Algorithmic Transparency as a Solution to the Problem of Envy
Envy between citizens tends to undermine mutual commitment to cooperative political institutions, such as democratic elections, legislative rules, and judicial norms. In a recent paper, Harrison Frye argues (a) that envy is an inevitable consequence of the opacity of market processes governing the distribution of social goods, and (b) that the solution is to promote cultural norms against dividing people into classes and against the expression of envy. In this paper I extend (a) to distributions run by AI technologies. If envy is a problem for ordinary human-constituted decision making, it is equally a problem for automated decision making, and possibly more so. The good news is that unlike human decision making (which is necessarily opaque), algorithmic systems offer the hope of at least some transparency. I survey the current approaches to explainable AI, and evaluate them against the goal of reducing envy through transparency. This allows us to tackle the problem of envy at its source, and improves upon (b), because it's not clear how we could change cultural norms effectively enough to prevent envy from undermining the democratic project.
Ruling Out: Making Sense of "No Ought From Is" without Sentence Categories
In the last 30 years, discussion of the Humean "no ought from is" thesis has coalesced around the idea of proving theorems to the effect that normative sentences can never be properly inferred from descriptive ones. But each existing theorem comes with significant costs, and what they all have in common is that they need to cleanly sort sentences into different types. But what if sentence-level categories are the source of the problem? I explore the possibility that we can make sense of "no ought from is" just by talking about terms and remaining agnostic about sentences.
