Information Technology Reference
In-Depth Information
someone in the audience always reacted by indicating: “young man, what a hap-
less attempt to put into numbers a complex phenomenon that requires a living and
experienced human being to judge”.
My reaction then was, oh yes, I see the difficulties, but that's exactly what we
must
do! And we will, I threatened them. I guess, looking back without anger, they
shut up and sat down and thought to themselves, let him have his stupid idea. Soon
enough he will realise how in vain the attempt is.
He did realise, I am afraid to say.
In the early 1960s, Birkhoff's quotient of order over complexity was taken up
again (by Bense, Frank, Gunzenhäuser, Moles, myself). It was given a promising
interpretation in information theoretic terms. Helmar Frank, in his PhD thesis of
1959, defined measures of surprise and of conspicuousness (of a sign, like a colour,
in an image). All these attempts were bold, strong, promising, radical. But they
were really only heroic: the hero always dares a lot, more than anyone else, stupidly
much, and always gets defeated and destroyed in the end.
I am sceptical about computer evaluations of aesthetics for many reasons. They
are a nice exercise for young people who believe in one-sidedness. Human values are
different from instrument measures. When we judge, we are always in a fundamental
situation of forces contradicting each other. We should not see this fact as negative.
It is part of the human condition.
Harold may be the one who, from his forty years of computational art practice
that took him so close to the heroes of AI, would be able to pave the way. But even
he is sceptical. “I don't know what it is that makes them (the computer-generated
images coming from his program) special”, he says. He continues to say he doesn't
know how to describe “what it is in computational terms”.
If we ever wanted to apply algorithmic methods to aesthetic evaluations, we must
first be able to describe what we want to measure. Such a description must be for-
mal and computable. So an explicitly formalised and algorithmic description is what
would be needed. And those descriptions would be of works that we are used to call-
ing “art”. We all know the situation where five of us are around a small collection of
pictures. We discuss them. We describe, bring in comparisons, develop our judge-
ments against the background of our own lives, and of the current situation and
discussion. We come up with a judgement in the end that doesn't totally satisfy any
participant of the meeting. But all of us feel quite okay. We think we can justify the
judgement. Tomorrow it could easily turn out to be different. This is how complex
the situation of an evaluation is.
In Toronto in 1968/69, I wrote a program that I proudly called
Generative Aes-
thetics I
. It accepted as input a series of intervals for information aesthetic mea-
sures. They defined boundary conditions that must not be violated. The algorithm
then tried to find a solution maximising the aesthetic measure against the boundary
conditions. Its result was, of course, only a (probability based) distribution of the
colours.
Just see what that program's task was: given a set of numeric (!) criteria, deter-
mine a “best” work that satisfies certain given evaluations. Isn't that great? I thought
it was. And I was 29 years old.
