“AI models will probably be able to extract abstract concepts from video data alone, without associated text, in the future.”