COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	Comments on "The Role of RAW POWER in INTELLIGENCE" by Hans Moravec
C00006 ENDMK
C⊗;
Comments on "The Role of RAW POWER in INTELLIGENCE" by Hans Moravec

#. In general, I doubt the conclusions of this paper.  It seems to
me that there is no guarantee that the current ideas in artificial
intelligence are adequate to produce human-level intelligence or
human-level vision with even a billion times the present processing
power, i.e. a factor of a thousand beyond what Moravec considers
necessary.  Moreover, I think that present computers well enough
programmed can achieve human-level intelligence and perhaps human-level
vision.

#. On the other hand, the shortest path to artificial intelligence
may well include the million-fold increase in processing power
Moravec proposes.

#. The argument on p.3 for neural nets on the grounds that animals
have them is fallacious.  Animal intellectual mechanisms have evolved
and not every possibility of design is also a possibility of evolution.
Nature as only recently evolved a mechanism for making improvements
by design.

	Perhaps I misunderstand the point, and neural nets are only
being compared with hormonal information processing which also evolved.

#. The conclusion that the cephalopod light show is for communication
requires substantiation.

Is the question about the giant squid rhetorical?

#. The speculation about intelligence-mediated conflict between sperm
whales and squid is far out.

#. Modern man probably is the most individually intelligent animal.

#. The two billion years of evolution were apparently mostly spent at
the lower levels - perhaps even at the prokaryote, pre-sex stage.

#. I agree with the general optimism, but "there are known brute forc
solutions to most AI problems" is simply false.

#. I demand more than a factor of 16 for not having to apply every
operator to every part of the picture, but at your estimates processing
an image should require only 160,000 seconds to fully process a
1000x1000 picture.  You have already used several times that amount
of computer time.
With the KL-10, this should go down to 32,000 seconds, so the
brute force solution of the vision problem should be demonstrable.