Designing Paranoid Machines: Kenneth Colby and the Tensions Between Error and Intelligence in 20th-century AI
450 Jane Stanford Way, Building 160, Stanford, CA 94305
Room 433A
Join us for the next CESTA lunch seminar with Johan Fredrikzon on the tensions between error and intelligence in 20th-century AI.
Talk Abstract
Using these conflicting discourses as framing, this talk turns to the historical period between them and a series of experiments aimed at creating paranoid computers (essentially software with delusional traits). In focus here are the studies carried out by medical doctor, psychologist, and AI pioneer Kenneth Colby working at the Stanford Artificial Intelligence Lab (SAIL) in the 1960s and '70s. I will argue that Colby’s undertakings capture essential aspects of both traditions mentioned, raising questions about what constitutes an error in a machine made to simulate human behavior. In highlighting the entanglements between error and intelligence in the history of AI, the talk notes their relevance for our current sociotechnical landscape. This includes not only the employment of chatbots to act as therapists for human subjects but also the creation of AI systems themselves as objects worthy of analysis by “AI psychologists”—specialists in interpreting the otherwise inscrutable, alleged, “inner lives” of models.
About the Speaker
Johan Fredrikzon is a researcher at the Division of History of Science, Technology and Environment at KTH Royal Institute of Technology in Stockholm. He is currently a visiting scholar at Stanford. In 2022–2024, he was a visiting postdoctoral fellow at the University of California, Berkeley. Fredrikzon holds a Master of Computer Science from Stockholm University where he also received his PhD in the History of Ideas in 2021. In his research, Fredrikzon has been interested in problems of erasure, disappearance, waste, and decay as conditioned by processes of data management, office work, and archival practices. During 2018–19, he was a research affiliate at Yale University. Fredrikzon's current research project is a three-year study of the history of artificial intelligence (AI) from the perspective of errors and mistakes in humans and machines respectively, funded by the Swedish Research Council.