Algorithmic Transparency: Exploration and Reflection from Theory to Practice

算法透明度:从理论到实践的探索与反思

Abstract

Although more and more important tasks and decisions have been entrusted to the algorithm, the computational complexity and opacity of the algorithm make it difficult for users to understand the decision-making process and results of the algorithm, resulting in their difficulty in trusting the algorithm, and even the phenomenon of "algorithmic aversion". Accordingly, algorithmic transparency is often seen as the foundation of trustworthy artificial intelligence and has received considerable attention in academic debates over the past few years. However, at the practical level, there are many challenges in implementing algorithmic transparency, which may even trigger certain ethical risks. Based on this, this study analyzes the challenges and risks at the practical level of algorithmic transparency, and points out that at least three dimensions of disclosure, review, and design can be used to solve the current practical difficulties.