Bridging near- and long-term concerns about AI

Abstract

Research and debate on the impacts of AI have often been divided into two sets of issues, associated with two seemingly separate communities of researchers. One relates to the short-term -- that is, immediate or imminent challenges, such as privacy and algorithmic bias. A second set of issues relates to longer-term concerns that are less certain, such as risks of AI developing broad superhuman capabilities. These two sets of issues are often seen as entirely disconnected. We argue that this perception of disconnect is a mistake. There are many connections between the shorter and longer-term issues, and researchers focused on one have good reasons to take seriously work done on the other. Long-termists should look to the short-term because research directions, policies, and collaborations developed on a range of issues now could significantly affect long-term outcomes. At the same time, short-termists could benefit from the long-termists’ big picture forecasting and contingencies work.