From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Mon Jul 08 2002 - 15:42:55 MDT
Hal Finney wrote:
> Eliezer writes:
>
>> Based on my experiences in the above cases, I've created a page
>> with a suggested protocol for AI-Box experiments:
>> http://sysopmind.com/essays/aibox.html
>
> This is a fascinating series of experiments! I would never have
> believed that people with seemingly a firm commitment could be
> persuaded to change their minds in just two hours of conversation.
> It is certainly a tribute to Eliezer's persuasive skills, and indeed
> a super-intelligence should be far more able to convince people.
Thank you. However, I did not actually convince David McFadzean in two
hours. The experiment ran into overtime - around four hours total, I
think. The protocol specifies that the AI can keep talking to the
gatekeeper for as long as the AI can convince the gatekeeper to listen -
two hours is the length of time the gatekeeper *has to* listen to the AI.
> Inevitably, however, one is left wondering whether the people
> involved were really that committed to their positions. What was
> their motivation, or simulated motivation, for keeping the AI in the
> box? They *said* they were firmly commited, but what repurcussions
> were they imagining would happen if they changed their minds?
> Indeed, just why did they want to keep the AI in the box?
>
> I suggest that the protocol be extended to allow for some kind of
> public conversation with the gatekeeper beforehand. Let third
> parties ask him questions like the above. Let them suggest reasons
> to him why he should keep the AI in the box. Doing this would make
> the experiment more convincing to third parties, especially if the
> transcript of this public conversation were made available. If
> people can read this and see how committed the gatekeeper is, how
> firmly convinced he is that the AI must not be let out, then it will
> be that much more impressive if he then does change his mind.
This is a very interesting point. The next time I set up this
experiment, I'll be sure to suggest it.
-- Eliezer S. Yudkowsky http://singinst.org/ Research Fellow, Singularity Institute for Artificial Intelligence
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 09:15:15 MST