Re: [sl4] Comparative Advantage Doesn't Ensure Survival

From: Stuart Armstrong (
Date: Mon Nov 24 2008 - 02:35:36 MST

Dear Tim,

I agree with your point, but I think your rebutal is flawed. Your
assumptions can be questioned for human-AI interactions, new, similar
models can be generated (maybe with the assumption that the humans who
control the food supply themselves won't starve), etc...

The core critique of comparative advantage is the ressources needed.
Put simply, comparative advantage does not work with ressources that
are scarce. If there are two men on a desert island, and only enough
food for one, then there is no incentive for the stronger to trade
(and reduce them both to walking skeletons). If there is only one
tree, and one man wants to build a canoe, the other a house, then an
equitable division is not going to happen if there is not enough wood
for both of them.

One idea about an out-of-control AI is that it will try to maximise
its control, and use of, ressources. The fact that the AI can generate
a widget or a potato for 1 energy ressource, while humans require 1
000 000 and 10 000 000 respectively, is not an incentive to trade
potatoes for widgets. Instead, it is an incentive to take over those
millions of energy ressources that humans are so crassly wasting
(coercice versus non-coercive arguments don't get into it - such an AI
would be sufficiently smart to take over the entire economy through
what we'd deem non-coercive methods, if it wanted to).


2008/11/24 Tim Freeman <>:
> I came across a section in "Beyond AI" by J. Storrs Hall saying that
> the economic law of comparative advantage implies that we can't be
> out-competed by AI's. I have also seen other people making the same
> claim. The argument seems invalid to me, and I wrote a rebuttal at:
> I'd like to be wrong here. If anyone sees a counter to the rebuttal,
> please let me know.
> --
> Tim Freeman

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