RE: MW QT AI please phone home; another path to a Singularity?

From: Damien Broderick (
Date: Sun Apr 21 2002 - 19:10:13 MDT

At 03:24 AM 4/22/02 +0300, Mikko Rauhala wrote:

>> but it seems to me like the whole scenario seems awfully
>> "this world"-centric. Reminds me of the whole Aristotle thing. If
>> communications across worlds is possible, then shouldn't we already
>> have been contacted by now?

>Unless, of course, this hypothetical act of communication to a parallel
>world in itself also splits the target world into branches where said
>communication a) occured b) didn't occur.

In this world, I've been off-line for two days while unimelb washed their
computer or something. Having now got back in touch with the real world,
it's been amusing watching everyone here recapitulate the lines of thought
that James and I ran through in about 10 minutes over sushi.

I do think that by now we should have been contacted, since a primitive
version has been tried (ineffectively in this branch, but arguably some
other versions might have been subtly superior). I should add, for anyone
wo failed to read between the lines of the first post, that I deliberately
masked some of the details in my earlier post.

The thing to remember, if I understand this correctly myself, is that, in
MW QT, branching in such cases sets out from *here and now*. Entanglement
needs to be retained. The Copernican assumption does not mean that
*anything* can follow from the state I'm in right now; only those events
that are consistent with my present history will occur (well, except for
some remotely unlikely tunneling effects).

Is it unconscionable to try such an experiment because it seems bound to
yield at least some proportion of hellish worlds? Forget Ben's cosmolethal
farts, this applies just as readily to Eliezer sitting in his room
thinking. In some quantum superpositions, evil programs emerge from his
meditations and slip past his guard, or subvert him entirely and generate
his evil twin. In others, despite the purity of his soul evil programs
result from an unnoticed coding error as his fingers slip on the keyboard.
In yet others, a bitflip does it. Still, the prospect of Bad Gods being
built is certainly one reason why I've delayed doing anything along these
lines myself. I'm hoping for a suitable ai theodicy to emerge from
discussions like this one; thanks to everyone for playing. (And my
justification for clogging up the list with it is that the issues do seem
to be somewhat related to ai creation by, coff coff, conventional means.)

Damien Broderick

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:38 MDT