Re: [sl4] Rolf's gambit revisited

From: Matt Mahoney (
Date: Mon Jan 12 2009 - 07:23:09 MST

--- On Mon, 1/12/09, Norman Noman <> wrote:

> Good lord, it doesn't have to be nearly this
> complicated. Here are some
> easier solutions:
> AI(pi) at time step 10000 is simulating AI(pie) at time
> step 99000, who is
> simulating AI(pi) at time step 98000, etc. As time advances
> and they eat
> more planets they grow, and so they have no trouble
> simulating earlier,
> smaller versions of each other and themselves.

Simulating past versions of each other does not allow 2 way communication.

> OR
> They simulate each other not just with a time lag, but
> actually slower than
> real time. In this case, the simulations can be carried on
> forever without
> eating significantly into the AI's processing capacity.

Same problem.

> OR
> They simply cut some corners! one simulated universe is
> full of pie, the
> other is full of computronium sifting pointlessly through
> endless random
> digits. In both cases, you can simply not simulate that
> part, and say you
> did, and spare yourself a lot of needless work.

Simulating approximations (which we always do in practice) means making guesses.

-- Matt Mahoney,

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