IDyOT: A Computational Theory of Creativity as Everyday Reasoning from Learned Information - Computational Creativity Research: Towards Creative Machines - page 140

stream, both of which might be considered restricted creative behaviours. In the
higher levels of the IDyOT representation, the same process allows semantic predic-
tions, inferring what the speaker intends, and potentially looking ahead to anticipate
the entire discourse. There is every reason to believe that this is a realistic simulation
of humans, who jump to conclusions, correct or otherwise, all the time.
Such a predictive mechanism necessarily begs the question: what happens when
there is no input to match with? In this circumstance, there is no existing sequence to
drive prediction, so it is driven directly from the memory, and the Global Workspace
may be thought of as an empty auditorium into which ideas may enter [ 55 ]. The
mechanism used above for chunking may also be used to select structures as they are
predicted, and it is important to remember here that there is no reason why prediction
cannot be performed at any level of representation that exists within IDyOT's mem-
ory. Thus, predictions conditioned only by the learned model, and not by current
input, can be made at any or all of the various levels from semantic context down to
phoneme, and the resulting distributions combined to make a single one [ 34 ], from
which a structure can be chosen by the throttle mechanism. Therefore, given the
necessary information-theoretic properties, such structures, which may or may not
be novel, may enter the Global Workspace without the need for external stimulus
[ 55 ], though, clearly, they will be conditioned by previous (learned) stimuli. As a
result of passage through the Global Workspace, the structures are added to memory,
and therefore become available for use in future generative inference. In such a way,
IDyOT can follow a cycle which is a simplified version of the process (probably)
described by Wolfgang Amadeus Mozart, characterising his own composition:
When I am, as it were, completely myself, entirely alone, and of good cheer—say traveling
in a carriage, or walking after a good meal, or during the night when I cannot sleep; it is on
such occasions that my ideas flow best and most abundantly. Whence and how they come,
I know not; nor can I force them. Those ideas that please me I retain in memory, and am
accustomed, as I have been told, to hum them to myself.
All this fires my soul, and provided I am not disturbed, my subject enlarges itself, becomes
methodized and defined, and the whole, though it be long, stands almost completed and
finished in my mind, so that I can survey it, like a fine picture or a beautiful statue, at a
glance. Nor do I hear in my imagination the parts successively, but I hear them, as it were,
all at once. What a delight this is I cannot tell! All this inventing, this producing takes place
in a pleasing lively dream. Still the actual hearing of the toutensemble is after all the best.
What has been thus produced I do not easily forget, and this is perhaps the best gift I have
my Divine Maker to thank for. [ 21 , p.317-318]
Some aspects of this description refer to what we defined in Sect. 7.3.3 as sponta-
neous creativity, where ideas appear unbidden in conscious awareness. Others are
deliberate: the selection and memorisation of pleasing elements. These latter are not
present in IDyOT, as a notion of “pleasingness” is currently absent, requiring as it
does a solution to the “hard” problem.
Therefore, our initial experiments with IDyOT as a creator will focus on the pro-
duction of sentences, melodies and concepts, based on a substantial body of learned
information, and selected by IDyOT's throttling mechanism. It remains to be seen
whether such a purely spontaneous system can create substantive, interesting struc-
tures, but early experiments with statistical generation in music have been promising