The Debate Over Intelligence (and Stupidity) in AI’s Large Language Models

This 90-minute Virtual ACtioN Discussion will confront the storm of interest generated by recent large language models including ChatGPT. What are they, how are they, and why are they so smart and so stupid at the same time? We shall organize the discussion around a paper by SFI President and William H. Miller Professor of Complex Systems, David Krakauer, and SFI Professor and Science Board Co-chair, Melanie Mitchell, on the heated debate in the current headlines and in the AI research community on whether large pre-trained language models, such as ChatGPT, can be said to understand language—and the physical and social situations language encodes—in any humanlike sense.

The meeting will begin with Professors Krakauer and Mitchell exploring the arguments that have been made for and against such understanding, and key questions that have arisen in light of these arguments. Then, two SFI Postdoctoral Fellows who work closely with Krakauer and Mitchell, Maell Cullen and Arseny Moskvichev, will present on 'Scaling and Timelines'. We will conclude with a practitioner discussion on the practical challenges and potentials of widespread integration of these diverse forms of technology and cognition.