The AI Ethics Journey Will Hit New Heights in 2021

The Department of Defense adopted its Ethical Principles for Artificial Intelligence in February 2020, a first for any military organization. These principles build on the foundational work performed by the Defense Innovation Board and is tied directly to one of the pillars of the DoD AI Strategy: Leading in military ethics and safety. The Joint Artificial Intelligence Center serves as the Department’s lead for coordinating the oversight and implementation of these principles.

Alka Patel, head of AI Ethics Policy for the JAIC, focuses on how to operationalize the five DoD AI Ethics Principles (Responsible, Equitable, Traceable, Reliable and Governable) and put them into practice in the design, development, deployment, and use of AI-enabled capabilities. However, to operationalize these principles throughout the DoD, the JAIC is turning to Responsible AI – an enterprise-wide framework that provides the DoD workforce and the American public the confidence that DoD AI-enabled systems will be safe and reliable, and will adhere to ethical standards.

Getting this done is not a simple exercise of writing and checking a “to do” list. “When we think about operationalizing these principles, we must do it from the lens of a broader Responsible AI framework,” says Patel. “Many of the five principles are rooted in sound engineering practices across the AI product lifecycle. However, we must also look at operating structures and organizational culture, such as increasing ‘Responsible AI’ literacy across the workforce and governance structures for escalation and decisioning.”

The JAIC looks to operationalize Responsible AI across the DoD enterprise through four pillars: People, Processes, Partnerships, and Policy. In 2020, the JAIC initiated efforts on all four pillars. Just a few examples of success in these areas include:

• People. The JAIC launched a Responsible AI Champions pilot program, which convened 15 cross-functional individuals from across the JAIC through an experiential learning journey to understand the AI ethics principles, identify tactics for operationalization, and seed a network of Responsible AI ambassadors. AI ethics training was also incorporated into the broader DoD Workforce Education Strategy prepared by the JAIC, as well as delivered during a workforce education pilot targeting acquisition and product capability managers in October. 
• Process. The JAIC created a model card for traceability and documentation of a trained model that was utilized in Project Salus as a test case. The goal is to integrate the use of a model card into existing workflows for all projects. Additional efforts include developing tools such as a harms analysis and risk assessment, which can be used across the product lifecycle, and finding ways to build and integrate such tools and corresponding checkpoints into the architecture of the Joint Common Foundation. Integrating into existing workflows and development pipelines will aid in decreasing the barrier to adoption of new tools and accelerate scaling of Responsible AI across the Department.
• Partnerships. The JAIC has been working on modifying procurement processes to incorporate Responsible AI principles into its RFI/RFPs, as well as working directly with acquisition partners to co-develop Responsible AI strategies. Most recently, the JAIC held a workshop with the World Economic Forum to consider practices around integrating AI ethics into the procurement process. Additionally, the JAIC launched the AI Partnership for Defense with like-minded military and defense forces from an initial 13 nations. The partnership aims to promote the responsible design, development, and use of AI by engaging in coordination and collaboration on AI technologies, governance, and policy to drive AI-enabled interoperability.
• Policy. In March of 2020, the Ethics Policy team launched the DoD-wide Responsible AI Subcommittee, and this diverse group of approximately 50 individuals representing all major components of the DoD has been meeting monthly throughout the year on a variety of efforts associated with policy and governance processes.

Patel and her team are also working closely with various JAIC-led DoD AI Subcommittees and the JCF Prime Integrator as they develop, integrate, and automate tools and processes into existing workflows, which are critical to scaling and adopting Responsible AI. The efforts also include external outreach and collaboration with academia, industry, acquisition partners, and allies to ensure that best practices are incorporated in applied AI engineering and evaluation for interoperability of Responsible AI implementation of procured/shared technologies.

Patel and her team will continue to push forward in rolling out the ethics principles across both the JAIC and the services. One focus for Patel will be increasing Responsible AI literacy across the DoD, as well as providing centralized guidance on operationalizing each of the AI Ethics Principles. As an example, this spring, she will launch a Responsible AI Champions pilot within the broader DoD.

Additionally, the JAIC will be producing a Responsible AI Strategy and Implementation document. “Understanding has to occur across the Department before we can put out policy, at least if we want it to be meaningful and impactful,” Patel said. “So we are working towards creating a responsible AI strategy and implementation guide that can be flexible—updated as necessary—and we will provide centralized direction at an enterprise-wide level while allowing services and other areas the autonomy necessary to implement these practices.”

Patel noted that just like when building and deploying AI capabilities, the AI ethics effort has to start in the design phase and requires ongoing vigilance and assessments throughout the lifecycle, including continual learning and iteration through a feedback loop.

“Ethics is really everybody’s job—not just the technologists—and as with any relatively new space where there is a lot of learning and change taking place, we all need to vigilant and have a constant growth mindset around AI ethics,” she said. “We don’t have all the answers, but we are going to try things, most likely make mistakes, but learn from them. For this reason, ethics is a journey, not a destination. That said, I’m really proud that the DoD adopted these ethics principles in 2020, putting that stake in the ground which allows us to take our first steps on this journey, and I’m excited about what 2021 will bring as we accelerate along this path.”