gilbert@cs.glasgow.ac.uk (Gilbert Cockton) (05/30/89)
If a system passes the Turing Test with one subject, but not with another, should it be considered intelligent? If 55% of a sample of hundreds say the system is intelligent, is it? What if the subjects are: a) drunk or on drugs (or a') the experimenters are :-))? b) mentally subnormal? c) polite and don't want to upset the experimenters (especially if a' applies too :-])? Then how valid is the Turing Test? Just what sort of Science did young Mr. Turing have in mind when he decided that subjective opinion could ever be a measure of system performance? How do AI types *REALLY* test their systems? -- Gilbert Cockton, Department of Computing Science, The University, Glasgow gilbert@uk.ac.glasgow.cs <europe>!ukc!glasgow!gilbert
cam@edai.ed.ac.uk (Chris Malcolm cam@uk.ac.ed.edai 031 667 1011 x2550) (05/31/89)
In article <3018@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) writes: >Just what sort of Science did young Mr. Turing have in mind when he >decided that subjective opinion could ever be a measure of system >performance? Young Mr Turing suggested in his 1950 paper "Computing Machinery and Intelligence" that it would be 50 years before what we now call "the Turing Test" and he called "the imitation game" could be applied. I see no reason to revise that estimate downwards. He also made it clear that he considered that everyday notions of "machinery" and "thinking" then carried too much excess semantic baggage for reasonable discussion of the question "can machines think" to be profitable, and consequently suggested the "imitation game" as a Gedanken experiment with which to clear the philosophical air. While there have been a few occasions in AI research when people were asked to play a Turing-like game to assess a program whose specific purpose was to articulate a model of a certain kind of of human behaviour, such as Colby's "PARRY", or the ranking of MYCIN's diagnostic responses compared to a panel of experts, it remains true that the primary purpose of the Turing Test in AI is as a gedanken experiment. As Turing pointed out in the paper in question: "The popular view that scientists proceed inexorably from well-established fact to well-established fact, never being influenced by any improved conjecture, is quite mistaken. Provided it is made clear which are proved facts and which are conjectures, no harm can result. Conjectures are of great importance since they suggest useful lines of research." The Turing Test was suggested by its originator as a source of useful conjecture. It still serves that purpose, and that tradition in AI is continued by Searle's Chinese Room argument, and Harnad's Total Turing Test, to name two examples recently ventilated on comp.ai. >How do AI types *REALLY* test their systems? This question seems to suggest that AI types _pretend_ to be using something like the Turing Test, but actually in the privacy of their labs are up to something quite different. I thought, Gilbert, that you had once in your career suffered some education in AI? The answer to how we AI types test our systems is as various as the kinds of system we build, and the reasons we build them. In many cases, as I am sure you know, what is interesting about our systems is why they don't work :-) That is not the kind of thing which is established by testing a system. -- Chris Malcolm cam@uk.ac.ed.edai 031 667 1011 x2550 Department of Artificial Intelligence, Edinburgh University 5 Forrest Hill, Edinburgh, EH1 2QL, UK
grano@kreeta.cs.Helsinki.FI (Juhani Grano) (06/01/89)
I think that Alain Turing did not want to define what 'intelligence' really is (that would indeed be hard), but rather to provide some kind of basis for empirical tests, and forget the unfruitful arguing about what defines intelligence. The test, although inaccurate, does provide some information about the intelligence of the object being tested. Furthermore, would you say that a drunk person is not intelligent? Or a child? Or someone mentally handicapped? The question of what defines intelligence is and will remain unsolved. -------------------- Kari Grano University of Helsinki, Finland email me at: grano@cs.helsinki.fi Department of CS.
bwk@mbunix.mitre.org (Barry W. Kort) (06/01/89)
In article <3018@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) wonders: > How do AI types *REALLY* test their systems? I suppose AI types test their systems much the same way teachers test their students. By giving them examinations consisting of puzzles or problems appropriate to the class of intelligence which the learning system has presumably acquired. Neural network classifiers are given samples to classify. Diagnostic expert systems are given the symptoms of ailments or fault conditions. Theorem provers are given candidate theorems to decide. The successful intelligent system (be it made of silicon or made of meat) gets a passing grade and goes ahead. The unsuccessful ones go back for remedial education or flunk out. (Or maybe they just end up as USENET junkies.) --Barry Kort
fransvo@maestro.htsa.aha.nl (Frans van Otten) (06/01/89)
Kari Grano writes: >I think that Alain Turing did not want to define what 'intelligence' >really is (that would indeed be hard), but rather to provide some >kind of basis for empirical tests, and forget the unfruitful arguing >about what defines intelligence. The test, although inaccurate, >does provide some information about the intelligence of the object >being tested. > >Furthermore, would you say that a drunk person is not intelligent ? >Or a child ? Or someone mentally handicapped ? The question of what >defines intelligence is and will remain unsolved. It seems to me that the use of the word "intelligence" is rather subjective, and the opposite of "dumbness". It is not an absolute property of a system or human or animal, at least not when normally used. I think that is why it is so hard to define "intelligence". And a definition based on subjective perceptions like these is probably not of much use in the field of ai. But it might be possible to describe a property which I might call "absolute intelligence". This could be described by something like this: 1. A set of rules, like "if big angry man coming towards me then run away"; 2. A machine to apply these rules to input data, resulting in output data and/or actions. Of course, ai languages like prolog use this kind of data representation, but they don't seem to be as successful as humans. On the other hand, we humans (at least I) do sometimes call a program intelligent (or dumb). This happens mostly when the used algorithm is a bit complex and/or the result is surprising. When I know exactly how the program works, I usually don't call it intelligent anymore. What does this mean ? I appreciate any comments on this random output of my brain. -- Frans van Otten | fransvo@maestro.htsa.aha.nl or Algemene Hogeschool Amsterdam | fransvo@htsa.uucp or Technische en Maritieme Faculteit | [[...!]backbone!]htsa!fransvo
b27y@vax5.CIT.CORNELL.EDU (06/02/89)
In article <3018@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) writes: >Then how valid is the Turing Test? > >Just what sort of Science did young Mr. Turing have in mind when he >decided that subjective opinion could ever be a measure of system >performance? > >How do AI types *REALLY* test their systems? The Turing Test is not really used to test AI systems at all. If The AI Guys today could get 55% of any (sober) group of people that thought the computer was intellegent, they would be dancing around the room. We are looking at a at least another few decades of research before we have anything "smart" enough to trick anyone. The Turing Test should be thought of as a Thought experiment: If you have some system that in every way seems and behaves in an intellegent manner, is it intellegent, or is it just some very complex series of brainless conditional responses? It brings up some fascinating philosophical questions. My favorate twist on the question: "How do you know if the person you are talking to is really intellegent?" The writer of this letter on the Usenet may SEEM like he/she/it has reasonable mental capabilities, (except for maybe spelling) but what if it is (I am) just some complex fancy generator. If your interested in a better discussion of the Turing Test, along with a very good counterexample (The Chinese Room example) I would read Robert Searle's paper called something like "Minds, Brains, and Program". ( I'm pretty sure that title is wrong). I think I saw some discusion on Searle lately on the net about him either in comp.ai or somewhere else. -----Misha------ Michael Gray /-------------------------------/ Misha Computing / "Save The Humans" / 526 Stewart Ave / / Ithaca N.Y. 14850 / Bumber Sticker / 607-277-2774 / / B27Y@VAX5.CIT.CORNELL.EDU /-------------------------------/
rwex@ida.org (Richard Wexelblat) (06/02/89)
Friends, go back and read Turing's paper again. It was a semi-technical (or perhaps non-technical) speculation on whether machines might ever be able to think -- and how we might be able to tell if they do so. Turing described something he called the Artificial Game in which a man, a woman, and an arbiter communicate by teletype. The arbiter cannot see either of the players, but they select between themselves which shall be required to tell the truth and which be permitted to lie. Then both try through conversation and Q and A to convince the arbiter that they are of the gender of the truth-teller. It might well be the case that over a broad selection of players and arbiters, there ought to be reliable statistics of the relative success of the truth-teller and liar. (At least within a group of similar age, education level, social class, nationality, etc.) Now, program a computer to "be intelligent" and give it "experience" sufficient to play the Artificial Game, replacing gender with human/ machine as the deciding factor. (I.e. sometimes the human will be the liar, sometimes the computer.) Turing POSITED that if the win/loss statistics for the human-computer game match those of the man-woman game then the computer might be said to "think." Bias has nothing to do with the test as it is statistical. Given a large enough sample of games played, individual bias can be made insignificant. Please note that Turing was not stating that a machine winning the game would be intelligent; rather he pointed to one whose win/loss statistics were commensurate to those of a human. My library is in transit so I can't check the original text of the paper. I believe, however that an equally valid interpretation of the "Turing Test" is to leave the game roles alone but just change the players. That is, now a man or woman plays one side, a computer the other. The goal, however, is still gender, not origin. I like this formulation better. Note also that the focus of the paper was methodology, not method. There's a big difference in that. -- --Dick Wexelblat (rwex@ida.org) 703 324 5511 (Formerly: rlw@philabs.philps.com)
childers@avsd.UUCP (Richard Childers) (06/02/89)
gilbert@cs.glasgow.ac.uk (Gilbert Cockton) babbles: >Just what sort of Science did young Mr. Turing have in mind when he >decided that subjective opinion could ever be a measure of system >performance? Gotta start somewhere. He wasn't measuring "system performance", he was measuring the relationship between human expectations and the reality, based upon an observed, but until then unspoken, set of rules already in place in human interactions. >How do AI types *REALLY* test their systems? Well, since you know so much, why don't you tell us a good metric for awareness, hot shit ? Seems to me the only metric for awareness is a boolean test for awareness, as measured by another point of awareness, with error checking carried out by consensus reality. I know some people will object to this as a way of generating quantifiable data, but I have never had any trouble integrating a series of 'yes'-'no' answers into a more detailed observation, and in fact it forms a major portion of my pool of problem-solving techniques in life. I see nothing shameful in such an effort. It may not meet _your_ criteria, but it meets everyone else's. Consensus reality is hard to argue with. But if you _like_ beating your head against a wall, well ... I don't think that's particularly suggestive of intelligence, however -- artificial or otherwise. There's nothing intelligent about unanswerable questions if they don't make any contribution to the discussion at hand. >Gilbert Cockton, Department of Computing Science, The University, Glasgow > gilbert@uk.ac.glasgow.cs <europe>!ukc!glasgow!gilbert -- richard -- * "We must hang together, gentlemen ... else, we shall most assuredly * * hang separately." Benjamin Franklin, 1776 * * * * ..{amdahl|decwrl|octopus|pyramid|ucbvax}!avsd.UUCP!childers@tycho *
gilbert@cs.glasgow.ac.uk (Gilbert Cockton) (06/02/89)
In article <1108@hydra.cs.Helsinki.FI> grano@cs.helsinki.fi writes: > > Furthermore, would you say that a drunk person is not > intelligent? Or a child? Or someone mentally handicapped? > The question of what defines intelligence is and will remain > unsolved. I think you may have misunderstood my point. The quality of a Turing test depends on the quality of the observing subjects. This is not true in the same way, or to the same extent, for proper experimental investigations. The bias here lies with the experimenter and the sample (both are revealed in replications, or results). The issue for the Turing Test is: what is an acceptable sample? If Turing didn't want to pin down intelligence, he should have used another word. I do not accept your version of history. Sources? In Turing's day, it was not as unreasonable to think of 'intelligence' as an out-there in-agents property. The term has no role today apart from common sense approbation. -- Gilbert Cockton, Department of Computing Science, The University, Glasgow gilbert@uk.ac.glasgow.cs <europe>!ukc!glasgow!gilbert
mbb@cbnewsh.ATT.COM (martin.b.brilliant) (06/03/89)
From article <952@maestro.htsa.aha.nl>, by fransvo@maestro.htsa.aha.nl (Frans van Otten): > ..... > It seems to me that the use of the word "intelligence" is rather > subjective, and the opposite of "dumbness". It is not an absolute > property of a system or human or animal, at least not when normally > used. I think that is why it is so hard to define "intelligence". > And a definition based on subjective perceptions like these is > probably not of much use in the field of ai. > > But it might be possible to describe a property which I might call > "absolute intelligence". This could be described by something like > ........ I wish I knew more about epistemology. I agree that the use of the word "intelligence" is subjective. On the one hand, definitions are subjective: what looks like a good definition to one person looks like a bad definition to another. On the other hand, the Turing test is objectively definable but is based on the subjective judgments of a participant. The subjectivity seems unavoidable. I see basically two approaches. One is to put the subjective part first, and then if we can only agree on it (ha, ha) then we can be objective thereafter. The "ha, ha" is, if I may say so, the joker. We argue and argue over definitions. The other approach is to put the objective part first. That's what Turing tried to do, in a sense. He suggested putting people and machines in an objective setting, and letting the people do what people do, namely, make subjective judgments. That way, we, the observers, at least can objectively watch other people being subjective, instead of being lost instantly in our own subjectivity. Turing's argument is based (I imagine) on the one thing (I imagine) we can agree on: that intelligence is what people have. All our definitions are based on the belief that people are intelligent. All our questions are based on wanting to know whether machines can do what people do. We just don't all know exactly what it is that people do. I think (ha, ha?) we can agree on that. 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.
mbb@cbnewsh.ATT.COM (martin.b.brilliant) (06/06/89)
From article <3039@crete.cs.glasgow.ac.uk>, by gilbert@cs.glasgow.ac.uk (Gilbert Cockton): > ..... > In Turing's day, it was not as unreasonable to think of 'intelligence' > as an out-there in-agents property. > > The term has no role today apart from common sense approbation. I don't want to sound stupid, but it seems to me that a statement like that takes all the meaning out of the term "artificial intelligence." What is artifical intelligence if intelligence is meaningless? If I take the statement literally, it means that if I like a machine, I can say it embodies artificial intelligence, and nobody (or anybody) can contradict me, because my statement is totally subjective. I really have trouble absorbing that. I sometimes find value in Gilbert Cockton's unconventional views, but at other times I find his excesses excessive. 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.
grano@kreeta.cs.Helsinki.FI (Juhani Grano) (06/07/89)
In article <952@maestro.htsa.aha.nl> fransvo@htsa.UUCP (Frans van Otten) writes:
:But it might be possible to describe a property which I might call
:"absolute intelligence". This could be described by something like
:this:
:
: 1. A set of rules, like "if big angry man coming towards me
: then run away";
: 2. A machine to apply these rules to input data, resulting
: in output data and/or actions.
:
:Of course, ai languages like prolog use this kind of data
:representation, but they don't seem to be as successful as humans.
The problem with these if .... then .... constructions is obviously that
they fail to achieve the necessary adaptability that humans have. When
something unexpected happens, a human *magically* knows the context or
'frame' which to adapt. This is of course related to experience and
knowledge about reality, but humans also seem to have the ability to
draw conclusions *very* heuristically...to err is human... It is not
clear to me whether the AI people are trying to make an intelligent machine or
a machine whose behaviour resembles that of human beings..:-)
The goal seems to run away every time we try to grasp it...I think it
only goes to show that trying to break 'intelligence' into discrete
areas is not very fruitful. Intelligence is more than the sum of the
features/abilities that are parts of it.
------------------------------
Kari Grano University of Helsinki, Finland
email to: grano@cs.helsinki.fi Department of CS
grano@kreeta.cs.Helsinki.FI (Juhani Grano) (06/07/89)
In article <3039@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) writes:
:The quality of a Turing test depends on the quality of the observing
:subjects. This is not true in the same way, or to the same extent,
:for proper experimental investigations. The bias here lies with the
:experimenter and the sample (both are revealed in replications, or
:results).
I agree.
:The issue for the Turing Test is: what is an acceptable sample?
I think anyone can see that e.g. statistically satisfactory sample of
testees yieds null result; that is, something so general that it says
nothing. Turing must have realized that! (no sources, sorry, I just
think so :-))
:If Turing didn't want to pin down intelligence, he should have used
:another word. I do not accept your version of history. Sources?
I haven't read the original paper, but according to a book of mine
(sorry again, it's finnish..) "Turing wrote, that the idea was to make
questions about the intelligence of machine uninteresting, NOT respond
to them." That also sounds reasonable to me - Turing wasn't an idiot.
:In Turing's day, it was not as unreasonable to think of 'intelligence'
:as an out-there in-agents property.
If I understood that...where is your sense of history? Take a look at
any book on the history of philosophy and see what's been said about
intelligence. Forty years is not that much...
------------------------------
Kari Grano University of Helsinki, Finland
email to: grano@cs.helsinki.fi Department of CS
gilbert@cs.glasgow.ac.uk (Gilbert Cockton) (06/07/89)
In article <1174@cbnewsh.ATT.COM> mbb@cbnewsh.ATT.COM (martin.b.brilliant) writes: >What is artifical intelligence if intelligence is meaningless? Exactly - can't we just get back together and have everyone just work on computer systems, and put away this silly distinction of AI versus non-AI systems? Where does it get us? >I really have trouble absorbing that. I sometimes find value in >Gilbert Cockton's unconventional views, but at other times I find his >excesses excessive. It is the idea that intelligence is a definable, measurable property which is a perversion. I am unconventional here, but not in much larger academic subcultures than than miniscule AI community. I suggest you look at the intelligence debate in psychometrics, and Herb Simon's "Sciences of the Artificial" - as someone in touch with psychologists, he has better sense than to want to use such a term as AI. If you are *SERIOUSLY* interested in this question, there is an enormous amount of work in psychometrics and educational psychology on this. Remember, IQ tests were originally devised to keep idiots out of the French infantry. Today they only confirm that armies have to take them anyway:-) -- Gilbert Cockton, Department of Computing Science, The University, Glasgow gilbert@uk.ac.glasgow.cs <europe>!ukc!glasgow!gilbert
cam@edai.ed.ac.uk (Chris Malcolm cam@uk.ac.ed.edai 031 667 1011 x2550) (06/07/89)
In article <3039@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) writes: >In Turing's day, it was not as unreasonable to think of 'intelligence' >as an out-there in-agents property. > >The term has no role today apart from common sense approbation. > Would you say the same sort of thing about 'consciousness'? Knowledge? Belief? -- Chris Malcolm cam@uk.ac.ed.edai 031 667 1011 x2550 Department of Artificial Intelligence, Edinburgh University 5 Forrest Hill, Edinburgh, EH1 2QL, UK
cam@edai.ed.ac.uk (Chris Malcolm cam@uk.ac.ed.edai 031 667 1011 x2550) (06/07/89)
In article <3039@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) writes: >The quality of a Turing test depends on the quality of the observing >subjects. This is not true in the same way, or to the same extent, >for proper experimental investigations. This is an excellent point. Why, only last week, when one of my research assistants came up to me saying "I've been playing with the latest version of the program for a week now, and I still can't tell the difference between it and Harry!" I had to explain to her - yet again! - the proper design of psychological experiments. Come on! What are we talking about? Turing suggested a gedanken experiment he doesn't think would in practice be possible for 50 years, there's no good reason to contract that estimate, and Gilbert is criticising the design of the experiment as though it were commonplace practice?? -- Chris Malcolm cam@uk.ac.ed.edai 031 667 1011 x2550 Department of Artificial Intelligence, Edinburgh University 5 Forrest Hill, Edinburgh, EH1 2QL, UK
gilbert@cs.glasgow.ac.uk (Gilbert Cockton) (06/08/89)
In article <407@edai.ed.ac.uk> cam@edai (Chris Malcolm) writes: >In article I write >> >>The term has no role today apart from common sense approbation. > >Would you say the same sort of thing about 'consciousness'? Knowledge? >Belief? Of course not. These are still very productive terms (e.g. race relations, Drucker's theories on knowledge workers, the Rushdie debate over respect for beliefs). Intelligence is dead outside of AI. Look at the psychometrics literature of the 1970s. Is anyone in psychology still studying intelligence and its measurement? -- Gilbert Cockton, Department of Computing Science, The University, Glasgow gilbert@uk.ac.glasgow.cs <europe>!ukc!glasgow!gilbert
gilbert@cs.glasgow.ac.uk (Gilbert Cockton) (06/08/89)
In article <408@edai.ed.ac.uk> cam@edai (Chris Malcolm) writes: >Come on! What are we talking about? Turing suggested a gedanken >experiment he doesn't think would in practice be possible for 50 years, So it isn't a test then? >there's no good reason to contract that estimate, and Gilbert is >criticising the design of the experiment as though it were commonplace >practice?? So what is the common practice? Again, how *DO* AI types test their systems? -- Gilbert Cockton, Department of Computing Science, The University, Glasgow gilbert@uk.ac.glasgow.cs <europe>!ukc!glasgow!gilbert
Gordon@ucl-cs.UUCP (06/08/89)
In article <3039@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) writes: >The quality of a Turing test depends on the quality of the observing >subjects. This is not true in the same way, or to the same extent, >for proper experimental investigations. [followed by comment from Chris Malcolm.] My two cents worth is ``Inteligence is the the mind of the beholder.'' Gordon.
cww@cadre.dsl.PITTSBURGH.EDU (Charles William Webster) (06/09/89)
In article <3075@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) writes: >It is the idea that intelligence is a definable, measurable property >which is a perversion. > >I am unconventional here, but not in much larger academic subcultures than >than miniscule AI community. I suggest you look at the intelligence >debate in psychometrics, and Herb Simon's "Sciences of the Artificial" >- as someone in touch with psychologists, he has better sense than to >want to use such a term as AI. My, my, my. Aren't we superior! You'll fool more of the people more of the time if you actually read the sources you superciliously cite (or at least represent them undistortedly). In "Sciences of the Artificial" Simon says: "At any rate, "artificial intelligence" seems here to stay, and it may prove easier to cleanse the phrase than to dispense with it. In time it will become sufficiently idiomatic that it will no longer be the target of cheap rhetoric." Simon may not pepper his writing with the phrase "artificial intelligence" but many of his students and collaborators are artificial intelligence reseachers, and he isn't nearly as mealy-mouthed about them as you are. > >-- >Gilbert Cockton, Department of Computing Science, The University, Glasgow > gilbert@uk.ac.glasgow.cs <europe>!ukc!glasgow!gilbert Chuck
cam@edai.ed.ac.uk (Chris Malcolm cam@uk.ac.ed.edai 031 667 1011 x2550) (06/09/89)
In article <1119@hydra.cs.Helsinki.FI> grano@cs.helsinki.fi writes: > It is not >clear to me whether the AI people are trying to make an intelligent machine or >a machine whose behaviour resembles that of human beings..:-) > Both extremes are research goals being pursued in AI. There are also some humble(?) AI people who are trying to make machines whose behaviour resembles that of insects. -- Chris Malcolm cam@uk.ac.ed.edai 031 667 1011 x2550 Department of Artificial Intelligence, Edinburgh University 5 Forrest Hill, Edinburgh, EH1 2QL, UK
andrew@berlioz (Lord Snooty @ The Giant Poisoned Electric Head ) (06/10/89)
In article <281@ucl-cs.UUCP>, Gordon@ucl-cs.UUCP writes: > My two cents worth is > ``Inteligence is the the mind of the beholder.'' See? Machines can't make errors of this type [sic] You are most definitely human! -- ................................................................... Andrew Palfreyman I should have been a pair of ragged claws nsc!logic!andrew Scuttling across the floors of silent seas ...................................................................
geb@cadre.dsl.PITTSBURGH.EDU (Gordon E. Banks) (06/10/89)
In article <3079@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) writes: >Again, how *DO* AI types test their systems? I suppose it depends on the system. I never heard of anyone using the Turing test, except perhaps with ELIZA in a modified way. INTERNIST was tested by giving it published tough cases that were also given to human experts. The program's results were compared to the those of the humans. Also, it was tested one-on-one against chairmen of departments of medicine when Dr. Myers was visiting their institutions for grand rounds.
cam@edai.ed.ac.uk (Chris Malcolm cam@uk.ac.ed.edai 031 667 1011 x2550) (06/13/89)
In article <3079@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) writes: >So what is the common practice? >Again, how *DO* AI types test their systems? As I've said before, in ways as various as the intended capabilities of the systems. For example, (you're not going to like this!) I'm developing a system which can plan how to assemble shapes out of parts. How do I test it? I tell it the shapes of the parts, the shape to build, and then watch the robot build the shape (or fail, as the case may be). The criterion is simple and indubitable. I cannot imagine there ever being any dispute about whether or not the robot succeeded (except trivial borderline pedantries). By developing I mean that I'm trying to extend the capabilities of the system. It is not a complicated system; there are probably thousands of ways in which it could be built. What is interesting is that some ways are very simple, whereas others are very complex. What is even more interesting is why, i.e., the interesting research questions concern good (simple, economical) architectures for building systems capable of successful thought and action in a world. -- Chris Malcolm cam@uk.ac.ed.edai 031 667 1011 x2550 Department of Artificial Intelligence, Edinburgh University 5 Forrest Hill, Edinburgh, EH1 2QL, UK
bwk@mbunix.mitre.org (Barry W. Kort) (06/14/89)
In article <3075@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) writes: > It is the idea that intelligence is a definable, measurable property > which is a perversion. Perhaps I am a bit perverted, but, when *I* use the word, "intelligence", I mean "the ability to think and solve problems". I define a "problem" as "an undesired state of affairs for which an appropriate idea has not yet been generated or agreed upon." I define "idea" as "a possibility for changing the state of affairs." I define "thinking" as "a rational form of information processing which reduces the entropy or uncertainty of a knowledge base, generates solutions to outstanding problems, and conceives goal-oriented courses of action." --Barry Kort
gilbert@cs.glasgow.ac.uk (Gilbert Cockton) (06/19/89)
In article <56041@linus.UUCP> bwk@mbunix (Barry Kort) writes: >Perhaps I am a bit perverted, but, when *I* use the word, "intelligence", >I mean "the ability to think and solve problems". Wow, that's tight! What if I can only solve some of your problems? What if I'm brilliant at some, and moderte at others? We can talk of intelligent behaviour (nearly always post-hoc), but never general intelligence - this was backed up by the psychometric work too, although that 'g' factor didn't always factor out. 'g' could be perceptual speed, confidence, insight, what have you, but it in no way guarantees success at an arbitrary problem. I'd love to staff a MacDonald's for a day completely with MENSA types to see how their IQ scores prepared them for all the problems of fast-food service :-) >I define a "problem" as "an undesired state of affairs for which >an appropriate idea has not yet been generated or agreed upon." Subjective, moral? Will AI solve all the world's problems? >I define "idea" as "a possibility for changing the state of affairs." Thus many ideas aren't. >I define "thinking" as "a rational form of information processing >which reduces the entropy or uncertainty of a knowledge base, generates >solutions to outstanding problems, and conceives goal-oriented courses >of action." And thus much brain life isn't thinking. What else do you want to proscribe? -- Gilbert Cockton, Department of Computing Science, The University, Glasgow gilbert@uk.ac.glasgow.cs <europe>!ukc!glasgow!gilbert
smoliar@vaxa.isi.edu (Stephen Smoliar) (06/19/89)
In article <56041@linus.UUCP> bwk@mbunix (Barry Kort) writes: >In article <3075@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk >(Gilbert Cockton) writes: > > > It is the idea that intelligence is a definable, measurable property > > which is a perversion. > >Perhaps I am a bit perverted, but, when *I* use the word, "intelligence", >I mean "the ability to think and solve problems". > >I define a "problem" as "an undesired state of affairs for which >an appropriate idea has not yet been generated or agreed upon." > >I define "idea" as "a possibility for changing the state of affairs." > >I define "thinking" as "a rational form of information processing >which reduces the entropy or uncertainty of a knowledge base, generates >solutions to outstanding problems, and conceives goal-oriented courses >of action." > >--Barry Kort Let us leave aside issues of perversion (and perversity) and go back to Gilbert's original remark. Note that, whether or not we accept "I define" as constituting a valid definition, Barry as left untouched the word "measurable," which I, for one, find to be a critical part of Gilbert's observation. Barry, do you wish to comment on how you would work measurement into your criteria? Actually, Gilbert's remark is very much in sympathy with what Minsky says about intelligence in THE SOCIETY OF MIND: A term frequently used to express the myth that some single entity or element is responsible for the quality of a person's ability to reason. (Besides, I prefer "myth" to "perversion.") After all, if you can isolate it as a single entity, then you have a leg up on being able to define or measure it. (You still may not succeed, of course. We still don't do a very good job when it comes to defining "chair.") ========================================================================= USPS: Stephen Smoliar USC Information Sciences Institute 4676 Admiralty Way Suite 1001 Marina del Rey, California 90292-6695 Internet: smoliar@vaxa.isi.edu "For every human problem, there is a neat, plain solution--and it is always wrong."--H. L. Mencken
jwi@lzfme.att.com (Jim Winer @ AT&T, Middletown, NJ) (06/28/89)
Barry Kort writes: > >Perhaps I am a bit perverted, but, when *I* use the word, "intelligence", > >I mean "the ability to think and solve problems". Gilbert Cockton comments: > Wow, that's tight! What if I can only solve some of your problems? > What if I'm brilliant at some, and moderte at others? ... > I'd love to staff a MacDonald's for a day completely with MENSA types > to see how their IQ scores prepared them for all the problems of > fast-food service :-) Jim Winer adds: It's *not fair* to *actually listen* to what somebody is saying and then comment on it. It's also *not nice* to pick on MENSA types and others with congenital defects. . But it's fun. Jim Winer ..!lzfme!jwi Those persons who advocate censorship offend my religion. Pax Probiscus! Unable to reply to email sucessfully. The opinions expressed here are not necessarily
bwk@mbunix.mitre.org (Barry W. Kort) (06/29/89)
In article <8683@venera.isi.edu> smoliar@vaxa.isi.edu.UUCP (Stephen Smoliar) writes: > In article <56041@linus.UUCP> bwk@mbunix (Barry Kort) writes: > > In article <3075@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk > > (Gilbert Cockton) writes: > > > It is the idea that intelligence is a definable, measurable > > > property which is a perversion. > > Perhaps I am a bit perverted, but, when *I* use the word, > > "intelligence",I mean "the ability to think and solve problems". > > I define a "problem" as "an undesired state of affairs for which > > an appropriate idea has not yet been generated or agreed upon." > > I define "idea" as "a possibility for changing the state of affairs." > > I define "thinking" as "a rational form of information processing > > which reduces the entropy or uncertainty of a knowledge base, generates > > solutions to outstanding problems, and conceives goal-oriented courses > > of action." > Let us leave aside issues of perversion (and perversity) and go back to > Gilbert's original remark. Note that, whether or not we accept "I define" > as constituting a valid definition, Barry has left untouched the word > "measurable," which I, for one, find to be a critical part of Gilbert's > observation. Barry, do you wish to comment on how you would work > measurement into your criteria? Fair question. And one to which I have not given much thought. Intelligence appears to be a multi-dimensional attribute. Some psychologists have identified as many as seven distinct kinds of intelligence. So if intelligence is measurable, the measure would probably have to be given as a vector. (We already know that conventional tests of scholastic achievement distinguish verbal intelligence from mathematical intelligence.) But if there is a measure of intelligence, it would have to be based on ability to consistently solve problems of varying categories and levels of difficulty and complexity. In this sense, intelligence is really measured in terms of achievement, relative to the population. Some kinds of intelligence, such as artistic creativity, or social skills are difficult to measure with any degree of precision. > Actually, Gilbert's remark is very much in sympathy with what Minsky > says about intelligence in THE SOCIETY OF MIND: > A term frequently used to express the myth that some single > entity or element is responsible for the quality of a person's > ability to reason. I agree that intellience is not a single entity, but as Minsky suggests, an agglomeration of interworking subsystems. > (Besides, I prefer "myth" to "perversion.") After all, if you can > isolate it as a single entity, then you have a leg up on being able > to define or measure it. (You still may not succeed, of course. > We still don't do a very good job when it comes to defining "chair.") Like "Chair", Intelligence is the name of a fuzzy set. Candidate systems have varying degrees of membership in the category of Intelligent Systems. > "For every human problem, there is a neat, plain solution--and it > is always wrong."--H. L. Mencken Wasn't it Lao Tse (or maybe Chang Tse) who said, "Think about right and wrong, and one immediately falls into error."? --Barry Kort
bwk@mbunix.mitre.org (Barry W. Kort) (06/29/89)
In article <3118@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) writes: > In article <56041@linus.UUCP> bwk@mbunix (Barry Kort) writes: > > Perhaps I am a bit perverted, but, when *I* use the word, > > "intelligence", I mean "the ability to think and solve problems". > Wow, that's tight! What if I can only solve some of your problems? > What if I'm brilliant at some, and moderate at others? Then you have a mixed score on the intelligence vector. Incidently, Intelligence includes the ability to learn, discover, and create. One can learn a specific method appropriate to an unfamiliar class of problems, or one can devise a novel method to solve a new class of problems. > We can talk of intelligent behaviour (nearly always post-hoc), but > never general intelligence - this was backed up by the psychometric > work too, although that 'G' factor didn't always factor out. 'G' could > be perceptual speed, confidence, insight, what have you, but it in no > way guarantees success at an arbitrary problem. The best method I know of for solving the arbitrary problem is the Socratic Method. At the very least, it leads one to boundary between one's knowledge and ignorance. > I'd love to staff a MacDonald's for a day completely with MENSA types > to see how their IQ scores prepared them for all the problems of > fast-food service :-) I wonder if they would do better than Hamburger Helpers staffing our universities and think tanks. > > I define a "problem" as "an undesired state of affairs for which > > an appropriate idea has not yet been generated or agreed upon." > Subjective, moral? Will AI solve all the world's problems? I imagine there will be some teamwork between silicon-based systems and carbon-based systems. > > I define "idea" as "a possibility for changing the state of affairs." > Thus many ideas aren't. In the Calculus of Ideas, the goal is to generate and select the ideas which, when applied to the Real World, successfully transform us from the Present State to the Goal State. If our World Models are sufficiently accurate, we stand a chance of achieving this level of competence. > > I define "thinking" as "a rational form of information processing > > which reduces the entropy or uncertainty of a knowledge base, generates > > solutions to outstanding problems, and conceives goal-oriented courses > > of action." > And thus much brain life isn't thinking. Correct. By antonymy, I define "worrying" as an emotional form of information processing which fails to reduce the entropy or uncertainty of a knowledge base, fails to generate solutions to outstanding problems, or fails to conceive goal-oriented courses of action. > What else do you want to proscribe? Domestic violence, state terrorism, and child abuse. --Barry Kort
dmocsny@uceng.UC.EDU (daniel mocsny) (07/07/89)
In article <58052@linus.UUCP>, bwk@mbunix.mitre.org (Barry W. Kort) writes: > In article <3118@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk > (Gilbert Cockton) writes: > > Wow, that's tight! What if I can only solve some of your problems? > > What if I'm brilliant at some, and moderate at others? > > Then you have a mixed score on the intelligence vector. I almost wish we did not have one word "intelligence." Just as the Greeks had three words for what we call "love," one word can't contain everything that jams into the concept of what minds do. Consider the far simpler problem of characterizing the performance of a computer system. If I take two comparable workstations, one built around an Intel 80386 and the other containing a Moto 68030, and I ask the simple question: "Which system is more powerful?" The answer is another question: "At what task?" Until we know how to unambiguously characterize our artifacts, we can hardly get a handle on ourselves. To accurately benchmark a computer system, you need some quantitative expression that contains terms accounting for the performance of every subsystem constituting the system. Then you need to be able to express exactly how a particular task makes demands on those subsystems. I don't see any obvious way to make this procedure any simpler than just running the task on the target system and watching the wall clock. > > I'd love to staff a MacDonald's for a day completely with MENSA types > > to see how their IQ scores prepared them for all the problems of > > fast-food service :-) > > I wonder if they would do better than Hamburger Helpers staffing > our universities and think tanks. Whenever we have a group of people involved in some competitive environment with some fairly solid performance metric (e.g., getting through engineering school, learning to fly combat aircraft, playing a musical instrument), we find that their scores usually describe something like a normal distribution. People responsible for managing large enterprises obviously want to have some way to predict individual performance in whatever skills they demand. How convenient if this predictor were to be a scalar as easy to discuss as "intelligence." However, actual performance is the only valid test, as no artificial testing procedure can accurately account for all the factors. We don't yet have much of an idea how given real-world problems make demands on our wetware. If we did, we might be able to isolate our brains' subsystems, attempt to characterize their performances, and then try to draw some conclusions about how well the parts might work together. But this seems absurdly beyond what we can meaningfully discuss just now. Dan Mocsny dmocsny@uceng.uc.edu
mbb@cbnewsh.ATT.COM (martin.b.brilliant) (07/09/89)
From article <1415@uceng.UC.EDU>, by dmocsny@uceng.UC.EDU (daniel mocsny): > In article <58052@linus.UUCP>, bwk@mbunix.mitre.org (Barry W. Kort) writes: >> In article <3118@crete.cs.glasgow.ac.uk> gilbert@cs.glasgow.ac.uk >> (Gilbert Cockton) writes: > .... >> > I'd love to staff a MacDonald's for a day completely with MENSA types >> > to see how their IQ scores prepared them for all the problems of >> > fast-food service :-) >> >> I wonder if they would do better than Hamburger Helpers staffing >> our universities and think tanks. I know I'm not answering Daniel Mocsny, who (I think) finessed the MacDonald's question. I think Gilbert Cockton asked it. But the question doesn't have to be finessed. IQ runs in my family. My children did well in the best schools and not so well in others. The two who best fit that mold also do well in menial jobs. They don't do MacDonald's, but they do very well waiting on tables in restaurants, after hours at college. One earned money as a farmhand when she was too young to get any other job. Of course they needed training. Staffing "for a day" doesn't make sense. But they trained better than their friends, and stayed on the job longer. I don't know about Mensa types - I quit Mensa because I didn't need what that organization offered - but IQ does not by itself disqualify you from hard work. It just qualifies you to get a better job. Unfortunately, it's a necessary qualification, not sufficient. People look at machines, etc. and say: "that can't be intelligent." So machines, etc. don't get a chance to show what they can do. 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.
jps@cat.cmu.edu (James Salsman) (07/10/89)
In article <2037@cbnewsh.ATT.COM> mbb@cbnewsh.ATT.COM (martin.b.brilliant) writes: > IQ runs in my family. Please do not interpolate that idea: if you do, then you will be running the risk of ENTRY racism (RAY'siz'uhm) n. SYLLABLES ra-cism MEANING 1. The notion that one's own ethnic stock is superior. 2. Discrimination or prejudice based on racism. I enjoy working with large collections of on-line texts. :James ::chgrp -- :James P. Salsman (jps@CAT.CMU.EDU)
dhw@itivax.iti.org (David H. West) (07/10/89)
In article <5453@pt.cs.cmu.edu> jps@cat.cmu.edu (James Salsman) writes: ]In article <2037@cbnewsh.ATT.COM> mbb@cbnewsh.ATT.COM (martin.b.brilliant) writes: ] ]> IQ runs in my family. ] ]Please do not interpolate that idea: if you do, then ]you will be running the risk of ] ] ENTRY racism (RAY'siz'uhm) n. Marty's statement is potentially refutable, and hence capable of refinement into a scientific hypothesis. It is thoroughly reprehensible to suggest that such a thing should be rejected by purely political criteria. We are descending into barbarity quite fast enough already, thank you.
mbb@cbnewsh.ATT.COM (martin.b.brilliant) (07/11/89)
From article <5453@pt.cs.cmu.edu>, by jps@cat.cmu.edu (James Salsman): > In article <2037@cbnewsh.ATT.COM> mbb@cbnewsh.ATT.COM (martin.b.brilliant) writes: > >> IQ runs in my family. > > Please do not interpolate that idea: if you do, then > you will be running the risk of > > .... racism ... 1. The notion that one's own ethnic stock is superior. 2. > Discrimination or prejudice based on racism. > > I enjoy working with large collections of on-line texts. > > :James > ::chgrp > -- > > :James P. Salsman (jps@CAT.CMU.EDU) I don't know what to make of that. I think it was sent in anger, because it doesn't make sense. And it looks like an attempt at a public insult. I hope it is not. Public insult is demeaning to a professional newsgroup. I didn't ask for special treatment, nor ask anyone to deny fairness to anyone else. I said family, not race. I made no reference to ethnic stock. I expressed no prejudice. I expressed an observation. I admire my father's intelligence, which was not fully reflected in the job he held because religious prejudice kept him out of a better one. I admire my late mother's intelligence. I enjoy the company of intelligent, fair-minded people. Please, somebody help me. What should I have said instead of what I did say? Or is something bugging Mr. Salsman? Is there perhaps something wrong with the notion that intelligence is a heritable trait? Or a skill teachable by parents to children? 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.
jps@cat.cmu.edu (James Salsman) (07/12/89)
In article <2061@cbnewsh.ATT.COM> mbb@cbnewsh.ATT.COM (martin.b.brilliant) writes: > From article <5453@pt.cs.cmu.edu>, by jps@cat.cmu.edu (James Salsman): > > In article <2037@cbnewsh.ATT.COM> mbb@cbnewsh.ATT.COM (martin.b.brilliant) writes: > > > > > IQ runs in my family. > > > > Please do not interpolate that idea: if you do, then > > you will be running the risk of > > > > racism ... 1. The notion that one's own ethnic stock is superior. > > I don't know what to make of that. I think it was sent in anger, > because it doesn't make sense. And it looks like an attempt at a > public insult. I hope it is not. Goodness, I was certainly not trying to be offensive in any way, but the logical extention of one's family is one's race, and if any person makes that *interpolation* of the concept, they are racist. You are not, and I am sorry if I offended you. I apologize for not making clear that I was trying to provide and example of a value-system conflict that I was reffering to in my "Value Systems for AI" post a few days back. I don't argue that IQ runs in families, but it is also heavily dependant on environmental factors, and everyone should be given an equal oppertunity for eduction, regardless of their family or race. :jps -- :James P. Salsman (jps@CAT.CMU.EDU) -- :James P. Salsman (jps@CAT.CMU.EDU)
Gordon@ucl-cs.UUCP (07/22/89)
>> ``Inteligence is the the mind of the beholder.''
What I meant to type was ``Intelligence is in the mind of the beholder.''
Presumably I was understood the first time.... any machines reading this?
Gordon.
Gordon@ucl-cs.UUCP (07/22/89)
> Both extremes are research goals being pursued in AI. There are also some > humble(?) AI people who are trying to make machines whose behaviour > resembles that of insects. Agreed. A few years ago AI people had done rocks and were working our way up to the bacteria. Gordon. "Robust code not insight into intellignce" - Mike Lesk.
krobt@nova.UUCP (Robert Klotz) (08/03/89)
In an article of <22 Jul 89 11:49:11 GMT>, Gordon@ucl-cs.UUCP writes:
"
" .... any machines reading this?
"
yes, many, however i am sure not many are understanding.
...robert--
------
{att!occrsh|dasys1|killer|uokmax}!mom!krobt | argue for your limitations
or -------------- | and soon you will find that
krobt%mom@uokmax.ecn.uoknor.edu | they are yours.