Re: [sl4] JAGI submission

From: Eliezer Yudkowsky (sentience@pobox.com)
Date: Wed Nov 26 2008 - 17:02:50 MST


On Wed, Nov 26, 2008 at 3:52 PM, Peter de Blanc <peter@spaceandgames.com> wrote:
> Matt Mahoney wrote:
>>>
>>> Matt, RSI gains O(log t) bits compared to a Kolmogorov
>>> prior, while evolution gains O(t) bits compared to a uniform
>>> prior. You can't compare these two quantities.
>>
>> No, evolution gain has an upper bound of O(t) measured by Kolmogorov
>> complexity. (It might be less). It does not matter how information is
>> inherited. Each fitness decision that results in death with 50% probability
>> transmits one bit of information.
>>
>> -- Matt Mahoney, matmahoney@yahoo.com
>
> OK, I agree that this is an upper bound. Unless you can prove a *lower*
> bound, this doesn't allow you to compare evolution favorably to RSI.

Not to mention that *Kolmogorov complexity is completely irrelevant to
intelligence*.

And evolution can be simulated in closed systems, so it's not going to
do anything that recursive self-improvement couldn't do; for one
thing, an RSI device could just run evolutionary tournaments inside
itself...

...sheesh. Try *visualizing* what the math is supposed to *mean*, please.

-- 
Eliezer Yudkowsky
Research Fellow, Singularity Institute for Artificial Intelligence


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:03 MDT