CISA Seminar: Michael Rovatsos

“Superintelligence” is to machine learning as “Mind” is to operant conditioning: Discuss

The Dartmouth conference that established AI as a field of scientific research was held 60 years ago this year. By the late ‘sixties Donald Michie, Christopher Longuet-Higgins and colleagues had established Edinburgh as a major international centre, with a research programme covering all the areas of AI as it was then construed (and which is now called Artificial General Intelligence): perception and language, reasoning and decision-making, learning and logic, programming and chess playing, minds and robots … The contemporary public view seems to be that AI as a cognitive science has failed and public discussion is apparently focused on just two themes: machine learning (which business and investors love) and “Superintelligence” (which Hollywood and the global elites alternately love and fear). I have spent a career in cognitive science trying to contribute to the “Dartmouth-Edinburgh project” and apply it for human benefit, particularly in medicine. On some da!

 ys I despair that the AI community has slipped back 30 years. On others I think the field is maturing in exciting ways - with all the opportunities and risks that implies. In this session I will draw on experience applying AI to medicine in order to motivate some general questions and invite discussion: 1. "Are the cognitive sciences converging?" 2. "Could we build a general artificial intelligence within seven years?" 3. "What should a modern AI development platform support?" 4. “Can we make AI safe?”