Re: answers I'd like from an SI

From: Stathis Papaioannou (
Date: Fri Nov 16 2007 - 17:30:43 MST

On 17/11/2007, Norman Noman <> wrote:

> I don't think anyone here would argue that a superintelligent AI could
> take over the internet with relative ease. There are always security
> holes that we don't realize, and as soon as you can manage a buffer
> overflow you can get a computer to execute any code you want. The
> human brain is messier, more complicated, and less modular, but it is
> still a "robust yet fragile" system. It has strange weaknesses that
> allow seemingly harmless types of input, such as words, to set off
> catastrophic changes within it. Just because I can't talk someone into
> a psychotic state doesn't mean it can't be done.

Maybe if you asked the AI it would just tell you that it was
impossible, or only possible in a subpopulation of people (about 1%)
who have a susceptibility to developing psychosis. Maybe it would also
tell you that a reliable strategy for taking over the world given just
an Internet connection is also impossible, or computationally
intractable, or that the question of its possibility is undecidable.
It's irrational to assume that an advanced enough AI would be able to
do *anything at all*.

Stathis Papaioannou

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:00 MDT