Loading ...
Sorry, an error occurred while loading the content.

Re: [GP] eurisko and genetic programming

Expand Messages
  • jackpark@thinkalong.com
    It s worth putting all this in context. The scathing review was limited to AM, which provoked Lenat to think long and hard about the nature of his invention.
    Message 1 of 10 , Apr 13, 2004
    • 0 Attachment
      It's worth putting all this in context. The "scathing"
      review was limited to AM, which provoked Lenat to think
      long and hard about the nature of his invention. The result
      was Eurisko, following a year's postdoc with Herbert Simon,
      which, in the end, ran into problems greatly similar to
      those AM had, and thus was born Cyc.

      It would be true, in some sense, to argue that Lenat's
      voluminous pontifications about AM, Eurisko, and heuristics
      as a theory, don't offer up all that much with which to do
      science (in the sense of reproducing his results), though
      someone at CMU (Weimen Chen, if memory serves) managed to
      duplicate most of AM's discoveries by running a logic
      program of some sort. I never saw the program, but did read
      his paper. Ken Haase (google that) did a kind of rational
      reconstruction of AM, or was it Eurisko (I forgot which) as
      his MIT dissertation. And, Lenat put the full, undocumented
      source code for AM on the internet for a while. In theory,
      people have had a chance to duplicate his work. I don't
      recall her name, but another dissertation was a rational
      reconstruction of AM running in a graph-theoretic domain.

      I spent some ten years chasing the elusive AM/Eurisko
      paradigm, and finally managed a kind of detant with Lenat's
      writing and crafted a program which combines what I think
      to be the most important contributions of AM/Eurisko:
      computational "introspection" in an agenda based feedback
      system, heuristic driven. I call the program The Scholar's
      Companion (r), which combines evolutionary programming in
      the sense that AM/Eurisko do that, with Ken Forbus's QP
      Theory. That program was used in the research and defense
      of a dissertation in polymer science, and has been used in
      studies in hyperbaric immunology and molecular biology. Did
      I "crack" AM? I make no such claim. Is AM/Eurisko
      crackable? Probably not. Just read everything you can, and
      start over.

      Finally, I think the use of the word "scathing" is an
      unfair, ad hominem attack on an otherwise productive and
      creative fellow scientist.

      Jack
      On Tue, 13 Apr 2004 12:43:09 -0500
      "Gordon D. Pusch" <gdpusch@...> wrote:
      >John R. Koza writes:
      >
      >> Don't feel bad if you can "find very little about how it
      >was done"
      >> because I have never run into anyone who could figure
      >out how AM or
      >> Eurisko actually worked from the published papers on it
      >in the early
      >> 1980s. This whole issue is scathingly discussed in an
      >article that
      >> appears adjacent (in the same issue of the same journal)
      >to the more
      >> widely cited apologia concerning AM and Eurisko:
      >>
      >> Ritchie, G. D., and Hanna, F. K. 1984. AM: A case study
      >in AI methodology.
      >> Artificial Intelligence. 23(3): 249-268. August 1984.
      >
      >A slightly less "scathing" review appears to be located
      >at:
      >
      ><http://web.media.mit.edu/~haase/thesis/node52.html#SECTION00700000000000000000>.
      >
      >
      >-- Gordon D. Pusch
      >
    • John Koza
      ... From: Mike Eggleston [mailto:mikee@mikee.ath.cx] Sent: Tuesday, April 13, 2004 10:58 AM To: genetic_programming@yahoogroups.com Subject: Re: [GP] eurisko
      Message 2 of 10 , Apr 13, 2004
      • 0 Attachment
        -----Original Message-----
        From: Mike Eggleston [mailto:mikee@...]
        Sent: Tuesday, April 13, 2004 10:58 AM
        To: genetic_programming@yahoogroups.com
        Subject: Re: [GP] eurisko and genetic programming


        Hi John!

        On Tue, 13 Apr 2004, John Koza wrote:

        > Jack (and everyone):
        >
        > Don't feel bad if you can "find very little about how it was done" because
        I
        > have never run into anyone who could figure out how AM or Eurisko actually
        > worked from the published papers on it in the early 1980s. This whole
        issue
        > is scathingly discussed in an article that appears adjacent (in the same
        > issue of the same journal) to the more widely cited apologia concerning AM
        > and Eurisko:
        >
        > Ritchie, G. D., and Hanna, F. K. 1984. AM: A case study in AI methodology.
        > Artificial Intelligence. 23(3): 249-268. August 1984.

        Is that article(s) on-line somewhere?

        Mike




        Yahoo! Groups Links
      • John Koza
        I would guess that AI journal is on-line. It s an Elsevier journal. John Koza ... From: Mike Eggleston [mailto:mikee@mikee.ath.cx] Sent: Tuesday, April 13,
        Message 3 of 10 , Apr 13, 2004
        • 0 Attachment
          I would guess that AI journal is on-line. It's an Elsevier journal.

          John Koza

          -----Original Message-----
          From: Mike Eggleston [mailto:mikee@...]
          Sent: Tuesday, April 13, 2004 10:58 AM
          To: genetic_programming@yahoogroups.com
          Subject: Re: [GP] eurisko and genetic programming


          Hi John!

          On Tue, 13 Apr 2004, John Koza wrote:

          > Jack (and everyone):
          >
          > Don't feel bad if you can "find very little about how it was done" because
          I
          > have never run into anyone who could figure out how AM or Eurisko actually
          > worked from the published papers on it in the early 1980s. This whole
          issue
          > is scathingly discussed in an article that appears adjacent (in the same
          > issue of the same journal) to the more widely cited apologia concerning AM
          > and Eurisko:
          >
          > Ritchie, G. D., and Hanna, F. K. 1984. AM: A case study in AI methodology.
          > Artificial Intelligence. 23(3): 249-268. August 1984.

          Is that article(s) on-line somewhere?

          Mike




          Yahoo! Groups Links
        • David vun Kannon
          Indeed, the deepest criticisms of AM and Eurisko are that they didn t shift their heuristics themselves, only with the interference of the experimenter (Lenat)
          Message 4 of 10 , Apr 14, 2004
          • 0 Attachment
            Indeed, the deepest criticisms of AM and Eurisko are that they didn't shift their heuristics themselves, only with the interference of the experimenter (Lenat) did they do their best or most interesting work. This has now been reframed as an excellent example of human-computer collaboration! So Eurisko is really a shorthand for Eurisko-Lenat. Of the two main components, I suggest trying to replace the Lenat component, unless Doug has signed on to your experiments.

            If you had all of Doug Lenat's notebooks that recorded each of his decisions to interfere (and NOT interfere), and you could categorise how he interfered, then you could use that as a training set against which to test gp-built individuals. I can't guess how many trials it will take to evolve something as good as Doug was at tweaking Eurisko.

            The alternative is to accept Doug's role on yourself. In effect, he WAS the evaluation function. Crikey!

            Cheers,
            David vun Kannon
            ----- Original Message -----
            From: Jack Park
            To: genetic_programming@yahoogroups.com
            Sent: Monday, April 12, 2004 10:15 PM
            Subject: Re: [GP] eurisko and genetic programming


            Mike Eggleston wrote:

            >I find the idea of eurisko fascinating. I've researched several times,
            >but I find very little about how it was done. How was eurisko programmed
            >and how can genetic programming be used to enhance it's abilities?
            >
            >Mike
            >
            >
            I think there are some parallels between eurisko and gp, but maybe not
            that many. Both use evolutionary methods to build populations, but
            eurisko may only create one or a few new concepts at a time, and that,
            typically, based on some heuristic that is greatly different than the
            whims of a random number generator. In fact, eurisko mutated its
            heuristics along with concepts. GP seems to want to make largish
            populations of players which compete in some tournament; eurisko, otoh,
            uses other kinds of fitness landscapes, perhaps fitness based on the
            behavior of a larger population of concepts given the new entry. For
            instance, eurisko once managed to roam around in a space of VLSI
            designs, and stumbled on a different architecture, one that eliminated
            the need for a gate, forming an improved VLSI design. The final test was
            whether the design could be fabricated. In another instance, the
            tournament was the Traveller's game, in which Eurisko managed to exploit
            holes in the rules. There actually is plenty of literature on both AM
            and Eurisko, but not all that much on the web. Where GP crafts tree
            structures which provide solutions to problems, Eurisko may roam about
            in a hypergraph of concepts (which include heuristics) either in search
            against some fitness function, but also, occasionally, under guidance of
            some randomly-selected heuristic, looking to add more "fodder" for study
            and manipulation. I don't see where gp would alter eurisko's
            performance, but maybe I missed something. Hope that helps.

            Jack



            ------------------------------------------------------------------------------
            Yahoo! Groups Links

            a.. To visit your group on the web, go to:
            http://groups.yahoo.com/group/genetic_programming/

            b.. To unsubscribe from this group, send an email to:
            genetic_programming-unsubscribe@yahoogroups.com

            c.. Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service.



            [Non-text portions of this message have been removed]
          • David vun Kannon
            Jack, My recollection of the HR article is that it was about as critical as it is possible to be in properly refereed journal. I think it is acceptable to use
            Message 5 of 10 , Apr 16, 2004
            • 0 Attachment
              Jack,

              My recollection of the HR article is that it was about as critical as it is possible to be in properly refereed journal. I think it is acceptable to use "scathing", "denouncatory", "the-emperor-has-no-clothes" to describe that article in this list, even though those words are emotive. These words express an opinion about the HR article, not about AM, Eurisko, or Doug Lenat.

              Cheers,
              David vun Kannon
              ----- Original Message -----
              From: jackpark@...
              To: genetic_programming@yahoogroups.com
              Sent: Tuesday, April 13, 2004 2:14 PM
              Subject: Re: [GP] eurisko and genetic programming


              It's worth putting all this in context. The "scathing"
              review was limited to AM, which provoked Lenat to think
              long and hard about the nature of his invention. The result
              was Eurisko, following a year's postdoc with Herbert Simon,
              which, in the end, ran into problems greatly similar to
              those AM had, and thus was born Cyc.

              It would be true, in some sense, to argue that Lenat's
              voluminous pontifications about AM, Eurisko, and heuristics
              as a theory, don't offer up all that much with which to do
              science (in the sense of reproducing his results), though
              someone at CMU (Weimen Chen, if memory serves) managed to
              duplicate most of AM's discoveries by running a logic
              program of some sort. I never saw the program, but did read
              his paper. Ken Haase (google that) did a kind of rational
              reconstruction of AM, or was it Eurisko (I forgot which) as
              his MIT dissertation. And, Lenat put the full, undocumented
              source code for AM on the internet for a while. In theory,
              people have had a chance to duplicate his work. I don't
              recall her name, but another dissertation was a rational
              reconstruction of AM running in a graph-theoretic domain.

              I spent some ten years chasing the elusive AM/Eurisko
              paradigm, and finally managed a kind of detant with Lenat's
              writing and crafted a program which combines what I think
              to be the most important contributions of AM/Eurisko:
              computational "introspection" in an agenda based feedback
              system, heuristic driven. I call the program The Scholar's
              Companion (r), which combines evolutionary programming in
              the sense that AM/Eurisko do that, with Ken Forbus's QP
              Theory. That program was used in the research and defense
              of a dissertation in polymer science, and has been used in
              studies in hyperbaric immunology and molecular biology. Did
              I "crack" AM? I make no such claim. Is AM/Eurisko
              crackable? Probably not. Just read everything you can, and
              start over.

              Finally, I think the use of the word "scathing" is an
              unfair, ad hominem attack on an otherwise productive and
              creative fellow scientist.

              Jack
              On Tue, 13 Apr 2004 12:43:09 -0500
              "Gordon D. Pusch" <gdpusch@...> wrote:
              >John R. Koza writes:
              >
              >> Don't feel bad if you can "find very little about how it
              >was done"
              >> because I have never run into anyone who could figure
              >out how AM or
              >> Eurisko actually worked from the published papers on it
              >in the early
              >> 1980s. This whole issue is scathingly discussed in an
              >article that
              >> appears adjacent (in the same issue of the same journal)
              >to the more
              >> widely cited apologia concerning AM and Eurisko:
              >>
              >> Ritchie, G. D., and Hanna, F. K. 1984. AM: A case study
              >in AI methodology.
              >> Artificial Intelligence. 23(3): 249-268. August 1984.
              >
              >A slightly less "scathing" review appears to be located
              >at:
              >
              ><http://web.media.mit.edu/~haase/thesis/node52.html#SECTION00700000000000000000>.
              >
              >
              >-- Gordon D. Pusch
              >


              ------------------------------------------------------------------------------
              Yahoo! Groups Links

              a.. To visit your group on the web, go to:
              http://groups.yahoo.com/group/genetic_programming/

              b.. To unsubscribe from this group, send an email to:
              genetic_programming-unsubscribe@yahoogroups.com

              c.. Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service.



              [Non-text portions of this message have been removed]
            Your message has been successfully submitted and would be delivered to recipients shortly.