Paper by ~MM on distributed self-awareness

http://www.cs.pdx.edu/~mm/self-awareness.pdf
by Melanie Mitchell (Hofstadter was her Phd. supervisor)

"1. Global information is encoded as statistics and dynamics
of patterns over the system's components.

2. Randomness and probabilities are essential.

3. The system carries out a fine-grained, parallel search
of possibilities.

4. The system exhibits a continual interplay of bottomup
and top-down processes."

Implications for Artificial Intelligence ...

"The four principles listed above, along with other general
principles abstracted from the study of decentralized complex
adaptive systems, can be a guide in designing articial intelligence
systems with decentralized architectures that have sophisticated
abilities for pattern perception and self-awareness.

In fact, these principles guided the design of the Copycat
system, developed by Douglas Hofstadter and myself (Hofstadter
& Mitchell 1994; Mitchell 2001), and its successor, Metacat,
developed by Hofstadter and James Marshall (Marshall
& Hofstadter 1998; Marshall 2002).

Copycat and Metacat are programs that perceive patterns
and make analogies in the domain of letter strings,
such as "If abc changes to abd, what is the analogous
change to iijjkk?" The purpose of both these projects was
to develop general mechanisms of high-level perception,
analogy-making, and self-watching that would be extensible
to many domains. Metacat, in particular, is able to monitor
its own processing so as to improve its perceptual abilities,
and is considerably more sophisticated than Copycat.

Limitations on space prevent a detailed description of
Copycat and Metacat here; interested readers should consult
the references given above. However, the philosophy behind
the programs can be summarized by the following principles
for modeling perception, which closely follow the principles
abstracted above.
(1) The perceptual process must be fine-grained, diverse,
redundant, and decentralized.
(2) Perception is guided by "fluid" concepts, which are
themselves shaped as the perceptual process unfolds.
(3) The perceptual process proceeds as an interplay of
bottom-up modes (driven by stimuli from the environment)
and top-down modes (driven by expectation, prior knowledge,
biases, and what has already been discovered). This
interplay is not preprogrammed, but is an emergent effect of
the collective actions of low-level components of the system.
(4) The perceptual process shifts over time from being
highly parallel, random, and bottom-up, to being more focused,
deterministic, and top-down. As in (3), this shift is
not pre-programmed, but rather is an emergent effect of collective
behavior in the system.
(5) The perceptual process must have a means of "selfwatching"
-- monitoring its own state and progress -- that feeds back to
affect behavior. In Copycat, this is implemented by a
computational "temperature". This is similar to the ideas
proposed by Segel described above concerning self-watching in
the immune system, as well as ideas of Orosz concerning
"variable connectivity" (Orosz 2001). In Metacat, self-watching
is a more explicit and sophisticated part of the system.

I believe that the work on Copycat and Metacat is the first
time these principles have been stated together in this general
format, and actually tested in computer models.
While Copycat and Metacat are limited to perceiving
analogies in a simple domain, they are based on quite general
principles. I believe that similar principles, gleaned from
mechanisms that produce adaptive behavior and self-awareness
in natural systems, will be necessary components of future
artificial intelligence systems with sophisticated
metacognitive abilities."


.