Handle: Engineering Artificial Musical Creativity at the “Trickery” Level - Computational Creativity Research: Towards Creative Machines - page 279

(LT), which entails that what the machine does cannot be anticipated by the designer
of this machine [ 8 ]. On the engineering side, it's enough for the storytelling machine
to trick human readers, inTuring-testing-style, into believing that the stories produced
by this machine were produced by creative humans . The Handle project

is, like Brutus, based on a direct analogue of this two-part position: viz.,

( creativity T )

P1 Computing machines can't be genuinely creative in the musical sphere. This means that
no AI system or agent can be a genuinely creative conductor or composer. Nonethe-
less ...
P2 from an AI-engineering point of view, (a) it's enough to aim for a machine conductor or
composer able to trick human listeners, in Turing-testing-style, into believing that the
music produced/guided by this machine was produced by genuinely creative humans;
and (b) such a creative T conductor and/or composer can in fact be engineered within
the foreseeable future.

The work described herein is of course directly in line with P1 and P2, and is intended
to empirically demonstrate, eventually, the truth of P2 (b).
Our “middle-ground” approach to mechnical creativity differs radically from the
approach advanced by Cope, a longtime researcher of the first rank working in
the intersection of AI and musical creativity, who abides by a “lower” definition
of creativity. To confirm this, we need only turn to Cope's Computer Models of
Musical Creativity [ 12 ], where he tells us that for him creativity is merely “[t]he
initialization of connections between two or more multifaceted things, ideas, or
phenomena hitherto not otherwise considered actively connected” (Cope 2005, 11).
Immediately after giving this latitudinarian definition, Cope provides a series of
examples of his brand of creativity in action. His last example is the solving of the
following puzzle:

I have three sons whose ages I want you to ascertain from the following clues. Stop me when
you know their ages. One, the sum of their ages is thirteen. Two, the product of their ages is
the same as your age. Three, my oldest-in-years son weighs sixty-one pounds.
Stop, says the second man, I know their ages.
What are their ages?

Under the assumptions that: (i) the second man is an adult, and hence—in our
culture—at least 21 years of age; (ii) the second man couldn't deduce the answer
after the second clue; and (iii) the second man knows his own age, it's possible to
provide an outright proof that the correct answer is 2, 2, and 9. In an informal nutshell
here, the reasoning runs as follows: Of the permutations of three numbers n , m , and
k that sum to 13 and have a product that's at least 21, the only two that produce the
same product (36) are: 1, 6, 6 and 2, 2, 9. Since in the former case there is no oldest,
we are left with the latter as the only possibility. Since, using standard formalisms in
logic-based AI [ 6 ], we have engineered a machine able to find and certify a formal
proof of the argument just given, it's clear that a theorem-prover-based program able
to solve this problemwould not be creative G . The reason is that the designer of such a
computer program wouldn't be surprised in the least when a formal proof expressing
the argument is found. In addition, such a program wouldn't be creative T ,forthe