Predicting SI Capabilities

From: Billy Brown (
Date: Wed Feb 23 2000 - 10:26:44 MST

Bostrom,N (pg) wrote:
> I have tried to think of some consideration or some test that could
> determine whether this would be possible to do in principle, but haven't
> come up with anything yet; it seems a wide open question. Any ideas on how
> one could settle the issue, short of building a superintelligence and
> trying?

Not in this specific instance, but I would like to offer a useful heuristic.
In human affairs we usually assume that something is impossible unless we
can see how to do it, or unless we know it has been done. This approach
works pretty well because humans can easily imagine all kinds of things that
they can't actually do, and surprising new discoveries are much less common
that hair-brained schemes that don't work.

When speculating about strong SI, OTOH, I think we can get a better
approximation of the truth by taking the opposite approach. If we can't say
for certain that something is forbidden by physical law, it seems reasonable
to assume that an entity with 10^20 times the processing power of our entire
species could find a way to do it. After all, cases in which we fail to
think of a solution that would seem obvious to an SI are going to be much
more common that cases in which we actually test and rule out every approach
that is allowed by physical law.

Billy Brown

This archive was generated by hypermail 2b29 : Thu Jul 27 2000 - 14:04:06 MDT