The complementary nature of human and algorithmic intelligence points to the need for an interdisciplinary approach that draws not only on computer science and statistics, but also on such fields as psychology, behavioral economics, and human-centered design. More effective and societally acceptable technologies will result if we shift focus from building “smart” machines to designing collaboration systems that enable forms of human-computer collective intelligence. Taking on board the implications of dual process psychology (System 2 “thinking slow” and System 1 “thinking fast” cognition) is a natural starting point for working towards “human-centered AI”. Regarding System 2, a key observation is that algorithms often excel at tasks that are difficult for humans (such as estimating probabilities or optimally weighing evidence); while struggle with things that come naturally to humans (such as understanding context, or using common sense). Regarding System 1, behavioral science teaches us that prompting smarter choices and decisions often involves more than providing information or setting up incentives. Often the manner in information is presented or choices arranged has surprisingly large effects on end-user behavior. This set of ideas, underutilized in AI, enables more effective and ethical forms of AI. Jim will discuss a variety of AI examples to illustrate the ideas.
James Guszcza is a 2020-21 fellow at Stanford’s Center for Advanced Study in the Behavioral Sciences. Jim has worked as data scientist for two decades and is the first person to be designated Deloitte’s U.S. Chief Data Scientist, is a former professor at the University of Wisconsin-Madison business school, and holds a PhD in philosophy from The University of Chicago. He serves on the scientific advisory board of the Psychology of Technology Institute. Several of Jim’s articles can be found here: https://www2.deloitte.com/us/en/insights/authors/g/jim-guszcza.html