Does black-box machine learning shift the US fair use doctrine?

Abstract

Machine learning (ML) has been widely used in not only daily life (eg online platforms) but also in professional fields (eg medical care, astronomy). However, it is noteworthy that most ML confronts a common Black-box Problem,1 which is deemed as one of the great policy issues with many ML.2 In Bathaee’s words, the Black-box Problem is defined as ‘an inability to fully understand an AI’s decision-making process and the inability to predict the AI’s decisions or outputs.’3 From a computer scientists’ standpoint, the Black-box ‘is an algorithm that takes data and turns it into something’ and often ‘detects patterns without being able to explain their methodology’.4 Put differently, the black-box decision model produces output without explaining why, because neither the stakeholders nor expert data scientist is capable of understanding the model.5 Therefore, the two main traits of the black box problem are unpredictability and...