Universal Beauty of Abstract Art Predicted by Deep Neural Networks

Why do certain artworks captivate us all? Despite the subjective nature of aesthetic experience, certain visual patterns evoke remarkably consistent beauty judgments across individuals and cultures. Using crowdsourced behavioral data across Western and East Asian cultures, we show that aesthetic ratings of well-controlled abstract paintings – artwork without explicit semantic references – are highly stable across people. We hypothesize that this consistent aesthetic perception arises from computational processes linked with perceived beauty during feedforward visual processing. To test this, we used pre-trained deep neural networks (DNNs) to predict perceived beauty in these paintings. Our findings reveal that efficient coding, reflected in reduced network activation level, predicts perceived beauty and emerges early in the cascade of visual processing, which is later complemented by mid-level beauty-related pattern matching. Combined, these findings uncover key neural computations underlying aesthetic perception, suggesting that perceived beauty may be evolutionarily rooted in energy-efficient sensory processing.