?

Log in

No account? Create an account

September 18th, 2007

Practical AI and Language (Again)

I was making a list of fictional AI machines (for my on-going interest in mind-mapping) and bumped into an awesome, vintage program: SHRDLU. The program's resurrected description verifies my insistence that language is inexorably related to the construction of reality.  We place immutable, metaphysical limitations on our world, as is evident by the way our words reflect our thoughts.  If SHRDLU could not comprehend something it was told -- like an extension to it's world-o-blocks --, then that world did not exist to the program, or at least it was inaccessible by any method available to the program.  Or in other words, "words matter".  This is a syntactic impediment, not a problem of meaning or creation.  SHRDLU's author, for example, evidently became disenchanted by problems of syntax, like parsing English [2], but it does not seem like there is any problem with creating more blocks or redefining the blocks.  Extending the world-o-blocks is a problem of developing something like a parser that could comprehend the extension directly and something like an interpreter to communicate its comprehension to us.  Ergo, intelligence is meaningful only when the thing can communicate its thinking to us in some predefined form, like Strunk's English.  It doesn't matter that person A believes in "asdfjkl;"; if they -- or it -- can't tell us that "asdfjkl" isn't anything more that a row on Querty's keyboard, then we'll say it's unintelligent. [1]

Now back to the list:

  1. Vicki from I Robot
  2. HAL from 2001: Space Odyssey
  3. LCARS from Star Trek: The Next Generation to Voyager
  4. Hybrids from Battlestar Galactica
  5. C-3P0 from Star Wars



1. I am not necessarily defending this conclusion, as is obvious by the entry on Hybrids.
2. Obviously; the inconsistency of natural languages is a problem that served as an impetus for my characterized "[Justin] has issues with words being able to mean different things in different contexts" posts.  The translation of this to standardized programming languages seems less of an issue than the logic required.