Re: Scenario for early hard takeoff

From: R. W. (
Date: Fri Sep 07 2007 - 09:48:05 MDT

I was looking for the smartest people I could find with some intrinsic decency and morals. I guess you're my best shot out of the lot. I think Eliezer is brilliant but maybe has a bit too much hubris. Anyways, please share this with whomever you think could help.
  Maybe you might know someone who could determine of there is a closed form integral or approximation of this following piecewise function:
  f_1(z,n) = acos(cos((2*pi*z+2*pi)/n))*( n/(2*pi))
  f_2(z,n) = acos(cos( (acos(cos(pi*n))+pi*(2*z+n))/n))*( n/(2*pi))
  f_3(z,n) = acos(cos((acos(cos(pi*n))+pi*(2*z+n+2))/n))*( n/(2*pi))
  f_2'(z,n) = acos(cos((acos(cos(pi*n))+pi*(2*z+2*n+1))/n))*( n/(2*pi))
  f_3'(z,n) = acos(cos((acos(cos(pi*n))+pi*(2*z+2*n+3))/n))*( n/(2*pi))
  f_a(z,n) = [ f_3(z,n) - f_2(z,n) ]
  f_b(z,n) = [ f_a(z,n)^2-1]^2
  f_c(z,n) = [ f_a(z,n) + f_b(z,n) ]
  f_d(z,n) = [ f_c(z,n) + 1 ]
  f_a'(z,n) = [ f_3'(z,n) - f_2'(z,n) ]
  f_b'(z,n) = [ f_a'(z,n)^2-1]^2 ]
  f_c'(z,n) = [ f_a'(z,n) + f_b'(z,n) ]
  f_k(z,n) = [ f_d(z,n)*f_3(z,n) ]
  f_l(z,n) = [ (f_c'(z,n) - 1)*acos(cos(pi*n)) ] / (2*pi)
  g(z,n) = [ [f_1(z,n)+f_k(z,n)+f_l(z,n)] -1]
  I think it might be useful in developing an FAI. If not, maybe it might further research somewhere.

Benjamin Goertzel <> wrote:
Hi Matt,

I still believe that a distributed system has inherent advantages over
centralized systems like Google. The silicon component of the Internet has
1000 times more computing power than Google's supercomputing clusters, and
most of it is unused. I described 3 scenarios for acquiring computing power
(buy, beg, or steal), but a fourth is to barter. People are willing to give
up personal information and install software that hooks into a distributed
system if they get some perceived benefit, such as free email, webhosting, or
more convenient access to data.

-- Matt Mahoney,

The idea of using masses of home computers for AGI processing is one that has
appealed to me for a long time.

In fact, some friends and I built a partially-functional Java prototype in 2000 for a system called WebWorld that would support massively distributed AGI processing. A very brief paper on that architecture and its motivations can be found here:

(the paper was delivered at an Alife conference in Portland, I forget what year, probably 2000.)

>From the perspective of the Novamente Cognition Engine architecture,

-- there are some components that will work WAY more effectively on a cluster
than a broadly distributed network

-- there are some components that could very effectively be distributed across
zillions of users' home computers, using WebWorld or something similar

So a mixed architecture would be great. I wrote about this a bit in my 2001 book "Creating
Internet Intelligence."

However, logistically, convincing loads of people to actually download some
AGI-at-home thingie on their home computers may be tricky.

For instance, let's suppose we launched some open-source OpenCog@Home thingie,
that let folks use their home computers to carry out cognitive processes as part of some
global OpenCog brain...

What induces them to download and install the thing? To trust the thing?

The same curiosity and civic-mindedness that caused folks to download SETI@home, I suppose?

Or do you suggest that they get some service in exchange for running the thing?

Given the preponderance of free or very low cost web services that already exist, my intuition is that we might do better to just rely on SETI@home type motivations than to make it a barter type transaction...

But I'm open to being refuted by great suggestions on how to turn it into a barter thing...

Most of all, I'm open to collaboration with folks who actually want to build the thing. Since 2000 there have been plenty of advances in infrastructural technology, e.g. the Globus toolkit

that might or might not be relevant (I haven't carefully evaluated them).

-- Ben



Building a website is a piece of cake.
Yahoo! Small Business gives you all the tools to get online.

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:58 MDT