Why Perfect Algorithms Can’t Create Masterpieces (But Flawed Human Beings Can)

A mere 347 years after the artist’s death, a new Rembrandt painting is suddenly in the news.  The subject of this painting is a man in his thirties.  He has dark, heavy-lidded eyes, slightly scruffy facial hair, and the pale, white-pink skin characteristic of sun-deprived Amsterdammers.  His garb is quintessentially Baroque – a dark shirt topped by a lace ruff at the neck that seems just a bit too tight.  The lighting, brush strokes, and ambiance all exemplify the style of Rembrandt van Rijn, the greatest of the Dutch Masters. 

The Next Rembrandt, a computer-generated painting imitating the style of Rembrandt van Rijn

The Next Rembrandt, a computer-generated painting imitating the style of Rembrandt van Rijn

Except it isn’t a Rembrandt.  It isn’t even truly a painting.  Created by a team of data analysts, developers, and art historians using a 3-d printer, the “painting” is a high-tech forgery.  The man in the painting is not a historical person but the product of algorithms.  The space between his eyes is not the space between any one person’s eyes but the average space between all the eyes of all the Rembrandt paintings the creators analyzed.  As such, the man in the painting looks distantly related to many of the other men Rembrandt painted. 

Detail of The Next Rembrandt

Detail of The Next Rembrandt

The mindset of those who created the painting is similar to that of the scientists who genetically engineered salmon.  The sense in which they are successful is obvious; they have created a product with the exact combination of qualities they desired.  In the case of the Rembrandt forgery, it is a combination of lighting, color, texture, and facial features.  In the case of the salmon, a combination of fast growth and delicious flavor.  In both cases, the new products are artificially similar but essentially different from the original products.  Once it is wrapped up into a sushi roll for consumption, no one knows if the salmon ever had to swim upstream or if he lived a short life in an industrial fish farm.  Similarly, once the fake Rembrandt is made into a pixelated image and distributed online, it is hard to say whether the artist ever sat in front of a blank canvas and envisioned what he would create.  All that we notice, at least at first, is the fact that these products have the right set of characteristics.  They pass.

One of the cheeky criticisms made of modern art is, “My kid could have made that.”  At the very least, people often look at modern works and say, “I could have made that.”  With the advent of these technologies, perhaps people will soon say the same thing about all works of art.  The day may come when a team of data-loving entrepreneurs enters the Sistine Chapel, looks up at the frescoed ceiling, and says to each other, “We can make that.”

Michelangelo, The Sistine Chapel Ceiling, 1508-1512, Fresco

Michelangelo, The Sistine Chapel Ceiling, 1508-1512, Fresco

If that day does come, will it be a great day for the visual arts?  Would a fake Michelangelo have the same power as the Sistine Chapel?  Would a computer-generated Van Gogh dazzle us in the same way that The Starry Night does?

In answering this question, it is worth analyzing how The Starry Night fits in with the rest of Van Gogh’s work.  The Starry Night is not a typical Van Gogh; it is an experimental Van Gogh.  He made the piece while he was in the asylum at Saint-Remy and suffering from mental illness.  Some of his other pieces had been painted from direct observation, but this piece was clearly a combination of observation and imagination.  The turbulence of his emotions was as significant a part of the subject matter as the actual scene depicted. 

With its coils of brushstrokes speeding through the sky, fireball stars, writhing Cyprus tree, and dots of warm light sprinkled throughout the cold, blue city below, The Starry Night is not an average Van Gogh predicted by an algorithm but a soaring Van Gogh in which he surpassed even himself.  The same could be said of many other masterworks.  They succeed not because they are just like everything else that artist made but because they transcend the limits the artist frequently bumped up against. 

Vincent Van Gogh, The Starry Night, 1889, Oil on canvas

Vincent Van Gogh, The Starry Night, 1889, Oil on canvas

There is a sense in which aberrations, exceptions, and changes, are actually the most interesting parts of an artist’s oeuvre.  Over the course of Rembrandt’s life, he went from being a young, brash, technically gifted portraitist to being an old man, broken down by the difficulties of life but capable of using loose brushstrokes to manifest his piercing insights into human nature.  The forces that sculpted him into a different person were painful ones.  His first three children died before reaching adulthood, and his wife Saskia also died young.  In addition, Rembrandt became bankrupt because of his excessive spending habits.  He was still well regarded, but his life was not what it once had been. The gravitas and heavy paint application of his later work were something he grew into over the course of his life, not something that any data analyst could have predicted.  These later works are some of his best and, far from lacking the technical finesse of his earlier works, benefit from the thicker, more sculptural application of paint.

Rembrandt van Rijn, Self-Portrait, 1629, Oil on canvas

Rembrandt van Rijn, Self-Portrait, 1629, Oil on canvas

Rembrandt van Rijn, Self-Portrait, 1669, Oil on canvas

Rembrandt van Rijn, Self-Portrait, 1669, Oil on canvas

This new fake Rembrandt is a fun example of what can be done with technology.  If I have the chance to see it, I will.  But it isn’t a masterpiece in the way that so many of Rembrandt’s actual paintings are.  The fake painting has all the right pieces, but they don't come together into something that is more than the sum of their parts.  For that, we need not a perfect algorithm but a flawed human being.