During the conference ‘Opportunities and Risks of Digital Transformation in Finance and Beyond’ organised by the Centre Responsible Digitality ZEVEDI on 2-3 June 2022, Mireille Hildebrandt will discuss the issue of proxies that is key to any Recommender System but more generally to any AI system. Proxies map what they stand for but they are not the same thing. Proxies are appropriate depending on the specific purpose and may have far reaching negative consequences if recycled for other purposes. Fairness requires a computational proxy to be taken into account by a machine learning system - and even by a multi agent system or a straightforward automated decision system. In her talk Hildebrandt will discuss how recommender systems optimise for two very different types of preferences: those of users and those the providers. In the end this means that proxies will be chosen in a way that rewards preferences that are lucrative from the providers’ point of view. Without a keen eye for the political economy of these kinds of systems it will not be clear that and how the preferences that are inferred are simultaneously nudged. Such nudging is often based on the design of a particular choice architecture that determines what users can and cannot act upon. EU law aims to configure another choice architecture, namely that of the providers of these systems, restricting their choice to provide recommender systems that do not manipulate users into dedicated behaviours but genuinely empower them to follow their own mind. See the underlying paper.
For more information about the event and full programme, click here.
For slides, click here.