General AI Systems and Interaction Design / GUI Semantic Modeling and Task Representation

Without manual annotation, can self-supervised learning generate effective semantic embeddings of GUI screens?