Is the EU human rights legal framework able to cope with discriminatory AI?

Abstract

The challenges introduced by AI for the EU anti-discrimination legal framework have been a widely discussed topic among the doctrine. In the light of the 20th anniversary of the EU Charter of Fundamental Rights, the Commission released a regulatory proposal to tackle AI. This paper seeks to determine whether the proposal successfully addresses the existent pitfalls of the EU framework. First, this paper explores the functioning of AI systems that employ machine learning techniques and determines how discrimination takes place. Second, the article examines intellectual property rights as one of the main barriers for accountability and redressal of violations committed by an AI system. Third, the state of the discussion concerning the pitfalls of the existent EU approach towards non-discrimination is addressed. The available academic literature suggests that discriminatory outputs produced by an AI will amount to indirect discrimination in most scenarios. In this sense, cases of indirect proxy discrimination will likely pass the proportionality test, therefore justifying the discriminatory output. The last section of this article studies the Commission’s regulatory proposal. Although the document seems to effectively tackle discrimination caused by biased training data sets, this paper concludes that intellectual property rights and proxy discrimination still constitute significant barriers for the enforcement of anti-discrimination law.