Jennifer Bradshaw and Amanda Van Dort (PeacePays.AI); Laylo Merali (DCAF)
As artificial intelligence (AI) becomes increasingly used across the security sector, including in policing, justice, border management, intelligence and the military, understanding and addressing bias in AI systems is becoming an important governance challenge.
This guidebook looks at how bias can enter AI systems and persist throughout their life cycle, from the initial definition of a problem and the collection of data to system development, procurement, deployment and monitoring.
It explains the different forms bias can take and how AI systems can reproduce and amplify existing inequalities, with potentially serious consequences for security decisions affecting people's rights, safety and access to protection.
Moving beyond the technical aspects of AI, the guidebook examines how institutional practices, governance arrangements and the way systems are used can contribute to these risks. It considers what this means for key principles of security sector governance, including accountability, transparency, the rule of law, participation, responsiveness and effectiveness.
It also highlights the relevance of the Women, Peace and Security (WPS) agenda in understanding and addressing the different ways AI-enabled systems can affect women and other groups.
Aimed at security sector leaders, policymakers, oversight bodies and practitioners, the guidebook provides practical guidance, checklists and tools to help identify, assess and mitigate bias at different stages of the AI life cycle. It emphasises that addressing AI bias is not simply a matter of improving algorithms, but an ongoing institutional responsibility.
Security institutions must retain responsibility for how AI systems are designed, procured, deployed and used, ensure appropriate oversight, and remain accountable for the decisions and outcomes they produce.
Ultimately, the guidebook seeks to help security institutions address the risks of AI bias while ensuring that the use of these technologies supports effective, accountable and inclusive security governance.
Dr Rumman Chowdhury (Humane Intelligence PBC); Dr Darja Schildknecht (Swiss Armed Forces); Nathalie Gendre, Dr Dawn Lui and Matteo Polizzi (DCAF); Katrina Fotovat (PeacePays.AI)