Professor Yoav Artzi believes that training neural nets differently, such as teaching them to focus on causal explanations and encouraging joint learning between language and the world it references, can improve language and world modeling in artificial intelligence.