Abstract invariants

From: Emil Gilliam (emil@emilgilliam.com)
Date: Sat Jul 17 2004 - 19:19:05 MDT


Again, quoting the Collective Volition page:

> Friendly AI requires:
>
> 1. Solving the technical problems required to maintain a
well-specified
> abstract invariant in a self-modifying goal system.
(Interestingly,
> this problem is relatively straightforward from a
theoretical standpoint.)
>
> 2. Choosing something nice to do with the AI. This is about
midway in
> theoretical hairiness between problems 1 and 3.
>
> 3. Designing a framework for an abstract invariant that
doesn't automatically
> wipe out the human species. This is the hard part.

It seems that in order to understand a Friendly abstract invariant with
the deepness of (2), and to understand what does or does not fit the
spirit of this abstract invariant as humans would judge it, a seed AI
would have to know an immense number of details about human brains. If
this is so, then there may be no practical way for the seed AI to know
all these details without scanning actual humans -- but, as SIAI's
strategy currently goes, we don't want it to have any capability of
this sort until takeoff time, and by that time the job of the Seed AI
programmers should be *done*.

Is "finding a way out of this deadlock" a useful way of characterizing
any part of (3)'s complexity?

- Emil



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:48 MDT