many of them symbolically as we can > anticipate. Let's say we have creatures living in the universe of {a,b}*. They have gotten pretty smart, in terms of language recognition, and have started to simulate their universe, which consists of patterns of 'a's and 'b's coming through a channel. It is true that they don't have a prayer; because the number a patterns, subsets of {a,b}*, is not only infinite, it is not even enumerable. Standing on the other end of the channel, the god of the ab-creatures can send any pattern it wants and always keep the creations guessing. Now consider that the god of the ab-creatures is itself an ab-creature being fed patterns from yet a higher level. At each level, as gods, they have total knowledge of the universe of the creatures immediately below them; and, as creations, find their own universe incomprehensibly complex and unknowable. Even when the levels of simulation are carried out to infinity, there is no qualitative difference in the intelligence of the occupants or their "virtual" reality between the successive levels of simulation. Thus the inhabitants can not prove at what level they are or even that they themselves exist within a simulation. -- John E Van Deusen III, PO Box 9283, Boise, ID 83707, (208) 343-1865 uunet!visdc!jiii
harnad@phoenix.Princeton.EDU (Stevan Harnad) (12/12/89)
Tom Blenko blenko-tom@CS.YALE.EDU of Yale University Computer Science Dept wrote: > You appear to be arguing both with the assumption that there simply is > no escape from this situation, and the related proposal that no escape > is necessary. If you wish to argue that there is any such thing as a > symbol grounding problem, I think you have to address both of these > views (which I understand to be widely accepted). Before you can argue about anything connected with the symbol grounding problem you first have to know what it is (preprint available by email): THE SYMBOL GROUNDING PROBLEM (Physica D 1990, in press) Stevan Harnad Department of Psychology Princeton University ABSTRACT: There has been much discussion recently about the scope and limits of purely symbolic models of the mind and about the proper role of connectionism in cognitive modeling. This paper describes the "symbol grounding problem" for a semantically interpretable symbol system: How can its semantic interpretation be made intrinsic to the symbol system, rather than just parasitic on the meanings in our heads? How can the meanings of the meaningless symbol tokens, manipulated solely on the basis of their (arbitrary) shapes, be grounded in anything but other meaningless symbols? The problem is analogous to trying to learn Chinese from a Chinese/Chinese dictionary alone. A candidate solution is sketched: Symbolic representations must be grounded bottom-up in nonsymbolic representations of two kinds: (1) iconic representations, which are analogs of the proximal sensory projections of distal objects and events, and (2) categorical representations, which are learned and innate feature-detectors that pick out the invariant features of object and event categories from their sensory projections. Elementary symbols are the names of these object and event categories, assigned on the basis of their (nonsymbolic) categorical representations. Higher-order (3) symbolic representations, grounded in these elementary symbols, consist of symbol strings describing category membership relations ("An X is a Y that is Z"). Connectionism is one natural candidate for the mechanism that learns the invariant features underlying categorical representations, thereby connecting names to the proximal projections of the distal objects they stand for. In this way connectionism can be seen as a complementary component in a hybrid nonsymbolic/symbolic model of the mind, rather than a rival to purely symbolic modeling. Such a hybrid model would not have an autonomous symbolic "module," however; the symbolic functions would emerge in the form of an intrinsically "dedicated" symbol system as a consequence of the bottom-up grounding of categories' names in their sensory representations. Symbol manipulation would be governed not just by the arbitrary shapes of the symbol tokens, but by the nonarbitrary shapes of the icons and category invariants in which they are grounded. -- Stevan Harnad Department of Psychology Princeton University harnad@confidence.princeton.edu srh@flash.bellcore.com harnad@elbereth.rutgers.edu harnad@pucc.bitnet (609)-921-7771
mbb@cbnewsh.ATT.COM (martin.b.brilliant) (12/12/89)
From article <8093@cs.yale.edu>, by blenko-tom@CS.YALE.EDU (Tom Blenko): > In article <Dec.10.11.48.06.1989.2717@elbereth.rutgers.edu> harnad@elbereth.rutgers.edu (Stevan Harnad) writes: > |.... As long as you allow yourself to interpret > |ungrounded symbols you'll keep coming up with "virtual reality." > |The only trouble is, what we're after is real reality .... The reference to real reality raised a ghost, and part of Tom Blenko's reply illuminated it: > ..... even if all > information from the environment were in principal available to a > putative intelligent entity (it makes no difference whether it is > artificial or not), there are necessarily limitations on what > information the entity could extract...... > ...... All sorts of entities (corporations, the > roach species, you and I, etc.) make imperfect use of incomplete > information in order to survive and reproduce...... In sophomore philosophy, I heard of a problem that Rene Decartes had. He had sensory information seemingly coming into his mind, but he wasn't sure it was really sensory in origin. He was sure only that he existed himself: "Cogito, ergo sum," I think, therefore I am. In other words, he knew of the existence of the symbols, but he didn't know whether or not they were grounded. He finally copped out, through some argument based on the concept of God, and convinced himself that God had to exist, and hence, somehow, that everything else did too. But is there any difference between Descartes's symbol grounding problem (which we all have to solve for ourselves, as individuals), and the problem of symbol grounding in an automaton? I think there is no logical way out, just as there was none for Descartes. And for practical purposes we don't need one, because we have got on for billions of years without one. M. B. Brilliant Marty AT&T-BL HO 3D-520 (201) 949-1858 Holmdel, NJ 07733 att!hounx!marty1 or marty1@hounx.ATT.COM Disclaimer: Opinions stated herein are mine unless and until my employer explicitly claims them; then I lose all rights to them. Notice: Communication will cease 12/30/89 due to retirement.
mnr@daisy.learning.cs.cmu.edu (Marc Ringuette) (12/12/89)
Stevan Harnad writes, > As the mint that coined the term, I think I speak with a little > semantic authority when I say that that's decidedly *not* the symbol > grounding problem but rather a symptom of it! Virtual worlds are not > real worlds, and what goes on inside a simulation is just meaningless > symbol crunching. The only way to ground symbols is in the real world. > > Let me add the prediction that whereas a "virtual world" may allow one > to ground a toy robot in the real world, it will never lead to what -- > for a psychobiologist, at any rate -- is the real goal: A robot that > passes the Total Turing Test. The reason is again the symbol grounding > problem: Virtual worlds cannot embody all the contingencies of the real > world, they can only capture as many of them symbolically as we can > anticipate. In ordinary AI this was known as the "frame" problem -- but > of course that's just another manifestation of the symbol grounding > problem. Mr. Harnad, I think you're taking a practical argument and couching it, misleadingly, in philosophical terms. Here are three arguments you seem to be making, and my commentary: 1. Existing symbolic AI systems live in simulated worlds which are a long distance away from reality. The only reason some people think they are grounded in reality is because they're projecting. [ I agree that the distance to reality is large in most cases. ] 2. It's impractical to produce a simulated world which is a short distance from reality. (My definition of "a short distance" would be that, if you were to replace the simulation with sensors and effectors which act in the real world, the AI system would be able to operate effectively in the real world if it does so in the simulation. The construction of the sensors/effectors must not be allowed to hide complexities of the problem, of course; the simulation is a "short distance" from reality only if the sensor/effector system is a fairly direct translation.) [ I think this is a reasonable argument for someone to make, but I disagree on empirical grounds: I expect that we will soon see AI systems which run equally well on the real world and on fairly-detailed simulations. Right here in our lab at CMU, we run a real robot with an extremely simple symbolic AI system. The abstraction done by the sensor/effector system is considerable; I would say that when we decouple our system from reality and run it under the simulation, it is a fairly "large distance" from reality, so it isn't a counterexample to claim (2). However, it's only a matter of degree, and powerful robotic performance systems must decrease that distance in order to be able to fully test their systems in simulation. I believe they will succeed. ] 3. No matter what the simulation is, if it's _completely symbolic_, then it's not _grounded_. [ This is the philosophical point. I don't think this matters at all. ] As a practicing roboticist, it's clear that when I consider an AI system I should not be asking the question "Is it grounded?" but rather the question "How interesting is it?", where _interest_ is positively correlated with _realism_. That's the way for me to produce good research results. Asking this question may lead me to ground my system in the lab, or it may not. I think it's up to you to give reasons why the AI community should care about the philosophical issue of _grounding_ rather than the practical issues of _interest_ and _realism_. And I think it behooves you to be more careful to identify which of your arguments are practical and which are philosophical. If you can't argue that the symbol grounding problem is worth considering on practical grounds, you might consider emphasizing your more practical arguments in this forum. For instance, if you wish to advocate connectionism, you might wish to cite the fact that connectionist systems typically have a "short distance" to reality, and that their _realism_ contributes to their _interest_. \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\ \\\ Marc Ringuette \\\ Carnegie Mellon University, Comp. Sci. Dept. \\\ \\\ mnr@cs.cmu.edu \\\ Pittsburgh, PA 15213. Phone 412-268-3728(w) \\\ \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\
harnad@phoenix.Princeton.EDU (Stevan Harnad) (12/12/89)
mnr@daisy.learning.cs.cmu.edu (Marc Ringuette) of Carnegie-Mellon University, CS/RI wrote: > As a practicing roboticist, it's clear that when I consider an AI > system I should not be asking the question "Is it grounded?" but rather > the question "How interesting is it?", where _interest_ is positively > correlated with _realism_... I think it's up to you to give reasons > why the AI community should care about the philosophical issue of > _grounding_ rather than the practical issues of _interest_ and > _realism_. As a practicing roboticist, you can be interested in whatever you like. But if you're doing real robotics, rather than virtual robotics, your robots better be able to do whatever they do in the real world. To the extent that symbol-crunching in a virtual world can actually be translated into robotic performance in the real world, none of my objections should worry you. To the extent it cannot, they should. For my part, my interest is in a robot that can pass the Total Turing Test; the symbol grounding problem for this enterprise is empirical and methodological. It isn't and never has been philosophical. -- Stevan Harnad Department of Psychology Princeton University harnad@confidence.princeton.edu srh@flash.bellcore.com harnad@elbereth.rutgers.edu harnad@pucc.bitnet (609)-921-7771
yamauchi@cs.rochester.edu (Brian Yamauchi) (12/12/89)
In article <691@visdc.UUCP> jiii@visdc.UUCP (John E Van Deusen III) writes: > >Intelligence is intelligence, no matter if it is accepting patterns in >{a,b}* or operating a fork lift on planet earth. It is only a matter of >degree. This strikes me as the core of what is wrong with a substantial proportion of AI research. The idea that the way to build intelligence is to start with abstract "language" (in the math logic sense of the word) recognizers on {a,b}* or theorem provers using FOPC and deciding that it is only "a matter of degree" to expand these systems to human intelligence. There is one clear example of how to develop intelligent systems incrementally -- it's called evolution. Wouldn't it make more sense to develop artificially intelligent systems in an analogous manner -- starting with simple, fully autonomous creatures, and progressively adding more advanced capabilities? True, the fact that nature used this course does not mean that it is the only course, but it does mean that it is a possible one. (And hopefully, the substitution of intelligent design for random mutations will cut down the required time by a few billion years.) _______________________________________________________________________________ Brian Yamauchi University of Rochester yamauchi@cs.rochester.edu Computer Science Department _______________________________________________________________________________
mnr@daisy.learning.cs.cmu.edu (Marc Ringuette) (12/13/89)
Stevan Harnad writes, > ... the symbol grounding problem for this enterprise is empirical and > methodological. It isn't and never has been philosophical. Let's look at the following statement from a methodological point of view. You wrote, > ... what goes on inside a simulation is just meaningless > symbol crunching. The only way to ground symbols is in the real world. What did you mean? If you're saying that an AI system running in a simulation is totally meaningless, then I'll argue that you're flat wrong. I get a feeling that you only meant 'meaningless' in the philosophical sense. In that case, my question stands: why make philosophical arguments if you really mean to discuss methodological issues? And would you re-state the methodological case you were trying to make? I couldn't hear it for the noise. /////////////////////////////////////////////////////////////////////// /// Marc Ringuette /// Carnegie Mellon University, Comp. Sci. Dept. /// /// mnr@cs.cmu.edu /// Pittsburgh, PA 15213. Phone 412-268-3728(w) /// ///////////////////////////////////////////////////////////////////////
mike@cs.arizona.edu (Mike Coffin) (12/18/89)
This is a single reply to multiple errors :-) From article <Dec.10.11.48.06.1989.2717@elbereth.rutgers.edu>, by harnad@elbereth.rutgers.edu (Stevan Harnad): > As long as you allow yourself to interpret > ungrounded symbols you'll keep coming up with "virtual reality." > The only trouble is, what we're after is real reality (and that > distinction is being lost in the wash). There's nobody home in > a symbol cruncher ... You omitted the first half of my posting, in which I pointed out that there is nothing (that I know of) that precludes the possibility that *we* might be simulated. In that case, what is the difference between real reality and virtual reality? On a more general issue, this "problem of symbol grounding" seems nothing more than a mantra that some people chant when they're faced with arguments they can't deal with otherwise. If the only thing that distinguishes reality and virtual reality is that symbols are grounded in one but not the other, what is the real difference? I mean, what observable effect does "grounding" the symbols have? Does a system begin behaving differently when symbols are grounded? Would *we* begin behaving differently if the entity that wrote our program suddenly said, "My God! they're ungrounded!"? -- Mike Coffin mike@arizona.edu Univ. of Ariz. Dept. of Comp. Sci. {allegra,cmcl2}!arizona!mike Tucson, AZ 85721 (602)621-2858
cam@aipna.ed.ac.uk (Chris Malcolm) (12/31/89)
Stevan Harnad wrote: > cam@aipna.ed.ac.uk (Chris Malcolm) wrote: >> my game is assembly robotics... The assembly agent is designed not only >> to succeed in its tasks, but to present a suitable virtual world to the >> planner, there is an extra constraint on the task modularisation. That >> constraint is sometimes referred to as the symbol grounding problem. > As the mint that coined the term, I think I speak with a little > semantic authority when I say that that's decidedly *not* the symbol > grounding problem but rather a symptom of it! Virtual worlds are not > real worlds, and what goes on inside a simulation is just meaningless > symbol crunching. The only way to ground symbols is in the real world. I think Stevan has taken me to mean the opposite of what I intended! When I said "virtual world" I meant something which bore the same relationship to the real world as does a virtual machine (such as a software code intepreter) to a real machine (such as the computer running the interpreter. In other words, this kind of "virtual world" is just as real as the "real world", the "virtual" prefix meaning simply that we are referring to a level within the organisation of a creature which is hosted by the various physical and information processing mechanisms which underly it. At the top level the virtual world of a creature is "what it is like to be" that creature, the umwelten of von Uexkull. A virtual world in this sense is most emphatically not any kind of simulation. If, however, one presumes that the symbol grounding problem can be solved by the kind of perceptual mechanisms Stevan has outlined in his various papers, with the occasional qualification that a similar system can be devised for the motor side, then it is true that the kind of virtual world I am talking about becomes rather like a simulated world, such as in the often cited example of the aircraft simulator being hooked up to the controls and sensors of a real aircraft, and thus crossing the great divide between simulation of flight and really flying. The reason for this similarity is the separation of the sensor and motor hierarchies until they are combined at a symbolic level. This separation omits the powerful facilities available to a creature by combining sensing and motor activity to create kinds of sensing and action otherwise impossible (or much more computationally expensive). One example of this kind of facility is the servo-mechanism, which uses feedback to create - as far as the functioning creature is concerned - a useful and perceptible stability from the otherwise unstable and ephemeral. Another example is the use of an unstable motor activity which the environmental conditions will tend to drive into one of two appropriate limiting conditions. In this case the unstable motor activity is used as a combination of sensing and response system. There are many other kinds of ways in which sensing and action can be combined to advantage, or subsituted for one another. Wherever sensing and action have been locally amalgamated in this sort of way a barrier is created to the extension of separate sensor and motor processing hierarchies. While these processing hierarchies definitely exist, and are most important, they can't be extended beyond such a barrier. An independent set of such barriers can be seen as constituting a level. There can be many such levels erected atop one another. When I say "virtual world" in the context of a biological or artificial creature I refer to such a level. These virtual worlds depend on the active functioning of the real creature situated in the real world, and in that sense they are thoroughly real and grounded. Let me add the prediction that whereas one certainly can ground toy robots with architectures which involve separate perceptual and motor hierarchies erected on top of a basic bottom level of feedback control, that we will not be able to create "intelligent" but dumb (speechless) robots, let alone a robot which could pass the Total Turing Test, without multiple levels of such virtual worlds, each one creating new virtual sensors and effectors, in terms of which the next level can be constructed. In between these levels, I expect the kind of categorical perception described by Stevan Harnad, and its motor analogue, will play an important role. But, IMHO, it cannot, on its own, accomplish all that is required of symbol grounding in a complex intelligent creature. -- Chris Malcolm cam@uk.ac.ed.aipna 031 667 1011 x2550 Department of Artificial Intelligence, Edinburgh University 5 Forrest Hill, Edinburgh, EH1 2QL, UK
lee@uhccux.uhcc.hawaii.edu (Greg Lee) (01/02/90)
From article <1781@aipna.ed.ac.uk>, by cam@aipna.ed.ac.uk (Chris Malcolm): > ... >Wherever sensing and action have been locally amalgamated in this sort of >way a barrier is created to the extension of separate sensor and motor >processing hierarchies. ... This seems a central point, but I don't understand it. Why is a barrier created? (And is the barrier in the creature, or in our analysis of it?) Greg, lee@uhccux.uhcc.hawaii.edu
cam@aipna.ed.ac.uk (Chris Malcolm) (01/16/90)
In article <5871@uhccux.uhcc.hawaii.edu> lee@uhccux.uhcc.hawaii.edu (Greg Lee) writes: >From article <1781@aipna.ed.ac.uk>, by cam@aipna.ed.ac.uk (Chris Malcolm): ... >>Wherever sensing and action have been locally amalgamated in this sort of >>way a barrier is created to the extension of separate sensor and motor >>processing hierarchies. ... >This seems a central point, but I don't understand it. Why is >a barrier created? (And is the barrier in the creature, or >in our analysis of it?) Good point. Now that I think it out carefully, it is not in fact a barrier, so much as an opportunity not be missed. One _could_ just go on extending the sensor and processor hierarchies, but the point is that local amalgamation, as in a feedback servo, creates (when seen with the right timeframe and granularity) a useful new feature of the "world" (umwelt) which cannot be seen either as the result of pure sensory processing, or as as some kind of macro-operation on top of effector processing. To take advantage of such a feature adulterates the purity of the twin hierarchy. For a criticism of the "vision-module" approach to "the vision problem" from this kind of standpoint, see Aaron Sloman's article in the latest (vol 1 iss 4) Journal of Experimental and Theoretical AI. Yes, it's in the creature. While it _might_ be possible to build creatures with separated sensory and actuator processing hierarchies, it would at least be very computationally costly. -- Chris Malcolm cam@uk.ac.ed.aipna 031 667 1011 x2550 Department of Artificial Intelligence, Edinburgh University 5 Forrest Hill, Edinburgh, EH1 2QL, UK
lee@uhccux.uhcc.hawaii.edu (Greg Lee) (01/18/90)
From article <1838@aipna.ed.ac.uk>, by cam@aipna.ed.ac.uk (Chris Malcolm): " ... " >>Wherever sensing and action have been locally amalgamated in this sort of " >>way a barrier is created to the extension of separate sensor and motor " >>processing hierarchies. ... " " ... it is not in fact a " barrier, so much as an opportunity not be missed. I believe there are two issues involving separation/modularity versus amalgamation noticeable here. One concerns whether sensing and action are separable, and the other whether lower levels are separate from higher levels -- i.e. whether there is a hierarchy. I must admit to having only a foggy idea about the nature of these "barriers", but I thought you were saying that there is a relationship between these issues, to the effect that the need to process both input and output efficiently requires the development of a processing hierarchy. If something like that is so, it is pertinent to whether we can expect to find that higher levels of human processing, "thought", are sufficiently independent of human sensing and acting organs to be emulated by a program that runs on non-human hardware. On a small scale, these matters arise in the analysis of language, where one might see words as the barriers between phonological and syntactic processing. Greg, lee@uhccux.uhcc.hawaii.edu