[ExI] More thoughts on sentient computers

[Giovanni Santostasi]

Who cares about the Turing test?
Even the idea that the machine has to "imitate" or fake human behavior. I
don't care about that and in fact it is a bad idea. We don't want
machines to fake anything. It is already obvious when machines try to
pretend to be humans in these convo people have where they provoke it for
example.
I would rather do more interesting experiments like test the theory of mind
of the AI like in this paper:
Theory of Mind May Have Spontaneously Emerged in Large Language Models
https://arxiv.org/abs/2302.02083?fbclid=IwAR2WbdZZRN5afuYWYKybjijTejN08I3EGUKkqt1__B793OWRGWid8slygeE

An another emergent property.

Giovanni

On Mon, Feb 20, 2023 at 8:46 AM Dave S via extropy-chat <
extropy-chat at lists.extropy.org> wrote:

> On Monday, February 20th, 2023 at 11:08 AM, spike jones via extropy-chat <
> extropy-chat at lists.extropy.org> wrote:
>
> The Turing test indicates only one class of sentience (he said, as we now
> have software passing the Turing test regularly (demonstrating our
> collective desperate determination to move the AI goal posts again.))
>
>
> There's no "the" Turing Test. Even Turing described at least two
> variations. And they weren't defined sufficiently to be practical
> tests--and likely weren't intended to be used that way. The Wikipedia page <
> https://en.wikipedia.org/wiki/Turing_test> covers all of the variations
> and problems with implementation pretty well.
>
>
> In my opinion, a useful variation would be for an examiner to have two
> rate-limited text chat sessions active: one with an AI and one with a
> human, neither of which would see the dialogue in the other chat session.
> The AI would be instructed to pretend to be a human. The examiner would
> attempt to determine which subject is human and which is AI. There'd have
> to be other restrictions because things like writing a haiku or limerick
> are easy for, e.g., ChatGPT, but many humans would have a hard time with
> that. A sufficient rate limit could fix that but would slow down the test
> too much.
>
>
> -Dave
>
>
>
> _______________________________________________
> extropy-chat mailing list
> extropy-chat at lists.extropy.org
> http://lists.extropy.org/mailman/listinfo.cgi/extropy-chat
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230222/fda905dc/attachment.htm>