Re: Simulation argument in the NY Times

From: Matt Mahoney (matmahoney@yahoo.com)
Date: Sun Aug 19 2007 - 18:44:03 MDT


--- Damien Broderick <thespike@satx.rr.com> wrote:

> At 12:14 PM 8/19/2007 -0700, Matt Mahoney wrote:
>
> >If you define the goal of evolution to be the continuation of
> >life
>
> If you define the goal of stars to be the irradiation of their planets
>
> If you define the goal of water to be its running downhill to the sea
>
> Stop hypostatizing "evolution." This is SL4, you should know better than
> that.

Sorry, I did not express myself correctly. I mean that a population of
organisms is intelligent, and evolution is the algorithm. My argument is that
the intelligence of the population is proportional to its algorithmic
complexity, the length of the shortest program that could output all of the
DNA of the population.

Legg and Hutter's definition of universal intelligence is the expected
accumulated reward signal of an agent interacting with a Solomonoff
distribution of environments. The agent and environment are modeled as a pair
of interacting Turing machines that exchange symbols at each step, and in
addition, the environment outputs a numeric reward signal. (The environment
thus defines a goal as well).

In this sense, a population of organisms (even those lacking nervous systems)
is more intelligent than (a model of) water flowing downhill, first because
living organisms can achieve the same goals in a larger set of environments,
and second, they can achieve a richer set of goals. For example, water in a
cup will fail to achieve its goal, but if I give a cup full of organisms the
same goal (by killing those which don't move downhill), then some of them may
crawl out, and future generations will be even more successful. Second, I
could similarly train a population to flow uphill, or to convert sugar to
alcohol, or a million other goals. The number of goal-environment pairs in
which the goal can be achieved increases with the biodiversity (information
content) of the population.

As for the earth falling into a black hole, yes, it is also possible for a
human to be eaten by a less intelligent shark. But on average, the more
intelligent system will usually win.

-- Matt Mahoney, matmahoney@yahoo.com



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