Re: How hard a Singularity?

From: James Higgins (jameshiggins@earthlink.net)
Date: Sat Jun 22 2002 - 18:55:02 MDT


At 07:00 PM 6/22/2002 -0500, you wrote:
>James, say you upload Ben as you suggest. Now he can think at electrical
>rather than neurotransmitter chemical speeds, lets say 1000 times
>faster. Ben can now maintain thousands of details simultaneously in
>consciousness rather than 4 to 7 as a biological. Now Ben can flawlessly
>cross correlate this data with no chance of logical error as computers do
>now. If you think biologicals can ovoid logical errors just try to adding
>five fifty-digit numbers in your head. And being in silicon gives Ben an
>ability he didn't have before, the ability to monitor and adjust his own
>cognitive functions, improving concentration, and memory accuracy, and
>expanding any bio bottlenecks in his cognitive architecture. The new
>silicon Ben is a wiz compared to the old bio Ben.
>
>Mike.

Sorry, I don't buy it.

You just jumped the whole point. If the first AI ran "1000 times faster"
and "could maintain thousands of details simultaneously in consciousness"
then it wouldn't be HUMAN EQUIVELANT, now would it? You skipped
human-equivalent AI and went straight to super-human AI.

Plus, your assuming an AI would inherit everything a computer could do just
because it runs on computing hardware. Nothing says that the AI can keep
more than 4-7 details in consciousness simultaneously. Who knows, maybe
the 1st successful human-equivalent AI can keep only 3 things resident at a
time but thinks 3 times faster (which ends up balancing out, thus still
being "human equivalent"). Also, the speed of the hardware is
irrelevant. Doesn't matter if it was running on optical chips that ran 10
million times faster than human neurons, if its human-equivalent it roughly
thinks of the same things we do given the same amount of time.

It is equivalent to a single human intellect. Thus it does nothing more
than add one more mind to the project (except, as I mentioned, that it can
work more than 60/hours a week). Then again, if it is truly
human-equivalent maybe it gets burned out and bored if it works on the same
problem more than 80/hours a week, which could throw a monkey-wrench into
everyone's calculations. We won't know until we get there...

See my post "How to gauge positive progress (was How hard a Singularity?)"
for further details.

James Higgins



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:39 MDT