Looking at Art in New Ways – How AI Is Critiquing, and Creating Art
The Mona Lisa by Leonardo da Vinci – one of the world’s most acclaimed pieces of art. The public queues for hours at the Musée du Louvre in Paris to catch a glimpse of this much studied portrait. But, according to AI, it’s not much to look at really. At least that is the outcome from a project being undertaken at the Art and Artificial Intelligence Lab at Rutgers University.
The team at Rutgers has been using AI to analyze and create art for the past five years, studying around 80,000 different paintings by over 1,100 artists. One of the first outcomes from the research was the replication of known painting styles. A special type of AI, known as a Generative Adversarial Network (GAN), was used to create new pieces of art in styles such as Baroque, Pointillism, Rococo, and Abstract Expressionism.
The system is split into two distinct parts. The “suggester” of the GAN is responsible for the creation of the artwork, essentially suggesting an image in the desired style. The “discriminator” then provides critique on the piece, making a distinction in favor of or against the artwork on the basis of that agreed style. The two sides of the GAN can be thought of as entering into a discussion on how the painting should be improved.
Besides creating art, the AI can also work as an electronic art historian. Provided with background information on when paintings were created, it is able to deliver its verdict on the creativity of the piece. If a painting is dissimilar to previous artworks, but similar to pieces that appeared after it, it is considered novel and influential. The AI for instance, considers the Mona Lisa to be too similar to works that appeared before it.
Similarities between works of art created decades apart can also be found by the AI. For example, Norman Rockwell’s ‘Shuffleton’s Barbershop’, from 1950, features a group of men playing music together in the back room of the barbershop. To the right, a wood stove provides warmth, while a chair takes a rest from that day’s customers. This is very similar in composition to Frédéric Bazille’s ‘Bazille’s Studio’, painted 70 years earlier in the 1880’s.
Of course, AI isn’t expected to replace art historians. But its capability can be used by them to analyze a range of artistic aspects across a broader range of time and style than they could without it.
So, could AI instead replace artists? The team at Rutgers applied a technique called Creative Adversarial Networks (CAN) designed to generate artwork that does not fit to known artistic styles. The AI-created pictures were reviewed by the general public, hidden in-between other Abstract Expressionist work and non-figurative pictures on view at Art Basel.
Around half of the respondents believed the CAN-created pictures were drawn by human artists. But, perhaps more surprisingly, they considered these images to be more intentional, better visually constructed, and both more inspiring and communicative than human-created artworks.
Without the ability to put brush to canvas itself, we won’t see AI replacing human artists yet. But like all technologies, it may find its space. Perhaps the artwork in offices or hotels you next visit will be created, or inspired by AI.