Page 266 - The Black Box Book

This is a translation (including minor revisions) of an essay entitled "Computerkreativität.
Maschinelles Lernen und die Künste Künstlicher Intelligenzen", first published in: Dotzler,
B. J. & Karpat, B. (2021). Götzendämmerung. Kunst und Künstliche Intelligenz. Bielefeld.
THE SO-CALLED AI REVOLUTION
In the debates on artificial intelligence (AI), one objection against attri-
buting creativity to computers has been raised regularly for decades. It is
based on the claim that they can only execute what has been designed
and programmed by humans. However, already the General Problem
Solver, developed by Allen Newell, John C. Shaw, and Herbert A. Simon, in
the second half of the 1950s did not define every state of its system
a priori, but linked the occurrence of rules to the conditions of their appli-
cability (Newell et al., 1959). So-called machine learning methods, however,
go a considerable step further in this respect. Their central characteristic
is that they enable computers to master different tasks (automatically)
without in principle being explicitly programmed to do so (Samuel, 1959).
Among the existing approaches to machine learning, artificial neural net-
works (ANNs) have given the research field of AI a new upswing after many
years of stagnation. Whenever innovations in AI have been discussed in
the last ten years or so, whether with reference to self-driving cars, to
the prediction of stock market prices or to the field of medical diagnos-
tics, ANNs were always involved (Sudmann, 2018). And because of this
technology, computers were also able to achieve hitherto unprecedented
performance in the field of creative artistic work: AI systems, we read
everywhere, can now compose music in the style of Bach or Beethoven,
produce images as if they had been painted by Matisse and Monet, or
generate texts that, at least at certain points, reveal a literary quality that
seemed completely impossible just a few years ago.
In the art world and creative industries, reactions to the recent AI
boom have, as expected, been very mixed. For some, whether artists, jour-
nalists or scientists, the current manifestations of AI represent a source
of inspiration, an interesting tool, or a productive intellectual challenge,
for others, an outrageous exaggeration or even an existential threat.
Admittedly, such contrary assessments have a long tradition. Regardless of
the specific topics and fields of application, fundamental debates about
the potentials and limits of AI have always seemed to be particularly
polarizing. Against this background, the following considerations are also
informed by the question of how not only the developments and promises
of AI, but also its critical perception and discussion, especially in connec-
tion with problems of art and creativity, can be critically reflected upon.
266
AI CRE ATIVIT Y – SOME
H I S T O RI C A L RE M A RK S
A reliable starting point for such an intervention is to situate current phe-
nomena in their respective historical contexts. For example, it is worth
remembering that the founding document of AI research, the proposal for
the Dartmouth Summer Research Project on Artificial Intelligence, written
1955 by John McCarthy among others, already highlights “randomness
and creativity” as a relevant field of research on intelligent machines:
“A fairly attractive and yet clearly incomplete conjecture is that
the difference between creative thinking and unimaginative com-
petent thinking lies in the injection of some randomness. The ran-
domness must be guided by intuition to be efficient. In other words,
the educated guess or the hunch include controlled randomness in
otherwise orderly thinking.” (McCarthy et al., 1955, p. 2).
From today’s point of view, the description of the connection be-
tween creativity and randomness may not be merely obviously incomplete,
but under-complex. After all, the impression is given here that random-
ness is a kind of ingredient which, as a component of a recipe to produce
a creative machine, merely has to be provided in the right dosage and for
which, in the end, the most important thing is to ensure that it proves to
be effective under certain conditions (“guided by intuition”).
Moreover, one could probably add in the spirit of that logic that it
also satisfies the criterion of efficiency. After all, even notorious skeptics
of AI are willing to grant a computer so much creative power that, given
sufficient computing time and capacity, it would be able to combine a cor-
respondingly large number of words by means of random combinatorics
alone and without the implementation of linguistic rules in such a way that,
at some point, a poem of the quality of Paul Celan’s poetry could emerge.
At the end of the 1950s, it was the computer poetry of Theo Lutz
which – indeed with random combinations, but still decisively rule-based
– developed stochastic texts on the electronic large-scale ZUSE Z22,
a computer operated at that time by Technische Hochschule Stuttgart
(Lutz, 1959; see also Bernhart & Richter, 2021).
From the 1960s onwards, artists and scientists such as Alison
Knowles, Margaret Masterman, Max Bense, James Tenney and Robin
McKinnon Wood each generated poetry on computers in their own way
(Higgins & Kahn, 2012).
It is also not a new phenomenon that art produced by a computer
looks like human-made. One of the well-known examples is described
267