How AI describes GLTR

Collaborative Semantic Inference (CSI) is a framework for co-designing the interactions and models in deep learning to enable visual collaboration between humans and algorithms. It exposes the intermediate reasoning of models, enabling semantic interactions with visual metaphors so users can understand and control parts of the model’s reasoning. The approach is demonstrated with a co-designed case study of a document summarization system and is associated with the VizHub platform (vizhub.ai) for visual AI collaboration.