ACM Technology Policy Council Releases Statement on Principles for Responsible Algorithmic Systems
November 1, 2022
The ACM Technology Policy Council (TPC) has released a new Statement on Principles for Responsible Algorithmic Systems authored jointly by its US and Europe Technology Policy Committees. Recognizing that algorithmic systems are increasingly used by governments and companies to make or recommend decisions that have far-reaching effects on individuals, organizations, and society, the Statement lays out nine instrumental principles intended to foster fair, accurate, and beneficial algorithmic decision-making.
The nine instrumental principles include: Legitimacy and Competency; Minimizing Harm; Security and Privacy; Transparency; Interpretability and Explainability; Maintainability; Contestability and Auditability; Accountability and Responsibility; and Limiting Environmental Impacts. In addition to the nine instrumental principles, the TPC recognizes that, despite putting in place a framework to discourage bias in algorithms, several factors could produce unfair systems. For these reasons, the Statement includes four recommendations which take into account the way data is processed and how a system is deployed:
- System builders and operators should adhere to the same standards in selecting inputs or architecting systems to which humans are held when making equivalent decisions
- AI system developers should undertake extensive impact assessments prior to the deployment of AI systems
- Policy makers should mandate that audit trails be used to achieve higher standards of accuracy, transparency, and fairness
- Operators of AI systems should be held responsible for their decisions regardless of whether algorithmic tools are used
The TPC also recognizes that a “one size fits all” approach to achieving responsible algorithmic systems would be ineffective and that context plays an important role in each developer’s decisions, and hopes that the Statement will prompt discussions among all stakeholders, initiate more research, and help leaders develop governance methods to bring benefits to a wide range of users while promoting the reliability, safety, and responsibility of algorithmic systems.