Re: Error detecting, error correcting, error predicting

From: Krekoski Ross (
Date: Wed May 07 2008 - 17:36:24 MDT

On Wed, May 7, 2008 at 10:02 PM, Matt Mahoney <> wrote:

> --- Krekoski Ross <> wrote:
> >
> > I've always wondered about that argument. Won't any given model have a
> > different set of inputs?
> It depends on the level of intelligence.

fair enough.

> If it is high enough, it will
> have a very detailed model of your environment.

Certainly. Of course it can never have a perfectly detailed model of my
environment, which leads me to wonder if perfect simulation is ever
possible. If quantum mechanics interacts at any level with classical
mechanics, it would be impossible for even a quantum machine to simulate a
physical environment of similar size to an arbitrarily high degree of
precision. As you say though, a 'very detailed' model is certainly possible.

> It could know (and be
> able to manipulate) the state of every neuron and synapse through
> simulation, direct observation, or both.

And of course we are possibly interacting in a simulation at this very
moment as well. This in fact would require less processing power than a
perfect simulation as only the physics of what is immediately observable and
causally interactive with me would need to be simulated. I would probably
not notice a difference if some of the quantum interactions get
'approximated'. Of course, under such a scenario, science in a simulation
would hit a complexity wall at a specific scale of observation.

> If it is at a moderate level
> (like humans are to animals), we should expect it to achieve these godlike
> powers very quickly (on our time scale) through recursive self
> improvement.

Assuming there are no barriers of which we have no conception yet, yes.


> -- Matt Mahoney,

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