Online Platforms: From Transparency to Observability

Abstract

One of the defining characteristics of (large) online platforms is the heavy reliance on algorithmic techniques (machine learning, network algorithms, etc.) for tasks like search, recommendation, and matching. These techniques pose a number of problems to both researchers and regulators interested in understanding how these platforms function and how they take part in shaping outcomes. Concepts like "fairness" and, in particular, "transparency" have been put forward to address the issue, but they run into both conceptual and practical limitations. This talk discusses these limitations and puts forward the notion of "observability" as a possible solution. While the notion of transparency hopes to "nail down" the internal mechanisms of an algorithmic system, observability focuses on the outputs of processes understood as distribted socio-technical achievments rather than purely mechanical procedures. The talk concludes by discussing a series of regulatory directions that would facilitate the concrete realization of conditions of observability.