The Ministry of Defence (MOD) has released JSP 936 Part 1, this policy is aimed at guiding the safe and responsible adoption of Artificial Intelligence (AI) within Defence and outlines the necessary steps and considerations for integrating AI technologies, while ensuring ethical and effective implementation. 

Here are the key takeaways:

Overview

• JSP 936 Part 1 provides guidelines for the safe and responsible implementation of AI technologies in Defence, ensuring productivity gains and maintaining military edge whilst addressing ethical and legal concerns. 
• The policy covers AI applications across Defence, including robotic and Autonomous systems (RAS) and digital systems, emphasizing the need for robust governance and ethical considerations.

Key Points:

AI Ethical Principles

• Human-Centricity: Assess and consider the impact of AI on humans, ensuring positive effects outweigh negatives.
• Responsibility: Establish clear human accountability for AI outcomes, maintaining meaningful human control.
• Understanding: Ensure relevant individuals understand AI systems and their outputs.
• Bias and Harm Mitigation: Proactively address and mitigate unintended biases and harms.
• Reliability: Ensure AI systems are reliable, robust, and secure.

Governance and Accountability:

• Roles and Responsibilities: Defence organizations must appoint Responsible AI Senior Officers (RAISOs) to oversee AI ethical assurance and accountability.
• Ethical Risk Assessment: Conduct AI ethical risk assessments to determine approval levels and manage risks throughout the AI lifecycle.
• International Collaboration: Promote interoperability and coordination with NATO and allies, ensuring shared ethical standards.

Human/AI Teaming:

• Human-Centred Design: Apply human factors principles to optimize system performance and user adoption.
• Training: Provide appropriate training for users to understand AI behaviour, performance, and limitations.

AI Lifecycle Management:

• Planning: Develop comprehensive plans addressing algorithm development, data management, and lifecycle transitions.
• Requirements: Clearly define performance and functional requirements, including hazard analysis and data needs.
• Architecture: Ensure AI Architecture is traceable to requirements whilst protecting against failures identified in hazard analysis.
• Algorithm Design/ Implementation: Choices must be justified and documented, and their output checked. 
• Model Development: Ensure the risks of the use, re-use and further development are understood. 
• Verification and Validation: Demonstrate AI performance across operational contexts, including boundary conditions and realistic edge cases.
• Integration and Use: Ensure AI systems perform reliably within their operational design domain (ODD) and monitor performance continuously.

Quality, Safety, and Security:

• Quality Assurance: Apply MOD quality policies, ensuring data integrity and compliance with ethical requirements.
• Safety Management: Integrate AI-specific safety risks into broader safety cases and risk assessments.
• Security: Adopt a Secure by Design approach, addressing unique AI security risks and adversarial threats.

Supplier Management:

• Competence and Assurance: Ensure suppliers demonstrate competence and provide assurance for AI products, including data provenance and compliance with MOD policies.

Conclusion: JSP 936 Part 1 emphasizes the importance of ethical, responsible, and reliable AI implementation in Defence, with a focus on human-centricity, accountability, and continuous improvement. It provides a comprehensive framework for managing AI risks and ensuring the safe and effective use of AI technologies in military applications. Part 2 - Guidance will provide guidance and best practice to assist users to comply with Part 1. 

If you would like to discuss AI and Defence, please contact Lucy Owens or Louise Dean. This article was written by Louise Dean.

If you would like to discuss how current or future regulations impact what you do with AI, please contact Tom Whittaker, Brian Wong, Lucy Pegler, Martin Cook, Liz Smith or any other member in our Technology team.