Designing AI Systems That Obey Our Laws and Values
Communications of the ACM, Vol. 59.9, September 2016
3 Pages Posted: 27 Aug 2016
Date Written: August 25, 2016
More and more machines -- from driver-less cars to weapons to search engines -- are using AI as part of their operational system. Because these systems are semi-autonomous and learning on their own from experience -- they are likely to stray from their programmers' instructions. These cars may learn to speed; AI-equipped drones hit wrong targets; and search engines discriminate.
We suggest that to prevent such problems, society needs a whole new kind of AI programs, supervisory ones, that will monitor, audit, or correct the operational AI programs. In some cases they will ensure that laws are observed. In others, that the machines follow the values of their users, a special kind of AI program, the ethics bot, that will discern these values by observing the users of the machines.
Suggested Citation: Suggested Citation