Design of Automated Filtering to Approach Efficient Error Avoidance

State-of-the-art automated filters cannot perfectly emulate the analysis necessary to evaluate whether an upload attempt infringes copyright law. Existing discussions on how to design filtering mostly neglect some facets and rarely apply an efficiency perspective. This chapter conducts the first comprehensive analysis of why filters produce errors and what design choices best approach optimal error avoidance. For this, it applies the error cost framework, which states that decisions under uncertainty should strive to minimise the sum of direct costs of decision-making and expected net costs of errors. The analysis finds that a central reference database should be established, right holders should be responsible for preventing deficiencies of the database, and the database should comprise features to reduce right holders’ costs of providing information. For other key questions, conclusive answers cannot yet be given. Platforms should thus experiment with different approaches, track and disclose probabilities of errors, and disclose filtering designs. The resulting insights would also reduce uncertainties that cause inefficient levels of divergence between national approaches to filtering. Furthermore, the analysis points to possible uses of the information collected by filters that go beyond blocking, namely, notifying right holders of potentially infringing uploads and improving selection into the manual review.