Information Technology Reference
In-Depth Information
HC:
A report from the front. A couple of weeks ago I decided I wanted to see
more saturated colour in AARON's output. I gave the program what I thought would
be a suitable colour profile (controlling the frequency with which it would choose
one of the nine possible combinations of lightness and saturation) and then watched
in increasing frustration as the program generated several hundred rotten images.
Yesterday I bowed to what I've always known to be the unyielding dominance
of value—lightness—over saturation, and substituted a different colour profile that
generated colours from very light to very dark. And this morning I selected forty
stunning images: my “aesthetic evaluation”? from more than two hundred mostly
excellent images.
What was I looking for when I made the selection?
A sense of place. All the images make use of the same set of form generators;
I chose those images that transcended mere arrangement of forms, those that gen-
erated the sense that they represented something external to themselves, those that
seemed to carry the authenticity of the thing seen.
What contributes to this sense of place?
There are relatively few variables in the program that exercise critical control
over the nature and reading of the output. One is the choice of colour profile. Others
are the scale of forms relative to the size of the image; the proportions of the image;
the background colour (hue, lightness and saturation) relative to what builds in the
foreground; the proportion of space allocated to background and foreground; the
mode of distribution of the forms.
You'll see that these are all quantifiable. (There are several possible distribution
modes, each of which is controlled by quantifiables.)
Is the nature and quality of the output—the sense of place—then quantifiable?
I am aware that there are no intrinsically good or bad values for the variables
that control the output. The sense of place—and everything else—results from the
combination of all the variable values. That's a multidimensional space with perhaps
fifteen or twenty dimensions that I know about; way beyond my own mathematical
capabilities if I thought that was a good way to go. But notice that the same set
of values generated more than two hundred images, of which I judged only forty to
have an adequate sense of place. Evidently there are other elements involved beyond
the variable settings; specifically, I suspect, the “clustering” of forms which emerges
from distribution and scale and population and all the rest.
Is this emergent property—clustering—quantifiable? I doubt it.
The implication seems to be that a program might be able to pick out the good
ones, but couldn't pick out the exceptional ones; which are, of course, the ones
I'm interested in. But even this might be going too far, partly because it may not
be possible to identify the original variable values from the output, partly because
in doing so it would only have identified this particular work as belonging to a
particular group and would reject any work that didn't belong to this or another
successful group. Clearly, that's not the way to go. The transcendent images that
don't belong to any group are precisely the ones I want.
The more important point to make, however, since we appear to be talking about
aesthetic evaluation, is that I've not said a word to suggest that beauty is an issue
