What Can AI Tell Us About Fine Art?

Whether it’s the mysterious playfulness of Mona Lisa’s smile or the swirling soft colors of Monet’s painting, there are qualities of fine art that attract viewers like a moth to a flame. What in these works captivated people for centuries? Researchers are now using machine learning algorithms to tease these subtleties and explore the relationship between aesthetics, sentimental value, and visual art memorability.
Eva Cetinic is an art enthusiast and researcher at the Ruger Boskovic Institute in Croatia. Although she believes that art is indescribable in many ways, she wanted to challenge her own point of view by exploring how machine learning can quantify art. “The rise of artificial intelligence forces us to rethink what values are specifically human, and understanding art is a particularly fruitful platform for this kind of research,” she explains.
For starters, Cetinic and her colleagues analyzed over 100,000 images from WikiArt. Their results, published on June 5 at IEEE Access, hint at common themes of what we find beautiful and fascinating.
The researchers took several existing models that are appropriately trained to analyze the aesthetics, sentimental value, and memorability of photos, and modified the models to make them more applicable to fine art. They then compared the predictive estimates of the models with those obtained by people in other studies, in which participants evaluated works of art on factors such as aesthetic quality, beauty, color, content and composition. From there, the researchers selected a predictive model that best matched human preferences for each category.
By choosing models that accurately reflect human preferences, AI can uncover the intricacies that influence human judgment when it comes to art.
It is not surprising that the model chosen for the analysis of aesthetics, found that daring and intense picture is the most pleasant, while dull and dull picture less enjoyable. But the factors that make art more appealing to the AI’s eye, such as color harmony (that is, if colors match well) and brightness, are actually negatively correlated with the sentimental value of the image. So instead of color, the model found that a person’s emotional response to the work is more closely related to things like flowers and smiling people, while outdoor scenes or sad or scared faces are less sentimental.
Perhaps, reflecting human nature, models found nudity especially memorable. It is interesting that abstract images also turned out to be memorable, which, according to the authors, may be due to the lack of objects that we recognize. Since we rarely encounter visual stimuli observed in abstract paintings, an image can attract the viewer’s attention more than a picture containing an object with which we are familiar.
So how are all these factors related to each other? Apparently, beautiful paintings are not necessarily memorable. In fact, while there was a correlation in model predictions between what is considered aesthetically pleasing and what is considered sentimental, both factors were negatively correlated with the memorability of the picture. For example, abstract paintings are remembered, but have a low aesthetic value; on the contrary, landscapes are pleasing to the eye, but not remembered.
The researchers also analyzed the data of the artist and the era. Ironically, William Turner, the artist who created the most visually appealing pieces of work, also scored the lowest score in terms of memorability. Ah, Woe to be an artist.
Cetinic notes that the model for sentimental value is awarded to the only female artist in this subset, Frida Kahlo, with a significantly higher score than other artists. “This may be due to the fact that Kalo’s paintings often include traits that the model has learned to identify as very positive, such as color brightness and colors,” she says. However, an observer familiar with Kahlo’s work knows that there is actually a lot of pain in her paintings and that a superficial understanding of the feeling of individual motifs cannot understand the emotional complexity of her artistic expression. This is an example that shows the current limitations of this approach, but also points to new possible directions for future research.”

Post a Comment

0 Comments