Audit of algorithms

Audit of algorithms

Rapid advances in machine learning algorithms have led to their increased use for a wide variety of applications in in organizations. More recently though, concerns have been raised about algorithms being applied without careful thought about potential biases being introduced or perpetuated by algorithms as well as fairness concerns and other ethical concerns regarding adverse consequences for some stakeholders. This implies that there is a need to better control the development and use of algorithms. Drawing on the financial auditing field, we present a comprehensive framework for the audit of algorithms. This assertion-based framework consists of 18 assertions in four categories: stakeholder accountability, control, adequacy of training data and algorithm development & impact and we provide an initial validation the framework with a group of highly experienced auditors. This framework offers a systematic process for a thoughtful application of machine learning algorithms that keeps the drawbacks are under control.

ECDA will organize several round tables sessions with the big four accountancy firms in the Netherlands to evaluate the framework and develop a common approach on audits of algorithms.