[PREV - WHAT_ABOUT_THE_WALL]    [TOP]

GRACE_MACHINES

                                             November 30, 2017

 "Machines of Loving Grace" (2013)
  by John Markoff                        The title is from a phrase
                                         by Richard Brautigan:
  This is a good (but not                "... and all watched over by
  great) book on some                    machines of loving grace."
  interesting subjects: Markoff
  spins the history of                      This works better for
  computers as a duel between               AI than IA.
  the visions of "artificial
  intelligence" and
  "intelligence augmentation",            The writing for this book had many
  exemplified by John McCarthy            low-level problems that seem peculiar
  and Douglas Englebart.                  for a pro like Markoff-- I think he's
                                          been relying (unconsciously?) on the
          ENGLEBART                       NYT copy editing staff.  Even worse    
                                          though is an inconsistency in tone and                              
          JMC                             perhaps a weakness in overall vision.          
                                          Is Markoff reporting on events,
  Though actually, Markoff                celebrating developments, or arguing
  quite aptly subdivides the              for an approach to the field...?
  "artificial intelligence"               
  side into two approaches:               There's too much Wired-magazine
                                          bizporn, and too much gosh-wow about
  McCarthy had always favored             things I was never much interested in
  an algorithmic approach                 and am already tired of hearing
  toward understanding the                about (robot cars!  Siri!).
  basis of intelligence.
  There's another approach
  toward AI that involves an
  adaptive learning              We have reliable handwriting and speech
  approach-- and that's          recognition now-- and that's something I
  actually the kind of AI we     was skeptical we'd ever see (back in the
  see working in the modern      80's we were still having trouble with
  world.                         optical character recognition of type-
                                 written text)-- but we don't actually know
  Markoff wants to tell a        how any of this works.
  story of augmented
  intelligence winning               These developments are both amazing and
  early on, and artifical            underwhelming, because there's no real
  intelligence winning               breakthrough in understanding that
  now, but then McCarthy             underlies them.  We owe our generation's
  doesn't work as an                 great triumphs of software to the
  icon:                              hardware engineers.

  McCarthy lost twice.               There's a Noam Chomsky interview
                                     available online, conducted at a
      Three times if you             Boston-area branch of Google: when
      count the micro/mainframe      asked about current trends in
      war-- McCarthy initially       linguistics, Chomsky commented that
      turned up his nose at          there was a peculiar phenomena now
      PCs, favoring the              where if some researchers got some
      multi-user shared              code working to predict a linguistic
      resource approach              pattern, they would act like they'd
                                     discovered something, but since they
      (Which he invented, by         don't really know how the code made
      the way-- circa 1980,          the prediction, the success can't act
      the young turks were           to confirm any underlying theory,
      having trouble                 because there is no underlying
      believing that JMC had         theory.
      come up with something
      as fundamental as the
      idea for time-sharing,
      but he really did                  There was an issue of the AAAS
      write the first paper              "Science" a few years back,
      about it.)                         which talked about a new field
                                         of research: trying to scrape
                                         some understanding out of a
                                         "neutral net" once you got it
                                         trained up and working...

                                         Perhaps notably, I haven't heard
                                         of any results from that approach
                                         since then.




--------
[NEXT - JMC]