From: Rafal Smigrodzki (rafal@smigrodzki.org)
Date: Thu Jun 19 2003 - 20:02:31 MDT
Kevin wrote:
> 2.) Even if it were a human on the other side of the conversation,
> some people are bound to think it is a machine.
### It only matters if the examiner can provide a statistically valid
differentiation between the AI and humans. If we state as null hypothesis
that the AI is indistinguishable from humans, then the expected success rate
of the examiner will be 50%. If the .95 confidence interval around his
success rate does not contain the expected result of the null hypothesis, it
means the AI has to try harder.
----------------------------------
> 3.) Different people would expect different responses to the same
> questions when they are deciding whether or not it is an AI or human,
> and finally,
> 4.) An AI that was nowhere near as capable as a human intellectually
> still may pass simply because even an educated human may just think
> they are dealing with an ignorant human on the other side.
>
> Am I way off here in thinking that we may never be able to divise a
> sufficient test simply because as human beings, we have a wide range
> of intellectual capabilities that is too broad to nail down?
### We are talking here about AIs with average human or better IQ - we do
not need to consider AIs with an IQ of 0 to 5 (profound MR, lying in bed and
drooling).
--------------------
Can we
> even use this test to ensure that a human is human?
### No. But we can prove that a non-human is indistinguishable from a human.
Rafal
This archive was generated by hypermail 2.1.5 : Thu Jun 19 2003 - 17:12:31 MDT