An old idea for mechanizing creativity is to divide it into two parts:
one is an exploratory part, which is highly random and throws out a lot
of "crazy ideas". In the visual arts, maybe it tries lots of different
combinations of colors and shapes. The other part is the judgement part,
which looks at the output of the first part and decides what looks good.
In some models this could then feed back and narrow the search space of
the first part.
To say that computers can't mechanize creativity, you either have to say
that this method won't work to produce creative artworks, or else that
computers can't do the first or second parts. The most questionable is
probably whether computers can ever be able to have esthetic judgement,
to be able to look at different artworks and say which ones people will
find attractive.
It's not clear when this will be possible, but making a computer a critic
sounds easier than making it an artist. Actually the two are not as
different as they seem.
Hal