Can a computer be creative?

Creativity is a wide field, but if we confine ourselves to creative creativity in the creation of art, I would say in today’s state of the art: No.The machines are now good in terms of craftsmanship, but they simply lack the intention.

Machine learning allows computers to create new ones from previous patterns, such as composing like Bach or turning a photo into an image like van Gogh’s.These are fascinating results that deceive even experts. But at the end of the day, it’s a form of imitation. But in turn.

The question of whether non-human things or animals can be (fundamentally) creative is old.The recent successes of Artificial Intelligence (AI) have given new impetus to the question of computer creativity. The press recently reported on images generated by an AI system that have generated significant sums at auctions.

Systems that produce such images today often consist of two components:

  1. A “generator” that generates potential new images based on previously fed real images.

This is typically done with artificial neural networks that have been trained for this purpose.

  • A “filter”, i.e. a second neural network trained to distinguish real human images from computer formations.
  • In the interplay of these components, plausible images are ultimately created by a generate-and-test method, i.e. roughly by trial and error with a built-in control body.

    This approach is reminiscent of the mathematical-philosophical theorem of the endlessly typing monkeys.This is available in different variants, in the 1930s it was roughly formulated in such a way that if one would let infinitely many monkeys indefinitely for a long time on mechanical typewriters, a monkey would type at some point in purely statistically, e.g. “Romeo and Juliet” and another, for example, the “boy in the bog”. Actually, a monkey that taps for an infinite amount of time is enough for this. It should be clear that the monkeys are not creative.

    Has the above-mentioned AI system now practically implemented the old dream of the typing monkeys?Within limits yes: The generator generates many potential images among which are also “real art”. And the filter is even able to select them from the “committee”. This goes beyond the thought game with the monkeys (here a lecturer would have to read the typed “works” for an infinite amount of time in order to pick out “Romeo and Juliet” and similar works of art).

    The really interesting question remains: Is creativity at play here?If so, it would have to be in the filter (or in the interaction of filter and generator). Since the filter only answers the question whether the result is indistinguishable from a human image and based on existing images, I would be more inclined to say that this is not creativity. In my way, creativity manifests itself not only in the result, but also in the way there and in it, for example with styles, expectations, cultural norms, etc. to play and thus create something new that appeals to, delights or provokes the consumer in some way. The famous bathtub of Beuys is a good example, the Pop Art a similar. This kind of creativity therefore requires a great deal of knowledge about art, society, cultural activity, current events and, above all, empathy. The filter simply does not have that.

    One could argue that the filter has implicitly learned all this from the “analysis” of the human images shown to it during training (compared to the computer-generated images or other “disturbing stimuli”).But my creativity doesn’t go so far as to believe that.

    In this interview, I also touch on the topic: Understanding Artificial Intelligence – HIIG

    Leave a Reply