Unveiling the "algorithmic veil"： reflection on building an algorithmic interpretation framework
(School of Law, Tongji University, Shanghai 200092, China）
A proper solution of the problem on the explainability of algorithm is an important legal issue in algorithm governance, limited by the expanding "explanation gap" between algorithm users and audiences. However, at this stage, the dilemma of the explainability of algorithm exists in the process of algorithm operation, decisionmaking formation and application, which is embodied in three aspects: imbalance of data identification, insufficient proof basis and generalization of damage results. Combined with the differences in interpretation requirements, explanation methods and criteria of different application fields, building an algorithm interpretation framework with the help of systematic thinking may become a breakthrough to solve the problem on the explainability of algorithm. Taking the object as the logical jumpingoff point, the explainability method is divided into three modes: directional notification, public disclosure and administrative reporting. This method can be applied to medical, information recommendation, finance and other fields based on the concept of "scene justice", and the degree of explainability and criteria are distinguished for different businesses and scenarios, so as to realize the explainability of algorithm.
Key words :
algorithm governance; automated decisionmaking; explainability; criteria for explainability