Mardi intime de la Chaire Hoover par Thomas Ferretti (University of Greenwich)

Algorithmic fairness refers to principles aimed at ensuring that decisions made by algorithms treat individuals impartially, avoiding various forms of discrimination, particularly in areas such as the labour market and recruitment. In the literature on algorithmic fairness (e.g., Cynthia Dwork, 2018), ‘individual fairness’ is often defined as meaning that ‘similar individuals should be treated similarly’, which is equivalent to a principle of fair equality of opportunity (Rawls, 2001). In a recruitment process, for example, people with similar talents should have equal chances of getting the job. In contrast, “group fairness” focuses on criteria such as statistical parity between demographic groups, to ensure that outcomes (e.g., the people hired) reflect the composition of the applicants or the population. In this article, I develop two arguments. First, I give reasons for rejecting group fairness and preferring conceptions of individual fairness such as fair equality of opportunity. Second, I highlight difficulties in achieving individual algorithmic fairness within a recruitment process. I propose that these difficulties reveal a flawed conception of how to achieve fair equality of opportunity in society, and I suggest a framework for resolving this.