Re: ESSAY: Program length, Omega and Friendliness

From: Brian Atkins (brian@posthuman.com)
Date: Thu Feb 23 2006 - 10:54:26 MST


Robin Lee Powell wrote:
>> And that attempting to directly compare compressed filesizes
>> output between two totally different compression systems makes
>> little sense?
>
> To an extent; there are strong theoretical boundaries on compression
> qua compression. Complexity of the substrate is something else
> entirely. This isn't really about compression; it's about the
> complexity of the underlying instruction set for the code in
> question.
>

I think of it as combination because you don't have one part (the compressed
code) working without the other (the system/environment).

If we think of music files, you can have one compression format such as a WAV
file where the music data is converted directly from an analog source into a
simple binary format that can later be replayed essentially directly back
through a simple hardware DAC to reproduce a close approximation of the original
analog signal.

But if you look at the MP3 format, the original analog information is compressed
much further into a digital file size 12x or more smaller than a WAV. This is
possible because the encoder has built in prior knowledge of the expected final
use environment (your ears/brain), knows what it can throw away safely from the
original signal, and also knows it will have the use of special playback
software so it can compress the data further by encoding it in a special way for
that specific playback software.

The end result of either case for the average music listener is they hear the
same song, but even in this simple case there are very different methods used
and 12x different data file sizes. Now attempting to go to an example even
further apart such as DNA vs. computer programs attempting to generate an
intelligence as the output, I am guessing the file size differences may be even
more than 12x, but it's hard to tell as far as I can see currently.

-- 
Brian Atkins
Singularity Institute for Artificial Intelligence
http://www.intelligence.org/


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:56 MDT