Design of a Brain-Computer Interface to translate emotional states into original paintings

Can Artificial Intelligence (AI) make Art? Researchers and artists are posing this question. In the last decade, the development of AI technologies has created genuine opportunities for artists, who are venturing into this world as experimenters or, perhaps, as pioneers. Like other innovations in the artistic field, the introduction of AI-related technologies is subject to a variety of criticisms and debates; among them, the legitimate doubt regarding whether an emotionless entity can make an effective contribution in a field in which the emotional sphere has a central role. In this regard, our thesis work has the objective to propose an AI system that generates paintings expressing human emotions. We designed a Brain-Computer Interface in which brain waves associated with different emotions are recorded in electroencephalographic (EEG) signals. The system relies on recent and state-of-the-art deep learning technologies, and it is divided into two main components. The first component is devoted to the automatic recognition of emotions in the EEG signals through a graph neural network, and it is trained on both a public EEG dataset and on signals we recorded with a commercial device. The second component is devoted to translating such emotions into original paintings, utilizing a generative adversarial network. The resulting paintings represent emotional states relying not only on stylistic features but also on expressive content and shapes. They represent, therefore, a promising opportunity for the employment of this technology in artistic applications. The AI system proposed and tested in this work is not only conceived as a mere technical tool, but also a creative actor. Through its interaction with a human, it succeeds in capturing and expressing the power and complexity of our emotional sphere.