The Ethics of the Algorithm: Autonomous Systems and the Wrapper of Human Control
48 CUMBERLAND LAW REVIEW 101 (2018)
34 Pages Posted: 20 Jul 2017 Last revised: 2 Jan 2018
Date Written: July 17, 2017
David Mindell notes in Our Robots, Ourselves, “For any apparently autonomous system, we can always find the wrapper of human control that makes it useful and returns meaningful data. In the words of a recent report by the Defense Science Board, ‘there are no fully autonomous systems just as there are no fully autonomous soldiers, sailors, airmen or Marines.’”
Designing and using “the wrapper of human control” means making moral decisions — decisions about what ought to happen. The point is not new as the “soldiers, sailors, airmen or Marines” references shows. What is new is the rise of predictive analytics, the process of using large data sets in order to make predictions.
Predictive analytics greatly exacerbates the long-standing problem about how to balance the benefits of data collection and analysis against the value of privacy, and its pervasive and its ever-increasing use of gives the tradeoff problems an urgency that can no longer be ignore. In tackling the tradeoff issues, it is not enough merely to address obviously invidious uses like a recent photo-editing app for photos of faces from a company called FaceApp. When users asked the app to increase the “hotness” of the photo, the app made skin tones lighter. We focus on widely accepted — or at least currently tolerated — uses of predictive analytics in credit rating, targeted advertising, navigation apps, search engine page ranking, and a variety of other areas. These uses yield considerable benefits, but they also impose significant costs through misclassification in the form of a large number of false positives and false negatives. Predictive analytics not only looks into your private life to construct its profiles of you, it often misrepresents who you are.
How should society respond? Our working assumption is that predictive analytics has significant benefits and should not be eliminated, and moreover, that it is now utterly politically infeasible to eliminate it. Thus, we propose making tradeoffs between the benefits and the costs by constraining the use and distribution of information. The constraints would have to apply across a wide range of complex situations. Is there an existing system that makes relevant tradeoffs by constraining the distribution and use of information across a highly varied range of contexts? Indeed, there is: informational norms. Informational norms are social norms that constrain not only the collection, but also the use and distribution of information. We focus on the use and distribution constraints. Those constraints establish an appropriately selective flow of information in a wide range of cases.
We contend they provide an essential “wrapper of human control” for predictive analytics. The obvious objection is that the relevant norms do not exist. Technological-driven economic, social, and political developments have far outpaced the slow evolution of norms. New norms will nonetheless evolve and existing norms will adapt to condone surveillance. Reasonable public policy requires controlling the evolution and adaption of norms to reach desirable outcomes.
Keywords: AI, artificial intelligence, predictive analytics, predictive models, robots, robot ethics, norms, informational norms, privacy tradeoffs
JEL Classification: Z18, C54, C55, K29
Suggested Citation: Suggested Citation