DPRG mailing list archives: [DPRG] SRS thread

Fri Apr 27 16:34:52 CDT 2007

On Fri, Apr 27, 2007 at 02:20:45PM -0500, dpa wrote:
> Howdy,
> 
> This post from a thread over on the SRS list server is
> worth a read:
> 
> <http://tech.groups.yahoo.com/group/SeattleRobotics/message/30845>

He writes
+The problem of the brain can be stated in a lot of ways. Here are mine.
+
+There are 10^15 neuron synapses in the 10^11 neurons of the brain. These
+are - literally - astronomical numbers. This is equivalent to 1,000,000
+gigabytes at 8-bits of resolution per synapse. 

The brain is a very, very complicated thing. But...

I am not convinced of 8 bits of resolution per synapse, so I think I
prefer Hans Moravec's estimate of an order of magnitude less storage.
Also, making hand-waving analogies to the information theory of what
bit error rates do to channel capacity, and hand-waving appeals to how
high the random error rate of the brain seems to be compared to what
we tolerate in our integrated circuits and mass storage, I suspect a
fairer analogy would knock off at least another order of magnitude.

And note that the brain at birth seems to be completely specified by
the genome, and while the raw human genome is very complicated (on the
order of a CDROM of data required to store it after using a basic
compression algorithm), it is not astronomically complicated, and not
all the genes apply to the brain, and not all the brain genes apply to
deep things (as opposed to shallow rules like predisposition to be
interested in faces, and to like sugar and symmetry, and to be scared
of heights, and so forth). My rough guess, from what little I know
about the mapping from genes to embryonic development, and from
suggestive evidence about how much of the genome doesn't seem to be
under intensive selective pressure, is that the stuff in the genome
describing the fundamental design of the brain is more like the
complexity of a 10Mbyte .tar.gz file of computer code than the
complexity of a 100Mbyte .tar.gz file of C code, and I will be very
surprised if it turns out to be as complicated as a 1000Mbyte .tar.gz
file of computer code. So it's still incredibly complicated, but it
looks like the complexity of a tightly written 10Mbyte tarball of neat
software (perhaps a multitasking operating system, relational
database, highly optimizing compiler, numerical algorithms library,
networking stack, and half a dozen other things) which happened to be
written by a team of madmen. There's plenty of imprecision in that
estimate, but not enough imprecision for it to be something
astronomical (like a team of madmen rewriting not only all the
software which has ever been written since 1950, but all the writing
of any sort that the human race has ever produced).

And this is an upper bound; we may be able to find more logical way to
do things. (The eye does all sorts of marvellous things, but our video
cameras do too, without having to do some weird stuff like
compensating for blind spots. Thus, it could turn out that the genes
which describe the design of the eye are rather more complicated than
one would think from the difficulty of designing a video camera.)

(Then, appealing to Moravec's estimates again, this time for the
computational speed of the human brain, you need to run the program on
a 100 teraMIPS computer for 10-20 years before it gets very
sophisticated about dealing with the world. So I'm not trying to
trivialize the problem of getting a computer comparable to a human
brain. But I am suggesting that the amount of design work required
could be less than many people think.)

-- 
William Harold Newman <william.newman at airmail.net>
PGP key fingerprint 85 CE 1C BA 79 8D 51 8C  B9 25 FB EE E0 C3 E5 7C
Ubi saeva indignatio ulterius cor lacerare nequit. -- Jonathan Swift's epitaph