park@usceast.cs.scarolina.edu (Kihong Park) (06/13/90)
I think the following points out the fallacy committed by Searle in a clear and simple fashion without need for "subjective" discussions: Searle's claim: Let M be a person who speaks only English. Let M reside in a room which has a Chinese(specialized) keyboard as well as a terminal which can display Chinese characters. Let M have access to a "rule" book B written in English which dictates what M is to type on the keyboard(answer) given some display of Chinese characters(question) on the terminal. Finally, let M follow the instructions of the book completely without reference to other extraneous sources and factors. Assuming book B exists based upon which the answers forwarded by M to questions in compliance with the above scenario pass the Turing test, it can be said that M with the help of B passes the Turing test without having understood the content of the discourse at all. Argument's fault: Searle is right in pointing out that in the above scenario M doesn't "understand" the content of the questions and answers at all. But this is precisely the merit by which modern day computers can be built. General purpose computers are physical realizations of Universal Turing machines. Universal Turing machines are devices which can simulate the behavior of other Turing machines. What a UTM basically does is perform sophisticated "book-keeping" operations, just as M does above. In doing so, the complexity of a problem solving procedure is separated from the necessary overhead needed for its implementation. The former is captured in what we call "programs" whereas the latter is hardwired in the hardware of the control unit of a "computer". But as every Computer Science student should know, programs can always be hardwired into the circuitry of a semi-conductor device. There is no essence in trying to distinguish software from hardware with respect to the final capability of the endproduct from a theoretical point of view. This is precisely where Searle makes a mistake. It's surprising that Searle should have aroused such controversy over a simple mistake, or lack of understanding of a basic result stemming back to 1936. If we belief that continuity and nondeterminism are nonessential aspects in the design of intelligent systems, then, yes, intelligent computers can in principle be built. The question is how to build one. If some people advocate that continuity and nondeterminism are absolutely essential properties of an intelligent system, well, then, it's a no-win situation. But that's not the topic of contention in the Chinese room argument.
frank@bruce.cs.monash.OZ.AU (Frank Breen) (06/13/90)
From article <3285@usceast.UUCP<, by park@usceast.cs.scarolina.edu (Kihong Park): < I think the following points out the fallacy committed by Searle in a clear < and simple fashion without need for "subjective" discussions: < < Searle's claim: < < Let M be a person who speaks only English. Let M reside in a room < which has a Chinese(specialized) keyboard as well as a terminal which can < display Chinese characters. Let M have access to a "rule" book B written in < English which dictates what M is to type on the keyboard(answer) given some < display of Chinese characters(question) on the terminal. Finally, let M < follow the instructions of the book completely without reference to other < extraneous sources and factors. < Assuming book B exists based upon which the answers forwarded by M to < questions in compliance with the above scenario pass the Turing test, it can < be said that M with the help of B passes the Turing test without having < understood the content of the discourse at all. < < Argument's fault: < Here's what I think is wrong with Searle's arguement. Of course M doesn't understand Chinese any-more than someone with half the speech centre of their brain missing would understand English. M is only a small part of the system - all the knowlege is stored in B and together M+B does understand Chinese. To put it another way - no single neuron in your brain understands anything - its only when you put them all together that there is any understanding, and-- No single part of the chinese room understands chinese - its only when you put it all together that it understands anything. It seems kind of obvious to me (now that I've thought of it) but I only caught the end of the arguement so I may be missing something. Tell me if I'm right. Frank Breen
park@usceast.UUCP (Kihong Park) (06/14/90)
In article <2410@bruce.cs.monash.OZ.AU> frank@bruce.cs.monash.OZ.AU (Frank Breen) writes: >Here's what I think is wrong with Searle's arguement. > >Of course M doesn't understand Chinese any-more than someone with half >the speech centre of their brain missing would understand English. >M is only a small part of the system - all the knowlege is stored in >B and together M+B does understand Chinese. To put it another way >- no single neuron in your brain understands anything - its only >when you put them all together that there is any understanding, >and-- No single part of the chinese room understands chinese - >its only when you put it all together that it understands anything. > >It seems kind of obvious to me (now that I've thought of it) but >I only caught the end of the arguement so I may be missing something. >Tell me if I'm right. > >Frank Breen Yes, you are basically right. But Searle is aware of this precise argument, in fact it is listed in his 1990 Sci. Am. article, yet still he dismisses it as incorrect. What you're doing above is basically falling into a trap whereby you are engaging in a discussion as to the validity of the statement that there is some fundamental difference between biological information processing systems such as the brain and any other artificial, mechanical counterpart. This is a statement which can't be proven or disproven at this very time. Either you accept it as a postulate or you don't. But he is making a different mistake in formulating his Chinese room argument which everybody can agree on to be faulty. Namely, his main point is that the person in the room(M), since he is essentially performing a table-lookup operation, does not understand the content of the question/answers. This is true. But from a theoretical point of view, given book B(program) and person M(control unit), there exists an equivalent Turing machine T which has B hardwired in its circuitry, and hence a pointing of fingers to the book-keeping entity M is not possible anymore. If you read his original articles, you will see that his Chinese room argument rests entirely on being able to point to M as the culprit. But his example is incorrect for the above reason. It's just a consequence of ther existence of Universal Turing machines. He probably would still like carry on with his core conviction of there being a fundamental difference between machines and brains, but he has to find another argument; Chinese rooms are impartial w.r.t. supporting his argument.
reynolds@bucasd.bu.edu (John Reynolds) (06/16/90)
In article <2410@bruce.cs.monash.OZ.AU> frank@bruce.cs.monash.OZ.AU
(Frank Breen) wrote:
Of course M [Searle] doesn't understand Chinese any-more than someone
with half the speech centre of their brain missing would understand
English. M is only a small part of the system - all the knowlege is
stored in B [the room] and together M+B does understand Chinese.
[I]ts only when you put it all together that it understands anything.
Tell me if I'm right.
park@usceast.UUCP (Kihong Park) replied:
Yes, you are basically right. But ... [w]hat you're doing above is
basically falling into a trap whereby you are engaging in a discussion
as to the validity of the statement that there is some fundamental
difference between biological information processing systems such as
the brain and any other artificial, mechanical counterpart.
reynolds@bucasd.bu.edu asks:
Am I missing something? That's not the trap he's falling into at all.
The idea that the components of a system acting in isolation may be
unable to carry out some functions they can achieve when working
together doesn't depend in any way on whether the components of that
systems are biological or not.
park@usceast.UUCP (Kihong Park) went on to add:
But he is making a different mistake in formulating his Chinese room argument
which everybody can agree on to be faulty. Namely, his main point is that
the person in the room(M), since he is essentially performing a table-lookup
operation, does not understand the content of the question/answers. This is
true. But from a theoretical point of view, given book B(program) and person
M(control unit), there exists an equivalent Turing machine T which has B
hardwired in its circuitry, and hence a pointing of fingers to the book-keeping
entity M is not possible anymore. If you read his original articles, you will
see that his Chinese room argument rests entirely on being able to point to
M as the culprit. But his example is incorrect for the above reason. It's
just a consequence of ther existence of Universal Turing machines.
reynolds@bucasd.bu.edu looks puzzled and types:
I don't see your point. And so by removing Searle, who you say
doesn't understand the content of the question/answers, and replacing
him and his book with circuitry you put intelligence into the system?
And in what way is M the culprit?cam@aipna.ed.ac.uk (Chris Malcolm) (06/27/90)
In article <36453@shemp.CS.UCLA.EDU> martin@oahu.cs.ucla.edu (david l. martin) writes: >In article <965@idunno.Princeton.EDU> markv@gauss.Princeton.EDU (Mark VandeWettering) writes: >> Searle's language is CRIMINALLY loose. Concepts such as understanding, >> causal powers, the distinction between syntax and semantics are >> not ever defined in any paper of his that I have read. >> his recent Scientific American article was not a "proof": he merely >> assumed that his conclusion was correct and proceeded. >Note that to the extent >that the above 2 claims were in fact made by AI researchers, they were the >ones who initiated the "loose" usage of concepts such as understanding, etc. Hold on a minute! "Understanding" etc. were being used "loosely" (in the sense of lacking precise definitions) by English speakers, psychologists, and philosophers long before AI was thought of, as were their cognates in other languages, such as Latin and Greek, long before the English language was invented. And lacking a precise definition is not necessarily a failure in a term: for example, despite centuries of wrangling, there is still no satisfactorily agreed definition of mathematics. -- Chris Malcolm cam@uk.ac.ed.aipna 031 667 1011 x2550 Department of Artificial Intelligence, Edinburgh University 5 Forrest Hill, Edinburgh, EH1 2QL, UK
forbis@milton.u.washington.edu (Gary Forbis) (06/27/90)
In article <25457@cs.yale.edu> blenko-tom@CS.YALE.EDU (Tom Blenko) writes: >Consider >again the artificial city, and suppose that someone does succeed in >constructing such a thing. So the artificial city replicates >externally, as closely as anyone can tell, just the behaviors that a >real city would. I'm not sure you believe this. I am pretty sure you do not. Please look again at your quote which follows. >Now, the question is, does the artificial city have "civic pride"? But >the architects of the artificial city are only concerned with inputs >and outputs, and when they deliver the desired transfer function, they >suppose, using your view, that they are finished. So there's no reason >for anyone to suppose that it's meaningful to talk about the civic >pride of an artificial city. It does not make sense to say a city has been replicated then say it has not. If an observer can tell the difference between the real and the artificial then as far as this city goes it has failed the Turing Test. If the architects are concerned with making the replica indestinguish- able from the real then if civic pride is important it must be replicated. > >Searle takes consciousness >and emotional states to be properties of the mind. His claim (indeed >his solution to the mind-body problem) is that these intentional >properties are identically the states of the underlying processor... Is Searle really a functionalist? I don't understand how he could be and still dispute the claims of Strong AI. --gary forbis@milton.u.washington.edu
llama@eleazar.dartmouth.edu (Joseph A. Francis) (06/28/90)
In article <25457@cs.yale.edu> blenko-tom@CS.YALE.EDU (Tom Blenko) writes: >Just as civic pride is a property of a city, Searle takes consciousness >and emotional states to be properties of the mind. His claim (indeed >his solution to the mind-body problem) is that these intentional >properties are identically the states of the underlying processor -- >that when one reports that someone is hungry, for example, one is >saying nothing other than that the current state of her neurons is one >element of the subset of all neuron states that has been labelled >"hungry". Therefore, he claims, the information-transducing properties >of an intelligent artificial entity do not suffice -- the artificial >entity must also reproduce the relationship between physical states and >mental states. Ahh. And here is the crux of the whole matter (to me at least). While others cite a variety of assumptions Searle makes in CR as the straw that breaks CR's back, I believe the following is a more telling problem: I don't think the CR can function as per CR WITHOUT having mental states. For instance, we would not say something passes the turing test if it can never remember the last question you asked it - so clearly the inards of CR do not have static comments - the man in CR must not be just reading symbols - applying rules from the book - and outputing symbols; he must also be WRITING things (in the book or on scrath paper or somewhere). Also, this system as a whole must be able to learn new Chinese words, learn how to play crazy eights and tic-tac-toe, and formulate opinions on the validity of Searle's CR argument, etc... The inards of CR are an extremely active, complicated, and evolving place. I claim (without any support - just a claim) that doing all this neccessitates mental states, self awareness, thought, and so on. So in essence, if Searle insists his CR has no mental states, I assert CR is impossible. If Searle doesn't mind attributing mental states to CR, then fine, but now he'll have to grant CR the property of intelligence. -Joe
blenko-tom@CS.YALE.EDU (Tom Blenko) (06/28/90)
In article <4490@milton.u.washington.edu> forbis@milton.u.washington.edu (Gary Forbis) writes: |It does not make sense to say a city has been replicated then say it |has not. If an observer can tell the difference between the real and |the artificial then as far as this city goes it has failed the Turing |Test. If the architects are concerned with making the replica indestinguish- |able from the real then if civic pride is important it must be replicated. A city has extentional properties (resources it consumes, products it produces) and intentional properties (I suggest civic pride as an example). The analogy is made to the mind. Searle says input/output relations do not suffice to reproduce the mind because they capture extentional properties while neglecting intentional properties (of which hunger might be an example). And he takes hoping, fearing, loving, hungering, and so forth, which are not objectively observable, to be essential and intentional states of any mind. |>Searle takes consciousness |>and emotional states to be properties of the mind. His claim (indeed |>his solution to the mind-body problem) is that these intentional |>properties are identically the states of the underlying processor... | |Is Searle really a functionalist? I don't understand how he could be |and still dispute the claims of Strong AI. I don't know what you mean by "functionalist". Certainly he is a physicalist. And (one of) his arguments against strong AI is that intentional properties arise not just from the program but from the processor, as I've outlined previously. Tom
kenp@ntpdvp1.UUCP (Ken Presting) (06/28/90)
In article <965@idunno.Princeton.EDU>, markv@gauss.Princeton.EDU (Mark VandeWettering) writes: > In article <593@ntpdvp1.UUCP> kenp@ntpdvp1.UUCP (Ken Presting) writes: > > > >Searle gets a lot of heat for using loose language, but there are > >important cases where he says just what he means, in no uncertain > >terms. > > Searle's language is CRIMINALLY loose. Concepts such as understanding, > causal powers, the distinction between syntax and semantics are > not ever defined in any paper of his that I have read. In particular, > his recent Scientific American article was not a "proof": he merely > assumed that his conclusion was correct and proceeded. Don't bother with the places where his language is loose. If we are serious about wanting to defeat this argument, we need to look at those few places where he is clear. We are all familiar with the process of "desk debugging" - simulating a program by hand to trace it's operation. Searle is not in a vacuum at UC Berkeley. He has talked to *plenty* of programmers, and knows about hand simulation. The CR does NOT get its staying power from the vagueness. This tar baby is sticky because it's based on everyday common sense. A programmer can runs through the steps of his program just as well (if not as fast) as a chip. And once you've "learned" a foreign language, you can do a lot more than just generate replies to written notes. But we need a stronger solvent than common sense. Let's talk logic: > >For example, he says that the CR example itself is presented only > >to show that semantics is not reducible to syntax. > > ... which has NOT been shown at all ... Searle doesn't really need to argue for this - he just wants a compelling example. Tarski has already proved it, and it is quite beyond debate. Tarksi's theorem states that no predicate P can satisfy the following criterion for al sentences S: P('S') is provable if and only if 'S' is true. Note that this criterion does not involve the provability of S, but only the provability of a sentence *about* S. Truth is a semantic property, while any "P" in the theorem is a syntactic property (because it takes a quoted sentence as its operand), so Tarski shows that truth cannot be reduced to any syntactic property. Pat Hayes has correctly pointed out that programs are more than syntax, which is very important. A single floppy disk with my program on it is physically different from the same disk with your program on it, and running the different programs will produce physically distinguishable output. This issue is important for the question of whether programmed computers have any specific "causal powers," but is independent of the question of understanding. > Searle is trying to prove the following: > > > For any program P whatsoever, and for any machine M whatsoever, > > the following inference is always invalid: > > > Machine M runs Program P, therefore Machine M understands. > > > . . . Searle is *not* trying to show that no program can > >think, or that no machine can think. He is too clever for that. He > >is not attacking the *goal* of Strong AI. > > Interesting distinction, but why would ever use the fact that program A > causes "understanding"? The only really valid test for understanding is > demonstrating it. > Suppose you want to sell your latest "Conversational Chinese" program. Wouldn't you like to claim that running your program will make the computer speak Chinese? It's a question of whether the way to make an AI is by writing programs or by building machines (or some combination). Strong AI says (according to Searle, and I think he is not far off) that given the right program, any machine that is big enough to run it will actually be intelligent, while the program is running. I think this claim is true, but Searle does not. He thinks something else must be said about the machine before we can conclude that it understands. Perhaps you agree with Searle? > >He is attacking the *argument* behind Strong AI. This is much easier > >to do, but almost as devastating. Write any program you want, and > >run it any way you want, on any hardware, parallel, serial, or cerebral. > >But if you want to claim that the system is thinking, you'll need a > >better reason than "It's running my program". > > Searle's Chinese Room was designed to attack the concept of the Turing test > as a valid test for intelligence. It simply fails. To Searle, it makes > no sense to say that the Chinese Room _understands_ Chinese. This is simply handwaving. Look - when I'm in my office, my office could pass the Turing test in English, but it still makes no sense to say that my office understands English. If you want to claim that Searle defies common sense in *his* argument, then you better not defy common sense in your *own*! Go back to the logic of the problem. Take any programmed Turing machine that can pass the Turing test. This machine can do nothing more than implement a syntactic algorithm, since the input tape can never contain anything other than a string of symbols. Therefore, by Tarski's theorem, (and Church's thesis) the machine could represent a syntactic predicate but not a semantic predicate, such as truth. Now, *whatever* it is that constitutes understanding, it must have something to do with knowing what words mean, and that certainly requires semantics. We have just seen that no Turing machine can represent the concept of "truth", so there is at least one word which no Turing machine can understand. This is a different conclusion than Searle wanted, but that's because I have used slightly different premises (getting Searle's original conclusion would require a more general concept of "semantic information"). I hope we can put an end to the vague language of both sides, in this thread. Ken Presting ("Let us calculate")
jim@se-sd.SanDiego.NCR.COM (Jim Ruehlin, Cognitologist domesticus) (06/29/90)
In article <25457@cs.yale.edu> blenko-tom@CS.YALE.EDU (Tom Blenko) writes: >Well, this is exactly one of the things Searle is disputing. Consider >again the artificial city, and suppose that someone does succeed in >constructing such a thing. So the artificial city replicates >externally, as closely as anyone can tell, just the behaviors that a >real city would. >If we go to a real city, we can pretty well arrive at an opinion about >how much "civic pride" it has. It is reflected in various tangible >elements of the city (parks, libraries, services) and less tangibly in >the attitudes and dispositions of its human inhabitants. >Now, the question is, does the artificial city have "civic pride"? But >the architects of the artificial city are only concerned with inputs >and outputs, and when they deliver the desired transfer function, they >suppose, using your view, that they are finished. So there's no reason >for anyone to suppose that it's meaningful to talk about the civic >pride of an artificial city. You may be carrying the metaphor of the city a little too far. What's the corelation between "civic pride" and something that happens in neurons (it's something intangible, I assume, but what?). And simply because cities have "civic pride" is no reason to assume that neurons have a coresponding phenomenon. In any case, we can linguistically define "civic pride". It has certain effects and, presumably, causes. It manafests itself in a certain way. As long as we can define a property like this, we can simulate it or duplicate it. - Jim Ruehlin
blenko-tom@CS.YALE.EDU (Tom Blenko) (06/30/90)
In article <3431@se-sd.SanDiego.NCR.COM> jim@se-sd.SanDiego.NCR.COM (Jim Ruehlin, Cognitologist domesticus) writes: |You may be carrying the metaphor of the city a little too far. What's |the corelation between "civic pride" and something that happens in |neurons (it's something intangible, I assume, but what?). And simply |because cities have "civic pride" is no reason to assume that neurons |have a coresponding phenomenon. The point about civic pride is that it ultimately is just a disposition shared by the inhabitants of a city. Concrete manifestations, (e.g. parks) may be taken as evidence for civic pride, but they can arise in the absence of civic pride, and they need not arise in the presence of civic pride. Searle's mind/brain hypothesis is that mental states (e.g. consciousness, hunger) are simply labels for collections of neural states. So, just as it is difficult to talk about the civic pride of a "city" lacking nearly-human inhabitants, it is difficult to talk about the "hunger" of a system defined by a program running on an arbitrary processor. Mental states are taken as a necessary property of a mind, artificial or otherwise. |In any case, we can linguistically define "civic pride". It has certain |effects and, presumably, causes. It manafests itself in a certain way. |As long as we can define a property like this, we can simulate it or |duplicate it. Searle thinks this view represents a major (and commonplace) misunderstanding (so do I). Let's accept that you can linguistically define "civic pride". Now, how do you duplicate it without using humans as elements of the system duplicating it? There are lots of other examples. You can simulate the aerodynamic properties of an aircraft design -- but there are always details missing, and some of them may prove critical to the aircraft's performance. Similarly, if you can provide an accurate simulation of the economy, which simply represents the aggregate behavior of a group of more-or-less independent actors, you can easily become the first billionaire on your block. Tom
Victoria_Beth_Berdon@cup.portal.com (07/02/90)
(Note: this article is being posted for Ken Presting) > In article <593@ntpdvp1.UUCP>, kenp@ntpdvp1.UUCP (Ken Presting) writes: > > [...stuff deleted...] > > Searle is *not* trying to show that no program can > > think, or that no machine can think. He is too clever for that. He > > is not attacking the *goal* of Strong AI. > > > > He is attacking the *argument* behind Strong AI. This is much easier > > to do, but almost as devastating. Write any program you want, and > > run it any way you want, on any hardware, parallel, serial, or cerebral. > > But if you want to claim that the system is thinking, you'll need a > > better reason than "It's running my program". > > Ken, the only article I have read by Searle was the one in the January > 1990 Scientific American. In that article, it sure seems to me that > Searle is claiming that the Strong AI position is provably wrong, not > only that arguments in its favor are incorrect. This is a very subtle point, and well worth getting straight. In the second paragraph of the Sci.Am. article, Searle says: The question that has been posed ... is, Could a machine think just by virtue of implementing a computer program? On page 27, his Conclusion 1 is: Programs are neither constitutive of nor sufficient for minds. Finally, later on the same page he says: Third, strong AI's thesis is not that, for all we know, computers with the right programs might be thinking, that they might have some as yet undetected psychological properties; rather, it is that they must be thinking because that is all there is to thinking. So you are right that Searle thinks that Strong AI is provably wrong. But I want to emphasize that for Searle, "strong AI" is *not* the straightforward claim that computers can think (someday). It is not even the slightly more sophisticated claim that the right program will make it possible for computers to think. Searle is attacking the claim that "running the right program, by itself, will CERTAINLY make computers think." When he uses phrases like "just by virtue of ..." or "that's all there is to ...", Searle is saying two things: The thesis of Strong AI is not just an atomic sentence, such as "Computers can Think", it is actually an inference, such as, "Programs can make computers think, BECAUSE <you name it>". The inference of Strong AI depends on exactly one premise, that the machine is running a certain program. No other premises are allowed. > . . . His Chinese room > argument seemed to be an attempt to prove that *no* program that only > uses symbolic manipulation can ever be said to understand. You are > certainly right that Searle did not go so far as to claim that no > machine can understand, but he sure seemed to be claiming that any > such hypothetical machine must be doing more than symbol manipulation. I guess we pretty much agree on this whole issue, but I thought I should beat it into the ground ... > > Searle's argument seemed to boil down to "if the man doing the > symbol-manipulation doesn't understand, then the Chinese room (man + > rules) doesn't understand", which is equivalent for computers to "if > the cpu doesn't understand, then the cpu running a program doesn't > understand". That claim, if you buy it (I don't) seems to me to > completely rule out the possibility of a computer understanding > anything. > > Daryl McCullough Searle does say (p.27) that he "has not tried to prove that 'a computer cannot think'", so I would say that if you are reading his argument in a way that commits him to the stronger position, you may want to look again. What makes Searle's weaker point (that mind cannot be inferred from programming) interesting is that he does not need to say that Strong AI will fail - but he is saying that there cannot be any reason to believe that it will succeed. That is, I'm not claiming that Searle is trying to undermine any arguments in favor of the goal of Strong AI being possible. What Searle is attacking can be viewed as the pratical arguments behind strong AI as a research program. "We want a smart computer, so let's write a program to make computers smart." Searle is saying that you can program from now till doomsday, and then Totally Turing Test The resulTs for a Thousand lifeTimes. But that will *not* entitle us to conclude that the computer is smart. So in addtion to writing programs, we should be doing something else, presumably related to "causal powers". > In article <593@ntpdvp1.UUCP> kenp@ntpdvp1.UUCP (Ken Presting) writes: > |precise. Searle is trying to prove the following: > | > | For any program P whatsoever, and for any machine M whatsoever, > | the following inference is always invalid: > | > | Machine M runs Program P, therefore Machine M understands. > > In article <25422@cs.yale.edu>, blenko-tom@CS.YALE.EDU (Tom Blenko) writes:> > This is much too strong, and you are arguing against yourself. Searle > claims that functional equivalence does not suffice, that intelligence > is an intensional property. This is presented as a counter to the > "machine-independent" property he ascribes to strong AI advocates. Why do you think this is too strong, or that I'm arguing against myself? > > I believe he makes the claim about biological versus silicon > implementations in his first paper, and I've certainly heard him make > that claim in person. On p.27 of the Sci. Am. article, Searle says: Second, I have not tried to show that only biologically based systems like our brains can think. Right now, those are the only systems that we know for a fact can think, but we might find other systems in the univers that can produce conscious thoughts, and we might even be able to create thinking systems artificially. I think this is a clear statement. Ken Presting ("Burn AFTER reading")
blenko-tom@CS.YALE.EDU (Tom Blenko) (07/03/90)
In article <31329@cup.portal.com> Victoria_Beth_Berdon@cup.portal.com writes: |(Note: this article is being posted for Ken Presting) |... |> In article <593@ntpdvp1.UUCP> kenp@ntpdvp1.UUCP (Ken Presting) writes: |> |precise. Searle is trying to prove the following: |> | |> | For any program P whatsoever, and for any machine M whatsoever, |> | the following inference is always invalid: |> | |> | Machine M runs Program P, therefore Machine M understands. |> |> In article <25422@cs.yale.edu>, blenko-tom@CS.YALE.EDU (Tom Blenko) writes:> |> This is much too strong, and you are arguing against yourself... | |Why do you think this is too strong, or that I'm arguing against myself? | It reads as FORALL P FORALL M NOT(M(P) ==> M understands) which says that no program running on any machine results in a machine that "understands" (should be system that understands). Searle's claim is closer to saying there is no universal intelligent program, i.e., NOT(EXISTS P FORALL M M(P) ==> M(P) is intelligent) which is logically equivalent to the much weaker (than yours) assertion FORALL P EXISTS M NOT(M(P) ==> M(P) is intelligent) I say that you are arguing against yourself because you attribute this claim to Searle (and the informal one it is intended to capture, saying that Searle denies the "relevance" of programs), yet it is at odds with your acknowledgement that Searle is not arguing against the possibility of an intelligent, artificial entity. |> I believe he makes the claim about biological versus silicon |> implementations in his first paper, and I've certainly heard him make |> that claim in person. | |On p.27 of the Sci. Am. article, Searle says: | | Second, I have not tried to show that only biologically | based systems like our brains can think. Right now, those | are the only systems that we know for a fact can think, but we | might find other systems in the univers that can produce | conscious thoughts, and we might even be able to create | thinking systems artificially. | |I think this is a clear statement. My point was that Searle believes not only the program, but the implementing processor, contribute essential properties to the resulting entity. Therefore it is relevant whether the implementing processor/system consists of hardware, software, or wetware. Tom
daryl@oravax.UUCP (Steven Daryl McCullough) (07/03/90)
In article <593@ntpdvp1.UUCP> kenp@ntpdvp1.UUCP (Ken Presting) writes: > Searle is trying to prove the following: > > For any program P whatsoever, and for any machine M whatsoever, > the following inference is always invalid: > > Machine M runs Program P, therefore Machine M understands. > If Searle were only trying to show that the inference above is invalid, then I would have no further argument with him; he would be correct. Furthermore, his Chinese Room argument would indeed be a convincing argument: If Machine M is the man in the Chinese room, then for any program P, the man could run program P and still not understand Chinese. However, the validity of the above inference is not claimed by Strong AI (or if it is, then they are just speaking loosely). The more precise claim would be that, for the right program P, one can infer Machine M runs Program P, therefore the system (Machine M running Program P) understands. This is closer to the strong AI position, and it seems that Searle has no good argument against it. For the Chinese room to count as an argument against this claim, it would be necessary to establish that the system (man + rules + room) does not understand Chinese. And Searle cannot establish this without offering *some* definition of what it means for a system to understand. (Comment: Searle's variant of having the man memorize the rules does not change anything; there would still be two systems: the man "acting himself" and the man following the rules. Establishing that one system does not understand does not automatically establish that the other doesn't.) Daryl McCullough
kenp@ntpdvp1.UUCP (Ken Presting) (07/10/90)
> |> In article <593@ntpdvp1.UUCP> kenp@ntpdvp1.UUCP (Ken Presting) writes: > |> |. . . Searle is trying to prove the following: > |> | > |> | For any program P whatsoever, and for any machine M whatsoever, > |> | the following inference is always invalid: > |> | > |> | Machine M runs Program P, therefore Machine M understands. > > Tom Blenko writes: > This is much too strong, and you are arguing against yourself... > . . . > It reads as > > FORALL P FORALL M NOT(M(P) ==> M understands) > > which says that no program running on any machine results in a machine > that "understands". . . . Tom, you are mistaken. You have overlooked the distinction between "valid inferences" and conditional assertions. In standard symbols, my version of Searle's thesis would read: (P) (M) - ( M runs P |= M understands ) The "|=" symbol denotes the logical relation called "entailment". The simple conditional form which you use here ignores Searle's repeated use of conjunctions like "must" and "simply by virtue of", which indicate a *necessary* relation between the antecedent and consequent. (I have neglected the object- vs. meta-language issue in my formula, but that should not lead to much confusion. I have also avoided the standard modal interpretation of "necessity", which should positively reduce confusion.) Since entailment is a stronger relation than implication, the negation of an entailment is weaker than than a negation of an implication, and my version of Searle's claim has similar truth conditions to the version you propose below. Since Searle is claiming (on my reading of him) that the running of any program will not *necessitate* the presence of understanding in any machine, he can proceed in two steps, the first of which is identical to your proposal: > . . . Searle's position > is closer to saying there is no universal intelligent program, i.e., > > NOT(EXISTS P FORALL M M(P) ==> M(P) is intelligent) > > which is logically equivalent to the much weaker (than yours) assertion > > FORALL P EXISTS M NOT(M(P) ==> M(P) is intelligent) > Notice that for Searle to support this last claim, he needs to demonstrate the existence of a single Machine such that no matter what Program it is running, it will not understand Chinese. He thinks he has done so with the Chinese Room. Perhaps he has not, but that is another question. If the CR example is successful, then he has his first step. Perhaps the difference between your reading of Searle and mine comes to this: You have formulated the conditions which he tries to meet with the CR example itself, while I am attempting to formulate the general conclusion for the argument of which the CR example is a part. The second step requires the application of a rule of inference which is analogous to "Universal Generalization" in natural deduction systems. If Searle is granted the assumption that there is no relevant difference between the case of a computer running a program and himself running the same program, then he can conclude for all machines that there is no necessary connection between the program it runs and its understanding. Searle thinks this assumption follows trivially from "Axiom 1: Programs are purely formal". Pat Hayes denies the assumption (with some justice, I think, but the issue is not simple). > I say that you are arguing against yourself because you attribute this > claim to Searle (and the informal one it is intended to capture, saying > that Searle denies the "relevance" of programs), yet it is at odds with > your acknowledgement that Searle is not arguing against the possibility > of an intelligent, artificial entity. > Even on your own formulation of my reading, this does not follow. From (P)(M) - ( Runs(M,P) -> Understands(M) ) it does not follow that (M) - ( Understands(M) ). All that follows is that running a certain program is not a sufficient condition for understanding. Now, you may object that if the question "What program is that machine running?" is not enough to decide the issue of the machine's intelligence, then no amount of additional information could ever establish that a general-purpose computer is intelligent. Many people do believe this (the Churchlands seem to), and propose that Connectionism is the only hope of AI. Whatever the status of that issue, the Chinese Room does not, by itself, establish that no programmed general purpose computer can understand. If it establishes anything, it establishes *only* that we must know more about a computer than what program it is running, before we draw any conclusions about its intelligence. > Tom > Thanks for your comments. I especially appreciate the formal direction you have given to this thread. If we can keep this up, we may get somewhere. Ken Presting ("Metastasis Before Modality")
dave@cogsci.indiana.edu (David Chalmers) (07/13/90)
In article <597@ntpdvp1.UUCP> kenp@ntpdvp1.UUCP (Ken Presting) writes: > >> Tom Blenko writes: > >> FORALL P EXISTS M NOT(M(P) ==> M(P) is intelligent) > >Notice that for Searle to support this last claim, he needs to demonstrate >the existence of a single Machine such that no matter what Program it is >running, it will not understand Chinese. Just for the record, this is fallacious. Such a strategy would be sufficient to support the claim, but not necessary. Take another look at the order of the quantifiers. Talk of "machines" tends only to confuse the issue, anyway. All we need is the notion of *program* (a formal object), and *implementation of program* (a physical system). It's not clear that all implementations will be describable as running on pre-existing machines. In this framework, the strong AI claim becomes: EXISTS P (program) such that FORALL S (physical system): S is an implementation of P => S is intelligent. Actually, even this may be too strong. Some might like to say "S produces intelligence" rather than "S is intelligent" -- the question of the "ownership" of the intelligence is somewhat vague. e.g. is your *brain* intelligent?; is your *body*?; such technical questions don't need to be answered to deal with Searle's argument. Anyway, with this in place, Searle needs to show FORALL P, EXISTS S such that S is an implementation of P but S does not produce intelligence, which is what the Chinese Room purports to show. Of course it doesn't show that, but that's another story. Suffice to reiterate the often-made point that the fact that the pre-existing machine (i.e. the person in the room) that implements the program fails to understand is quite irrelevant. Implementing machines aren't what counts: implemented systems are. >program, then he can conclude for all machines that there is no necessary >connection between the program it runs and its understanding. Searle thinks >this assumption follows trivially from "Axiom 1: Programs are purely formal". >Pat Hayes denies the assumption (with some justice, I think, but the issue >is not simple). Actually, I think that programs are indeed purely formal (or purely syntactic, or whatever you like). However, *implementations of programs* certainly aren't. They're concrete physical systems with all kinds of interesting internal causal structure. The fallacy of "programs are purely syntactic, minds are semantic, syntax isn't sufficient for semantics; therefore implementing an appropriate program cannot be sufficient to produce a mind" argument is best brought out by a corresponding argument: (1) Recipes are completely syntactic. (2) Cakes are tasty (or crumbly, or heavy, or...) (3) Syntax is not sufficient for tastiness (or crumbliness, or heaviness...) (4) Implementing the appropriate recipe cannot be sufficient to produce a cake. I hope that even Searle would see the fallacy here. Recipes are syntactic, but *implemented recipes* are not. Of course, one needs a meaningful interpretation procedure to go from the recipe (formal specification) to the cake (physical implementation). But one has such a procedure (it's hanging around in the head of (good) cooks, and could presumably be mechanized.) Exactly the same goes for programs. Programs are syntactic, implemented programs are not. Implemented programs are physical systems, derived from formal programs through an interpretation procedure (either a compiler or an interpreter, in practice, or both.). The role of the compiler/interpreter is precisely analogous to the role of the chef. >Now, you may object that if the question "What program is that machine >running?" is not enough to decide the issue of the machine's intelligence, >then no amount of additional information could ever establish that a >general-purpose computer is intelligent. Many people do believe this >(the Churchlands seem to), and propose that Connectionism is the only >hope of AI. This statement seriously misconstrues the nature of connectionism. The issue of Connectionism vs. Traditional AI is quite orthogonal to the issue of Strong AI vs. Searle. Personally, I'm a dyed-in-the-wool connectionist (or, more generally, a subsymbolic computationalist), but I'm also a dyed-in-the-wool Strong AI supporter. The two positions are quite compatible. Most connectionists believe that implementing the right program is enough to give you intelligence -- they just happen to believe that the program you need will be of a particular kind, compatible with the principles of connectionism. The notion that connectionism rejects, say, the Turing notion of computation is quite prevalent in some circles, and can even be found in print from time to time. It's quite fallacious, though. Personally, I think that the Turing notion of computation is the greatest thing since sliced bread. It's just that people in traditional AI placed far too heavy a restriction on the kind of computations they allowed in (by making a deep prior commitment about the ways in which computational states could carry semantics). Connectionism advocates removing this heavy semantic commitment (note: it doesn't advocate removing semantics, it just remains silent about the level at which the semantics might lie), and thus returning to the full-fledged, unrestricted class of computations that Turing allowed. Most connectionists believe in Strong AI, without a doubt. Only the *class* of sufficient programs is in dispute. Sorry about this... and I had vowed "never again". Chinese-Room withdrawal symptoms, I guess. One of these days I'm going to write a paper called "Everything You Wanted to Know About the Chinese Room but Were Afraid to Ask". Searle's arguments are deeply fallacious, but they raise an enormous number of interesting issues. -- Dave Chalmers (dave@cogsci.indiana.edu) Concepts and Cognition, Indiana University. "It is not the least charm of a theory that it is refutable"
kenp@ntpdvp1.UUCP (Ken Presting) (07/18/90)
In article <50741@iuvax.cs.indiana.edu>, dave@cogsci.indiana.edu (David Chalmers) writes: > > Talk of "machines" tends only to confuse the issue, anyway. All we need > is the notion of *program* (a formal object), and *implementation of program* > (a physical system). It's not clear that all implementations will be > describable as running on pre-existing machines. > > In this framework, the strong AI claim becomes: > > EXISTS P (program) such that FORALL S (physical system): > S is an implementation of P => S is intelligent. > This version of the Strong AI claim is anticipated by Searle, on p.29 of the Sci.Am. article: The thesis of Strong AI is that any system whatsoever ... not only might have thoughts and feelings, but _must_ have thoughts and feelings, provided only that it implements the right program, with the right inputs and outputs. Notice two (small) differences: 1) Searle is insistent on the issue of programs "constituting" minds, and physical objects "causing" thought. So the arrow above must not be read as simple material implication - it must be necessary implication, or entailment, or some other counterfactual (eg causal). 2) As I read him, Searle is lumping together all automata which compute the same function, independently of the algorithm they use. If I follow Daryl McCullough's last article, he for one would not accept this version of Strong AI. Probably few would. I myself balk at (2), because I believe a thinking thing must contain a representation of the concept of "truth", which cannot be finitely defined in I/O terms. Searle is perfectly willing to face Strong AI defined in terms of "implemented systems", but there is probably a difference between his concept of implementation and Dave's. The usual software engineering concept of "implemented system" involves: a) an independently pre-existing "machine" which can run most any "program" b) a "machine-readable" copy of a program c) a mechanical process for "loading" the program into the machine d) a user-initiated process of "running" the program. Searle seems to view "implementation" in this fashion, which I'll call GOFR, for "Good Old Fashioned Running" (apologies to John Haugeland). Note that Hilary Putnam claims to have shown that every physical system is an *instantiation* of every finite automaton. Unless Dave's concept of "implemented system" can be distinguished from Putnam's "instantiated automata", Searle could object that Dave's version of Strong AI implies panpsychism. "Go ahead," Searle would say, "write your magic program. Now find a system that *doesn't* implement it, or else explain why all these implementations lying around on the ground still act so stupid." GOFR is not subject to Putnam's argument, because to identify an object as a "machine" requires the concurrent specification of the processes of loading and running the program, and a coding scheme for machine- readable copy. (See _Representation and Reality_ for the argument) This is the first problem with the Systems Reply - the high powered abstract concept of implementation reduces Strong AI to an absurdity, while the traditional GOFR concept does not neatly excise the "pre-existing machine" and its gripes about not understanding its data. > > Anyway, with this in place, Searle needs to show > > FORALL P, EXISTS S such that S is an implementation of P but S does not > produce intelligence, > > which is what the Chinese Room purports to show. Of course it doesn't show > that, but that's another story. Suffice to reiterate the often-made point > that the fact that the pre-existing machine (i.e. the person in the room) > that implements the program fails to understand is quite irrelevant. > Implementing machines aren't what counts: implemented systems are. Let me try to give a formal analogue of this objection, in the forlorn hope of clarifying the issue once and for all. Let U(n,m) be the function computed by a Universal Turing Machine, where 'n' is the Goedel number of an arbitrary TM, and 'm' is the Goedel number of an arbitrary starting configuration of an input tape. Let C(m) be the function computed by a TM that, when implemented, can pass the Turing Test in Chinese. Let 'c' be the Goedel number of some TM that computes C(m). Finally, suppose for a moment that we have made sense of the concept of "implementation", and let S and R be implements. Note first that on any acceptable concept of "implementation": For any S, S implements U(c,m) if and only if S implements C(m). Now, according to Systems Repliers, all that Searle shows is: There are S, R such that S implements U(n,m) and R implements U(c,m) and S does not think. And of course, this is irrelevant. We only care whether R thinks. If I finally got it straight, this is the gist of Bob Kohout's "Viola" (:-) article last spring. NOBODY, NOT EVEN SEARLE, IS THAT STUPID. The Putnam-based objection I gave above is never raised by Searle, because he has a more straightforward counter on p. 30 of the Sci.Am. article: The point of the original argument was that symbol shuffling by itself does not give any access to the meanings of the symbols. But this is as much true of the whole room as it is of the person inside. Searle's point is that there is nothing a Universal TM can do with a program that he cannot do just as well himself. When he says "But I still don't understand Chinese", he is not just reporting a subjective state of ignorance. He is correctly emphasizing that the syntactically specified operations he is performing on the symbols are unrelated to their semantics. Everyone agrees that Searle-without-books not understanding is completely irrelevant. The controversial issues are: Problem 1: How much do we have to add to Searle in order to get an "implemented system", and what "causal powers" will that system have? Problem 2: Once we have an "implemented system", what connection, if any, is there between the operations of the system and the semantics of the symbols? There are (min) two distinct threads in the CR debate. One is not an issue for AI at all - Problem 1. General Computer Science should be able to handle that issue, but not until the semantics of programs is understood and the concept of implementation is cleared up. Problem 2 is specific to AI, but it can only be studied if we make some general assumptions about how symbols get their meanings. IMO, we can get these assumptions from Quine and Davidson, so there's no need to think in a vacuum. Ken Presting ("Anybody else abhor a vacuum?")