Re: Property rights

From: Stuart Armstrong (
Date: Tue Apr 22 2008 - 09:38:50 MDT

> > What is interesting about a "super AI" is that it could run an
> > efficient command economy. And it could run it in ways that minimise
> > the negative philosophical consequences to people. Would we want that?
> It can do so iff people can express demand curves. If it knows I will buy 3
> apples at $1 each, it has enough information to manage the status quo (which
> the market can do just fine already). What it needs to know is how I would
> react to apples at $2 or $4 -- and that's not even getting into how I react
> when oranges fluctuate, too. A partial solution is for all of us to start
> ceding control of our small purchases to automated agents that will, for
> example, order more of stuff we like when it's cheap, hedge against the risk
> of huge price increases, etc. I imagine that this will happen with fuel
> first -- I suspect that most drives would prefer using derivatives to get a
> flat price for gas to paying a fluctuating price at the pump.
> The problem with this is that the AI is suddenly redundant. The information
> it gets from the agent transactions is already being used by the agents to
> accomplish the same goals as the AI. All it can do to make things run better
> is to create central clearinghouses for this information. And why would we
> need an AI to know that collecting this information and managing
> transactions would be a great business. All one has to do is look at the
> current market value of the NYSE or CME or NASDAQ (the companies running the
> exchanges, not the products traded on them) to know that this is an
> incredibly profitable business.

Yes, but an AI could incorporate externalities, and probably
out-perform the market (real markets are not pareto-efficient, and
even when they are approximately so, there may be non-local peaks that
AI's could move to).

But agreed - the AI would have to be monstrously beyond us to run the
whole economy. But it is possible that a 90% AI run-10% free market
(or some other proportions) would turn out to be the most efficient.

> when there is a *collection* of AIs

This gets interesting, but scary for humans. The law of comparative
advantage does not work when there are scarce resources involved; even
if uploaded, we would probably consume much more resources than the
AI's would find useful. If there are competitive AI's, then we may be
deemed expendable...


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:02 MDT