Volume 287, October 2020, 103349
We propose a framework for incorporating public opinion into policy making in situations where values are in conflict. This framework advocates creating vignettes representing value choices, eliciting the public's opinion on these choices, and using machine learning to extract principles that can serve as succinct statements of the policies implied by these choices and rules to guide the behavior of autonomous systems.
From the Discussion
In the general case, we would strongly recommend input from experts (including ethicists, legal scholars, policymakers among others). Still, two facts remain: (1) views on life and death are emotionally driven, so it’s hard for people to accept some authority figure telling them how they should behave; (2) Even from an ethical perspective, it’s not always clear which view is the correct one. In such cases, when policy experts cannot reach a consensus, they may use citizens’ preferences as a tie-breaker. Doing so, helps reach a conclusive decision, it promotes values of democracy, it increases public acceptance of this technology (especially when it provides much better safety), and it promotes their sense of involvement and citizenship. On the other hand, a full dependence on public input would always have the possibility for tyranny of the majority, among other issues raised above. This is why our proposed method provides a suitable approach that combines the utilization of citizen’s input with the responsible oversight by experts.
In this paper, we propose a framework that can help resolve conflicting moral values. In so doing, we exploit two decades of research in the representation and abstraction of values from cases in the service of abstracting and representing the values expressed in crowd-sourced data to the end of informing public policy. As a results, the resolution of competing values is produced in two forms: one that can be implemented in autonomous systems to guide their behavior, and a human-readable representation (policy) of these rules. At the core of this framework, is the collection of data from the