wsaba@cscs.uwindsor.ca (Walid Saba) (06/14/91)
It seems to me that the psychology - philosophy debate in AI is
not leading anywhere. The reason is that it is not discussing the
real issue. The question should be "what sort of reasoning do
we need to simulate, in order to simulate intelligence?". The
debate should not be about the question of whether we should
use logical reasoning or not? OF COURSE WE SHOULD! But what
sort of reasoning? is the right question.
For example, when being told
Sam is pregnant.
an "intelligent" system should infer some other facts.
For example,
=g=> Sam is a female
=g=> Sam is alive
=g=> Sam's age is within a certain range
(depending on the context, indicating
whether Sam is human, or perhaps a cat).
...etc.
Is this "logical" reasoning? Of course.
Whatever `logic' is used (default, common sense, etc.), we
are using some logic. As a matter of fact, every formalism
has its own logic.
The point here is that the argument against some (constrained)
logics should not be extrapolated to the point were we ignore
the simple fact that ANY FORMALISM HAS ITS LOGIC. How suitable
this logic is beside the point.
Did we ever stop to think how we (humans) compute the square
root of a number. We know that the routines we built compute
the square root in nano-seconds, and everything is fine...
Is it? Well, these are numerical computations and in that
sphere things are manageable. But couldn't the same be true of
symbolic computations??
Walid Saba
School of Computer Science
University of Windsor
Windsor, Ontario, Canada N9B 3P4
(519) 253-4232 ext. 3003/3005
email: wsaba@cs.uwindsor.cacam@aifh.ed.ac.uk (Chris Malcolm) (06/17/91)
In article <549@schoenfinkel.cscs.uwindsor.ca> wsaba@cscs.uwindsor.ca (Walid Saba) writes: >The >debate should not be about the question of whether we should >use logical reasoning or not? OF COURSE WE SHOULD! But what >sort of reasoning? is the right question. >For example, when being told > > Sam is pregnant. > >an "intelligent" system should infer some other facts. >For example, > > =g=> Sam is a female Illogical, Captain. Ectopic pregnancies can occur in males, and the deliberate induction of ectopic pregnancies in human males is now being considered as a way of sharing motherhood between the sexes. > =g=> Sam is alive Illogical, Captain. Pregnant females do die, and if close enough to term the live child can be safely removed from the dead pregnant mother. >Is this "logical" reasoning? Of course. Of course not. The unfortunate problem with logic is not that it is deficient as a system of reasoning, but that there are so few collections of english language propositions about the world to which it can be usefully applied. Usually it just gives silly results, as in the above example. If logical reasoning had survival value we -- not to mention dogs and cats -- would be a lot better at it. As it is, the kinds of informal reasoning we naturally employ are pretty well fitted to the kind of world we inhabit. And what kinds of reasoning do we employ? This is an interesting open research topic. -- Chris Malcolm cam@uk.ac.ed.aipna +44 (0)31 667 1011 x2550 Department of Artificial Intelligence, Edinburgh University 5 Forrest Hill, Edinburgh, EH1 2QL, UK DoD #205
wsaba@cscs.uwindsor.ca (Walid Saba) (06/18/91)
In article <1991Jun17.032758.14030@aifh.ed.ac.uk>, cam@aifh.ed.ac.uk (Chris Malcolm) writes: > > ... > The unfortunate problem with logic is not that it is deficient as a > system of reasoning, but that there are so few collections of english > language propositions about the world to which it can be usefully > applied. Usually it just gives silly results, as in the above example. The example I gave is not the test of all tests. Of course, given that the `alive' and `female' can not be assumed from `pregnant' (which I take it to be your main objection), we would modify our set of inference rules... What is the problem? My main point was that whatever set of beliefs we have, and however knowledge is representd we do (WE MUST) use some kind of reasoning when *processing* this information. Well, you seem to have no problem with that. Right? You do say: > If logical reasoning had survival value we -- not to mention dogs and > cats -- would be a lot better at it. As it is, the kinds of informal > reasoning we naturally employ are pretty well fitted to the kind of > world we inhabit. And what kinds of reasoning do we employ? This is an > interesting open research topic. So you suggest that it is some *informal* kind of reasoning. Again you missed my point (I think). By "formal" I am suggesting any logic in particular. I am also not suggesting that we do have that logic, or know anything about. It could turn out to be the LOGIC OF CHEO WAWA... All what I am saying is that it must be goverened by certain rules, and if it is the way we reason, we must be able to explain certain things using this logic/language... As for cats and dogs, well may be their logic is THE LOGIC OF WAWA CHEO. But since most of them can expected to behave in a certian way, they sure have some sort of WAWA CHEO reasoning system! WALID SABA wsaba@cs.uwindsor.ca School of Computer Science University of Windsor Windsor, Canada N9B 3P4 (519) 253-4232 ext. 3003/3005
berry@arcturus.uucp (Berry;Craig D.) (06/21/91)
wsaba@cscs.uwindsor.ca (Walid Saba) writes: >In article <1991Jun17.032758.14030@aifh.ed.ac.uk>, cam@aifh.ed.ac.uk (Chris Malcolm) writes: >> >> ... >> The unfortunate problem with logic is not that it is deficient as a >> system of reasoning, but that there are so few collections of english >> language propositions about the world to which it can be usefully >> applied. Usually it just gives silly results, as in the above example. >The example I gave is not the test of all tests. Of course, given that >the `alive' and `female' can not be assumed from `pregnant' (which I >take it to be your main objection), we would modify our set of inference >rules... What is the problem? >My main point was that whatever set of beliefs we have, and however >knowledge is representd we do (WE MUST) use some kind of reasoning >when *processing* this information. [...] >So you suggest that it is some *informal* kind of reasoning. Again you >missed my point (I think). By "formal" I am suggesting any logic in particular. >I am also not suggesting that we do have that logic, or know anything >about. It could turn out to be the LOGIC OF CHEO WAWA... All what I am >saying is that it must be goverened by certain rules, and if it is the way we >reason, we must be able to explain certain things using this logic/language... I have a fundamental objection to refering to our cognitive processes as a logical system of any type. A formal logical system must be reproducible - e.g., if I conclude that "pregnant => female" given my current internal state today, I should conclude the same thing given the same state at any other time. My assertion is that the underlying uncertainties of brain physiology negate this criterion. What I had for breakfast could skew synaptic responses just enough to move around a few truth values, without any new evidence or such being available.
costello@DEC-Lite.Stanford.EDU (Tom John Costello) (06/21/91)
In article <1991Jun20.201004.2277@arcturus.uucp>, berry@arcturus.uucp (Berry;Craig D.) writes: |>|> |> I have a fundamental objection to refering to our cognitive processes as a |> logical system of any type. A formal logical system must be reproducible - |> e.g., if I conclude that "pregnant => female" given my current internal |> state today, I should conclude the same thing given the same state at any |> other time. My assertion is that the underlying uncertainties of brain |> physiology negate this criterion. What I had for breakfast could skew |> synaptic responses just enough to move around a few truth values, without |> any new evidence or such being available. | I would certainly think that a few truth values moved around is easily sufficent to say that you are in a different state. The assumption that all input must be of a verbal form is wrong. Also formal logical systems do not have to be reproducible, non-deterministic logics exist, not that I find them in the least useful. The problem with arguing against logic is that for anyone to take you seriously you have to use logic to argue with. Arguments from physics claiming that matter is continuous or the brain is a wave function and can therefore not be described, are not an argument against logic merely they assert the need for non countable langauges in which to express the logic. In short, in order to study or discuss anything we must abstract it, and give it a name, and qualities. Thus the concept of concept assumes that we are formalising, as to extract a concept from our sensory perceptions is to map a certain formalism to those perceptions. Tom
wsaba@cscs.uwindsor.ca (Walid Saba) (06/21/91)
In article <1991Jun20.201004.2277@arcturus.uucp>, berry@arcturus.uucp (Berry;Craig D.) writes: > > [deleted stuff...] > > I have a fundamental objection to refering to our cognitive processes as a > logical system of any type. A formal logical system must be reproducible - > e.g., if I conclude that "pregnant => female" given my current internal > state today, I should conclude the same thing given the same state at any > other time. My assertion is that the underlying uncertainties of brain > physiology negate this criterion. What I had for breakfast could skew > synaptic responses just enough to move around a few truth values, without > any new evidence or such being available. Every one who replied to the original article (AGAINST FORMALISM?), except Smoliar, is interpreting "logical reasoning" out of context. I tried to say that by "logical reasoning" I only mean some type of LOGICal reasoning. That is we don't know what LOGIC it is, it could be a logic that has much more powerful tools than logical syatems have at the moment... For example, consider the following: - he failed the test (1)... Dr. Frost told his student that - he is taking a sabbatical next year (2)... - he should be at his office in the morning (3)... In (1) he is a reference to the student, in (2) to Dr. Frost, and in (3) it could be to either one... This of course is based on contextual world knowledge. Yep!!! No need to understand synaptic responses to solve these types of problems. Formal logic can accomodate this context-dependent interpretation. So, we perhaps extend logical systems soem more... May be context must be ivestigated furthur. Context is very essential. And Logics can deal with contexts, although no formal account of context is yet available for the general contextual analysis (I am still looking forward to hear or read about McCarthy's efforts in this subject. He did say that his main concern is to formalize the notion of context...). After all this introduction, you are right, "pregnant => female" could be accepted at some "internal state" (which I will simply call state), and the same implication might not hold in a different state! EUREKA *** What did we discover? This is an old and already solved problem in logical semantics. Interpretation could be done with respect to states, possible worlds, etc. For example, there could be mythical world where "pregnant" implies "male" (!). Intensional, and modal logics are powerful enough to accomodate for this situation. I do not know much about the brain physiology, but I think the problem in network, *PDP*, biological, ... , models is that they are missing the point. WE ARE NOT INTERESTED IN THE HARDWARE (Brain) rather WE ARE INTERESTED IN SOFTWARE (Mind, this big program that runs the every action of an individual). The only problem that I see in formal, logical semantics are that they are at the moment not flexible enough to cope with contextual world knowledge. There is however progress in this regard. And whatever the case may be at the moment, we should never forget that No progress in any sientific inquiry can be made unless it is based on solid foundation. Logical systems allow us to make a giant leap, they allow abstraction, generalization, and they describe the right way of doing things... I must say I like what Minsky states every now and than. It has in effect the following meaning (sorry if the qoute is not accurate enough): logic is perhaps useful is describing certain things, but not in explaining how and why things happen in a certain way. But if logical syatems can (from the expressive power point of view) describe a certain phenomena, they can be used to make giant steps (extrapolate, generalize) to explain general rules. They can be used then to "deduce" how larger things are operating, etc. In nay case, I am still hoping that the logic of CHEO WAWA will do the job !!! WALID SABA wsaba@cs.uwindsor.ca Comp. Sci., Windsor, Ontario, Canada.
dirish@csc-sun.math.utah.edu (Dudley Irish) (06/21/91)
The problem in this discussion is that we keep runing together the notions of reasoning rationally and formal logic. Remember the FORM in formal means that we operate only on the form of the expression, NOT on the meaning. It is not clear to me how a formal system can have any meaning. This is a very complex issue and it is by no means well understood. I am in no position to try to argue about languages of thought, rule based reasoning, or any of the other complex issues from philosophy of mind. I just want to remind people that the usefulness of formal logic is that it does not depend on the meaning of the terms. This in my opinion restricts its usefulness in explaining the meaning of thoughts. -- Dudley Irish / dirish@math.utah.edu / Manager Computer Operations Center for Scientific Computing, Dept of Mathematics, University of Utah The views expressed in this message do not reflect the views of the Dept of Mathematics, the University of Utah, or the State of Utah.
wsaba@cscs.uwindsor.ca (Walid Saba) (06/24/91)
In article <DIRISH.91Jun21074642@csc-sun.math.utah.edu>, dirish@csc-sun.math.utah.edu (Dudley Irish) writes: > The problem in this discussion is that we keep runing together the > notions of reasoning rationally and formal logic. Remember the FORM > in formal means that we operate only on the form of the expression, > NOT on the meaning. But meanings (according to some theories) have a well defined "form", which means that one could investigate a formal analysis of meanings. (the structure, or form, of meanings is, of course more complex, and to this point no single view is full proof... ). For example, Carnap-Cresswell-Montague like theories suggest that the srtructure of meanings is what is called *intensions*, while Kaplan have a slightly different view (information contents), Barwise suggested *situations*, etc... >It is not clear to me how a formal system can > have any meaning. In a way you are right. Everything is done by manipulating symbols, according to some well defined set of rules. Which means, we are only doing syntax. If you mean that we don't know how meanings are represented in our minds, than of course you are right. However, we do know that we can associate symbols, pictures, sounds, images, etc. with some "peice of information" that corresponds to reality... This could serve as meaning, what do you think??? Now FORM in FORMAL does correspond to the well-formedness aspect of a language, ie to form, (or structure), which is the syntax. However, languages have *semantics* and if these are not *formal* than that is where the people who are AGAINST FORMALISM do not see where the trouble lies.... Essentailly, one must justify his semantic theory, and unless it is formal (sound!!!!) not much could be done, if anything at all ****** > This is a very complex issue and it is by no means > well understood. I am in no position to try to argue about languages > of thought, rule based reasoning, or any of the other complex issues > from philosophy of mind. I just want to remind people that the > usefulness of formal logic is that it does not depend on the meaning > of the terms. This in my opinion restricts its usefulness in > explaining the meaning of thoughts. I understand your point about logic being essentially a formal language for manipulating symbols, independent of what these symbols mean. Essentially what you are saying is that EVERYTHING is syantx... This is an interesting argument, which I discussed with lots of people before. But recall, that inference rules, and even axioms in logic are usually justified semantically, ie they do correspond to some view... Of course, once these have been accepted, than all we need to do afterwards is syntactic manipulation, without any semantics. But recall, even a simple rule as (A or ~A), is considered to be true in all contexts, since we have a *semantic* justification for it. It is intuitively justified... etc. Is that true or not???? WALID SABA wsaba@cs.uwindsor.ca University of Windsor School of Computer Science Windsor, Ontario, Canada N9E 1X3 (519) 253-4232 Ext. 3003/3005