Re: FW: [Fwd: FW: Interesting fact]

[Subbarao Kambhampati <rao>]

All that these examples illustrate is that signal interpretation (perception)
involves  prediction in large part. If you can predict/expect what is 
going to happen next, then you don't have to interpret the signal 
bottom-up. 

You will have top-down predctive models constraining the possible 
interpretations of the signal.

(the stuff about semantic nets etc is bogus... all intelligent artifacts,
whether natural or artificial will do it anyways. A robot considers a
particular sonar pattern to be a wall because it sort of expects a 
wall there; a robosoccer robot interprets an orange blob to be a 
soccer ball. when these work, we consider these intelligent. When these
don't (as in the robot following a orange-shirted guy), we laugh at it..
but the principle--of predicting what you are going to see/hear/read--
is an important means by which an agent with limited sensory and
computational capabilities can live efficiently in an environment..


Rao
-----------

Paris in
in Spring time