Theorem: General Intelligence Entails Creativity, Assuming - Computational Creativity Research: Towards Creative Machines - page 51

which cannot be understood from, let alone derived from, antecedents e.g., see [ 5 ]. 1
In the case of Leibniz, Zeno's paradoxes of motion stood iron-strong for century
upon century, and then suddenly new infinitary concepts arrive on the scene, and
soon thereafter ordinary physical motion makes perfect mathematical sense. At any
rate, whether or not Hutter is right, the fact remains that while his Universal Artificial
Intelligence is a certified research-grade proposal for what general intelligence, in
man or machine, is, the topic's Index contains no entry for creativity; and therefore
at the very least we have no reason to think, on the basis of Hutter's topic, that what
I'm declaring to be a theorem is one.
Let's try a third tack. In keeping with so-called Psychometric AI (PAI, rhymes
with ' '; [ 3 , 6 , 7 ]), according to which AI consists in the engineering of artificial

agents capable of high performance on well-defined tests of various vaunted mental
powers in the human sphere, we can quickly see that, once again, general intelligence
doesn't seem to entail creativity: Let a be an agent able to perfectly answer every
question on every established, psychometrically validated test of general (human)
intelligence. 2 And now pull off the shelf every single established test of creativity
used by psychometricians and psychologists. 3 Next, does our assumption of a 's
prowess enable us to deduce that a will score at a high level on the selected test of
creativity? No. Indeed, the negative here is so obvious, and so firm, that I will not
trouble the reader with any details, and will instead sum up the psychometric chasm
between tests of general human intelligence and tests of human creativity by giving
this telling, representative fact: It doesn't follow from the proposition that some agent
is able to achieve perfection at digit recall 4 that that agent can quickly invent new
things to do with tin cans. 5

ˀ

3.2 The Setup

Despite the foregoing, in point of fact it is possible to show that high general
intelligence, whether of the human, alien, or machine variety, does entail creativity—
as long as certain assumptions are made. I find these assumptions to be eminently
reasonable, but lay no claim that they in fact are: Readers are invited to judge for them-
selves. As this “theorem” is a conditional, a natural way to construct the proof of it is to
assume, with respect to arbitrary instances of all the categories over which quantifiers

1 This point should not be interpreted so as to deny the brute fact that contemporary AI, viewed
pluralistically, is agent-based, where the agents often have goals/aims, and attempt to reach them
by straightforward processing of antecedents. Confirmation is provided by the widely-used [ 16 ].
2 Two ideal pickswould beRaven's ProgressiveMatrices [ 15 ], and theWAIS discussed in connection
with AI in [ 7 ].
3 One good choice would be the Torrance Tests of Creative Thinking (TTCT; [ 19 ]),targetedinthe
AI work reported in [ 4 ].
4 Digit recall is a sub-test on the WAIS (see note 2). On this sub-test, the test-taker attempts to
repeat back a string of digits given to him by the tester.
5 A typical question on TTCT. See note 3.