What would be Hilbert’s problems of AI?

Michael Felsberg, Fredrik Heintz, Fredrik D. Johansson, Danica Kragic, Fredrik Lindsten, Amy Loutfi, Alexandre Proutiere, Serge Belongie, Virginia Dignum, Fredrik Kahl, Kathlén Kohn, Bernhard Schölkopf

Abstract: Hilbert’s 23 problems in mathematics published in 1900 proved to be very influential for 20th-century mathematics. The attempt to solve these problems did not only consolidate the scientific field, but also led to new areas of research. Consolidation and structuring is definitely required in the field of AI, which has been growing in an untamed way during the past dozen years. Much of this research has been driven by large companies, which even when supporting the basic research are primarily interested in solving problems from application domains rather than addressing the fundamental questions related to the advancement of a scientific field. Those questions have to be addressed by publicly funded basic research that is focused on contributions to the scientific field that might enable long-term benefits of their applications. This continuously evolving white paper addresses the goal of starting an initiative that will formulate the program equivalent to Hilbert’s problems for the field of AI. It will be open to public comments, forming the basis for future revisions and extensions. Also, a public website with the status of the problems is available. The long-term impact of the initiative will be an urgently needed complement to the predominantly short-term results produced by applied research.