S3 Ep 10 Yejin Choi: Teaching AI Common Sense and Morality (Host: Pieter Abbeel)

Summary

In this podcast episode, host Pieter Abbeel interviews Yejin Choi, a noted expert in natural language processing (NLP) and AI research. Choi holds a Ph.D. from Cornell, is currently a professor at the University of Washington, and serves as Senior Research Director at the Allen Institute for Artificial Intelligence. She has received numerous awards and recognition for her work, including features in The New York Times, The New Yorker, TED Talks, and a MacArthur Fellowship.

During the conversation, they discuss the power and limitations of large language models. These models are trained on vast amounts of internet data, helping them learn language patterns and reasoning, but there are still challenges in developing models that exhibit intelligence and common sense capabilities. They explore the importance of reinforcement learning, supervised training, and human feedback in fine-tuning these models using pre-training and inference time algorithms, which focus on improving the quality of data rather than quantity.

Choi emphasizes the potential of AI-generated data to improve models, but also highlights the need to understand the limitations of AI algorithms and promote AI literacy to avoid misguided actions, especially in critical fields like medicine and legal aid. They also address the role of common sense knowledge in AI models, allowing language-based models to reason better and handle real-world moral and social norms more effectively.

The discussion moves on to Choi's perspective on language and reasoning, with her advocating for language as the best medium for teaching AI common sense and moral reasoning. They address both the challenges and excitement surrounding teaching machines common sense and moral reasoning skills. The potential applications of language models are explored, including multimodal applications that assist robotic planning and navigation, as well as education and remote medical assistance.

Finally, the conversation touches on Choi's personal background, her transition from a software engineering job to becoming a prominent NLP researcher, and her hobbies, which include hiking, mountain activities, and reading books on moral philosophy and cognitive science. Overall, this episode offers an engaging, in-depth look into the current state of NLP and AI research, and provides valuable insights into the possibilities and challenges of teaching AI common sense and moral reasoning.