[ExI] turing test, was: RE: ai class at stanford
stefano.vaj at gmail.com
Tue Aug 30 10:52:10 UTC 2011
On 30 August 2011 03:35, spike <spike66 at att.net> wrote:
> is a far more interesting conversationalist than at least some humans
This is why "success" in a Turing test has no meaning without a context, and
I prefer to speak of "performance" in Turing test (such as: what percentage
of interviewers you manage to deceive for how many exchanges, in average).
I am also pretty sure that there human beings which do sound more like
emulations (and perhaps poor ones...) than others.
Let us say that a fully Turing-qualified emulation would be one whose
results in any finite number of exchanges would be indistinguishable from
the average results of human beings, so that its positives and negatives
would be just the matter of wild guesses from the side of interviewers.
Another more interesting issue is that of Turing-specific tests as opposed
to Turing-generic, that is the test where the emulation does not simply
pretend to be human, but pretend to be a given individual, say, Mr. Smith.
This is a game which we can also play, and I am pretty sure that emulations
will surpass other human beings soon enough.
When an emulation will perform as well as Mr. Smith returning from a
five-years travel around the world or the hospital after a serious accident,
I think that everybody will recognise him as Mr. Smith, period.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the extropy-chat