Provocation | Neural Networks Do Not Paint Black Squares

In the early 20^th century, mathematicians Emile Borel and Francesco Cantelli formulated and proved a theorem in Probability Theory, according to which any event with any positive probability of occurrence (i.e., that has a chance – even the slightest – to happen), repeated infinitely, would eventually happen. In absolute certainty. The more familiar and popular version of this mathematical concept, states that if we put a thousand monkeys to a thousand typewriters (can also be just the one), they will eventually produce all of Shakespeare’s texts, the Bible, the base code of the Internet, and even the coalition agreement between Bibi Netanyahu and Benny Gantz.

There was another relevant revolutionary event that took place in that same period of time; the Russian artist, Kazimir Malevich, painted one of the most significant works of art of the 20^th century – a black square.

Black Square, by Kazimir Malevich, 1915

Both of these works formulate well, in their unique abstract, minimalist and precise language, the difference between human and artificial intelligence.

Every other day we hear about a new achievement by artificial intelligence in every possible field: medicine, research, transportation, the stock market, playing Chess and Go and nowadays, of course, art as well. Algorithms now paint portraits, write sonnets and scripts, compose music and what not. That being said, the CoVid-19 crisis will not be solved by artificial intelligence. Why? Because artificial intelligence relies on mountains and mountains of data, collected over time and categorized by human intelligence. I.e.: artificial intelligence knows how to process the past, provided sufficient quality data was collected about it. A common misconception is that artificial intelligence can predict future processes. But how will it predict future process if it does not have past information about them? This could be thought of as “retro-futurism”: the future, as it was predicted in the past, for example, going to the opera in Paris via flying steam engines.

La Sortie de l’Opéra en l’An 2000 (Exiting the Opera in the Year 2000), Painting by Albert Robida, 1882

Another misconception about artificial intelligence is the creativity attributed to it. Van Gogh and Rembrandt-style paintings, musical works that complement Schubert and resurrect Bach. Did we really manage to bring Rembrandt back from the dead? In practice, the machine (a neural network) learns the painter’s common brush movements, color palettes, compositions and objects of painting, and applies them in a “new” painting. But would this “new” painting that leans on a simulation of 17^th-century techniques, technology, and culture, still be considered a pioneering masterpiece today?

The power of the digital machine is in its ability to repeat the same simple action countless times, not getting bored, not deviating, and at very high speeds. It was the blend of these skills that enabled such a machine to defeat the then world champion in Chess, Gary Kasparov. Deep Blue, the artificial intelligence machine that defeated Kasparov is not a “genius”, but a bully – a computational bully. In chess there are about 10¹²⁰ possible moves. A fast enough machine makes the relevant calculations for every possible situation of the board and plays the move that would yield the highest score.

But what happens when the laws are not well-defined? Or when there is an option to change the rules throughout the game?

This takes us to Malevich’s black and iconoclastic square. This painting, which recently celebrated its 100^th anniversary, is still considered revolutionary and relevant today as it was a century ago. Malevich himself describes in his writings his radical act as another (and perhaps the ultimate) step in the ongoing relentless process of corrupting the image and continually breaking the traditional laws of painting, from the days of ancient Greece, through the work of Cezanne, Cubism, Futurism, and of course, Suprematism. Malevich’s work and perception of his work is the result of studying the past and applying the results in the present (the present being 1915). Is there any artificial intelligence today that can emit such output? Is there a sophisticated neural network today with revolutionary tendencies that, after learning all the ground-breaking and iconoclastic styles from the mid-18^th century to the mid-20^th century, would produce an output of a black square, and on the way also explain – why?

The portrait of Edmond de Belamy was produced in 2018 by a deep learning algorithm. It was made famous as the first piece generated by artificial intelligence sold at the Christie’s arts auction house, for a whopping 432,000 US$.

Let’s get back to our monkey and replace his creative writing skills with painting. The painting monkey, much like Deep Blue, will paint, in the infinite time at its disposal, all possible paintings, among them there’ll also be – “Black Square”. But will there be anyone who will recognize the greatness of that painting and charge this new “Black Square” with the interpretation and meaning that Malevich gave his “Black Square”, placing it within a broader context of art, culture and society?

We, humans, are transient. The quantity of our output is limited, we can play a limited number of moves in Chess, paint a limited number of paintings, compose a limited number of musical pieces. This is way we have to be precise and focused in choosing our next move on the board and the next painting we paint. Should the next work we paint abide by the existing rules? Should it slightly change them? Or rather is it time to break those rules and offer a new model? Or maybe it’s even time to abandon the existing medium and create a new medium, with new rules?

Malevich did not have any illusions, not about processes that took place in 1915, nor about processes that were to take place in his future. Art, according to Malevich, is subject to the influence of present forces just as much as any other field. Artificial intelligence will indeed change the world, as it will change art, but not by creating an infinite number of Rembrandt-like paintings.

Understanding and designing the new rules and concepts of this new tool is a task for us, mere humans. Our time is fleeting and thus we don’t reconstruct, reproduce or recreate – we create.