From: Dani Eder (danielravennest@yahoo.com)
Date: Tue Jan 22 2002 - 08:55:54 MST
An interesting report on how single neurons encode
facial recognition:
http://www.eurekalert.org/pub_releases/2002-01/m-pis011802.php
This type of research should eventually resolve
Penrose's
hypothesis. If it can be shown that all the
computation
that occurs in the brain can be explained by the
visible
neuron interactions (synapse pattern, connection
strength,
firing rates, perhaps some internal structural
features
in the neuron), then there is no reason to invoke
'spooky quantum' computation.
If there is no easier way to build an AI (easier
defined by the amount of computer hardware required
to implement it), then I expect a sufficiently large
neural network will eventually work. This is, in
fact, how I personally estimate the time to the
singularity.
If you look at chess playing computers, they beat
the best humans when they had specialized hardware
that was estimated to be 3% of a human brain in
total capacity. The implication is that the best
human players devote 3% of their brain to chess,
and the rest to the usual stuff our brains do.
Thus, I take a lower bound on 'when will the
Singularity happen' as when we have computers at
that capacity level available for specific tasks
like optimizing code and designing chips. This
could be sufficient for a runaway acceleration
in computer power. Specifically, 3 billion
neurons x 10,000 synapses/neuron x 100 Hz firing
rate x 1 bit/firing = 3,000 Tbits/sec = 100 Tflops.
When 100 Tflop machines reach the 100th place
in the Top500 supercomuter list I would say they
are available for specific tasks on a full time
basis. This chart:
http://www.top500.org/lists/2001/11/slides/PerformanceDevelopment2.pdf
projects reaching that level in 2012.
If you then say that instead of taking 3% of a
brain, it takes a full brain, that gives you a
date of 2018. Now if you say 1 synapse firing =
1 floating point operation rather than 1 bit, you
get a date of 2024. So if asked, I would say I
expect the singularity about 2018 +- 6 years.
Note that the 2 sigma early date would then be
2006, so I would estimate a small but non-zero
chance of a singularity in the next 4 years.
Daniel
__________________________________________________
Do You Yahoo!?
Send FREE video emails in Yahoo! Mail!
http://promo.yahoo.com/videomail/
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT