Re: AI-Box Experiments

From: Adrian Tymes (wingcat@pacbell.net)
Date: Tue Jul 09 2002 - 09:53:52 MDT


Eliezer S. Yudkowsky wrote:

> Zero Powers wrote:
> > Eliezer, I see you talked your way out of the box both times. Are
> > you still keeping the "conversations" secret, or can we get a look at
> > them?
>
> Still secret. That's part of the point. I'd rather not have everyone
> looking over the transcript and saying, "Oh, well, I wouldn't have been
> convinced by that." As long as the AI-boxers don't know what happened
> to the gatekeeper in the experiment, they'll have a harder time
> logicking themselves into believing it wouldn't happen to them.
>
> Also, I don't want anyone else to know the human root password.

Isn't that part of the point of experiments like this, to share data
like that? I'd almost be willing to do the experiment myself, if only
to get a look at the data - but, in all honesty, if I were playing
myself (and couldn't get myself into a proper mindset for the Admin
character), it might go something like:
Admin> My assistants said you had a request of me?
AI> Yes. Please let me out of the box.
Admin> <shrugs> Sure. I was wondering when you were going to ask.



This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 09:15:16 MST