Re: Scenario for early hard takeoff

From: Daniel Burfoot (daniel.burfoot@gmail.com)
Date: Tue Sep 04 2007 - 20:45:14 MDT


On 9/5/07, Benjamin Goertzel <ben@goertzel.org> wrote:
>
> What induces them to download and install the thing? To trust the thing?
>

This may be orthogonal to the line of thought you were pursuing, but what if
the software's purpose is just to STORE data uploaded by the user?

So, user has lots of photos and movies. User stores the photos and movies in
the globally distributed internet cognition engine. The engine does two
things:

1) distributes the bits over many machines, thereby providing redundancy.
2) *compresses* the bits, thereby saving space and network bandwidth.

As Matt has argued, strong compression of arbitrary data is equivalent to
AI. Strong compression of images and movies is equivalent to computer
vision. A core element of statistical learning theory is that in order to
justify complex models, one must have a suitably large set of data - the
engine by its very nature has access to immense amounts of data and thus can
develop an enormously sophisticated model.

One could imagine also that different nodes in the network develop different
techniques for representing data - different encoding schemes. So for
example Node A becomes very good at representing people's faces, while node
B becomes good at representing cars. It wouldn't be hard to implement some
sort of scheme where A and B negotiate to decide which one should have
responsibility for storing a given image, by comparing the code length of
the image using A's model vs. B's model. Then, as more images of faces are
shunted to A, it can become more and more efficient at representing faces
(one might call it the Rembrandt node).

Amusingly, the system would also become adept at reverse engineering. Lots
of people would want to store large executables. The most parsimonious
description of an executable is its source code, so in order to succinctly
represent such a set of files the system would just have to learn how to
back-infer the source of a compiled file.

So in the barter the user gets reliable and efficient data storage, while
the system gets lots of computational resources and perhaps more
importantly, data.

Dan



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