Algorithmic decision-making has spread throughout the public and private sector, and with it, growing concerns of unfairness. These concerns have led to calls for greater transparency, especially when the processes are generated by artificial intelligence and premised on personal information. Such calls have joined actual and proposed legal requirements.

A common approach to resolve transparency concerns calls for “breaking open the black box”; providing detailed information as to the algorithm’s inner workings. However, algorithm developers often strongly object to such measures, while arguing that they undermine their trade-secrets and compromise integrity by enabling gaming. These responses have led to considering auditing methods which examine the algorithm’s inputs and outputs to establish verifiable measures of accuracy and fairness. The panel will strive to establish which of these two strategies provides an optimal balance of competing equities and constraints in various contexts.

• What solutions is industry already providing, and are they sufficient?
• Can firms meet disclosure requirements sufficiently by providing counter-factual-based explanations?
• What are the limits and benefits of auditing methods analyzing the inputs and outputs of algorithmic processes? Does a shift to this form of evaluation reflect a broader structural change from relying on procedure to examining outcomes?
• Are there lessons to be learned from the managing of the COVID-19 crisis, as to how information about AI systems should be revealed?