A new working paper by Andrea Ferrario (Mobiliar Lab for Analytics at ETH, Chair of Technology Marketing at MTEC ETH), Michele Loi and Eleonora Viganò (UZH, Digital Societies Initiative) titled “Transparency As Design Publicity: Explaining and Justifying Inscrutable Algorithms” is available on SSRN.
The paper criticizes the modern approach to machine-learning model interpretability (with a focus on counterfactual explanations) and proposes a novel approach to algorithmic transparency denoted by “design publicity”. This is directly related to the justification of the use and design of the algorithms, and of each individual decision stemming from them.