On Fri, Mar 8, 2013 at 7:57 PM, Alan Grimes <[email protected]> wrote:
> Make 10^n copies. ;)

Really, is self replicating robots the only way to make AGI?

But my original question. What would you do with a single human-level
intelligence? Play it against Watson or Deep Blue. Sorry, it is only
human level so it loses. Submit it to the Loebner prize? Sorry, the
software has to run on their computers and all it will say is "Unable
to allocate 10 PB of memory". Rent it out to do work? Sorry, it costs
$1000 per hour for electricity. It's cheaper to hire humans. Give away
your 100M lines of code? Who has a supercomputer that could run it?

This is what happened to Cyc and OpenCog. There are some initially
promising results, then years of slogging away with no apparent
progress, to produce something that nobody can use. All because they
didn't know how to estimate the cost.

> I think you are being pessimistic about your hardware estimates, but fine.
> Let's talk about
> hardware then. There are two basic approaches, 3D nano-crystals and quantum
> computing. You can also jack up performance, by a factor of 100, using
> custom designed chips.

Yes, that's exactly what we need to do. Shrinking transistors like we
always have is not going to work. To get a factor of 10^5 improvement
you would need to make them smaller than atoms. So where are the
hardware designs that we need?

-- 
-- Matt Mahoney, [email protected]


-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com

Reply via email to