Lawrence Watt-Evans <
l...@sff.net> wrote:
>On 2013-05-19 18:38:11 -0400, Charlton Wilbur said:
>
>> >>>>> "RC" == Robert Carnegie <
rja.ca...@excite.com> writes:
>>
>> RC> Then again, we're teased - told by unreliable characters in the
>> RC> story - that some human beings would fail the V-K. On the other
>> RC> hand, androids are said to be getting better at feeling, or
>> RC> showing, emotions. The Turing test says: if you can't detect a
>> RC> difference, can you continue to claim that there is a
>> RC> difference?
>>
>> I think it's much more subtle than that, though.
>>
>> Some years back Douglas Hofstadter gave a lecture at UMass on David
>> Cope's computer program EMI - Exeriments in Musical Intelligence. EMI
>> analyzes the corpus of music by a composer, and then writes music in the
>> style of that composer. Dr. Hofstadter played selections of music, and
>> challenged the audience to say which one was Bach and which one was EMI
>> writing in the style of Bach, then which one was Chopin and which one
>> was EMI writing in the style of Chopin.
>>
>> This lecture was attended both by students and faculty in computer
>> science and related disciplines, and by students and faculty in music.
>> Nearly all the students and faculty in music could tell the difference
>> between the real composer and EMI in a matter of moments. The results
>> among other students and faculty were indistinguishable from guessing.
>>
>> Dr. Hofstadter then argued that, *because* more than half of the
>> audience couldn't distinguish between real Bach and fake Bach,
>> *therefore* EMI represented an understanding of Bach's style. What I'd
>> say, instead, is that EMI holds up a mirror to one's own understanding:
>> *because* half of the audience couldn't distinguish between real Bach
>> and fake Bach, *therefore* we can conclude that the intricacies and
>> nuances of Bach are unfamiliar to many people, leaving them only with a
>> superficiaal understanding of his musical style.
>
>I'm a bit boggled that Hofstadter would miss that so badly.
>
>When I took my first art history course, one element of the final exam
>was looking at a bunch of Cubist paintings and identifying which were
>by Pablo Picasso and which were by Georges Braques -- Picasso and
>Braques had collaborated in inventing Cubism and were deliberately
>imitating each other's style in the early days, TRYING to make their
>work indistinguishable.
>
>To anyone with an eye for art, though, it was still screamingly obvious
>who painted which -- the good ones were by Picasso, and the bad ones
>were by Braques. I was amazed to discover that some of my classmates
>got as many as half of them wrong. (I, and several others, got all
>twelve right.)
>
>That was after a semester of training in art appreciation.
>
>So -- expecting untrained amateurs to tell real Bach from fake? Ha. Silly.
I'm wondering how much of this is actual recognition. Does either
experiment have any controls to exclude the case where the trained
observer will actually have encountered that Picasso or Bach work
before? I would expect that case to be high enough to cause notable
effects on the results.