Information Technology Reference
In-Depth Information
to describe what it is in computational (or any other) terms. The limitation is my
own, not the program's.
Evaluation of a work in progress is directed to how to proceed. Evaluation of a
finished work is directed to whether it's any good. The procedures required to satisfy
the two are likely to be quite different, even when the same aesthetic is informing
the procedure in each case.
I think it's very unlikely that “in-line” evaluation
3
can be done algorithmically.
The simplest case I can think of would be to determine whether the work is finished.
Even that is much harder than one might think. It could only be done algorithmically
if one could provide an evaluation function—highly unlikely—which would be, in
any case, a shifting target with respect to the many different goals a program (or
human artist) may have.
For the general case the problem is much more difficult. The program can't de-
termine how to proceed unless it knows what it has done, and knowing what it has
done—the object so far—involves the notoriously troublesome “new term” prob-
lem. It knows it has done
a
,
b
and
c
, but can't know that it has introduced a novel
and unanticipated relationship between
a
and
c
. Which is exactly what should be
the determinant to the next step.
It is true, of course, that many human artists proceed algorithmically—you do
this, then you do that, and after you've done all the thises and thats you have an
artwork. No evaluation is required; your job is simply to do all the steps well. In hu-
man terms this algorithmic approach results in what we call “academic art”, which
I think has no place in a discussion on creativity.
Post-hoc evaluation is no less troublesome, and I suspect it's likely to be impos-
sible for a program that didn't itself make the artwork. A formal colour evaluation
4
that doesn't take account of the possibility that all the well-balanced colour har-
monies may add up to a portrait of an oddly-dressed man making rude hand ges-
tures, for example. It also implies that there are canons for colour distribution and
the rest, and evaluation simply measures conformance to those canons. (Impression-
ism good, German Expressionism bad?)
For most artists, making art is an on-going affair, not a series of isolated (art-
work) events. Consequently, the completion of each work provides an extension of
the feedback-driven consideration operating in the in-line evaluation. It is similarly
concerned with direction, not aesthetics (except to the degree that, for the completed
work, the artist must decide at least in part on aesthetic grounds, whether to accept
it or reject it).
That is quite different from the aesthetic evaluation by any other agent, who is
not engaged in that on-going process. In this case direction is clearly not an issue,
acceptance/rejection is not an issue, and the aesthetic principles brought to bear on
the work are unlikely to have much correspondence to those of the artist.
3
By “in-line” Cohen is referring to evaluation of aesthetic decisions as a work proceeds.
4
At the time of writing this statement, Cohen was very focused on translating his theories of colour
and colour harmony into algorithms that AARON could use to colour abstract shapes.
