An Assurance Case for the DoD Ethical Principles of Artificial Intelligence

The Ethical Principles of Artificial Intelligence (AI) [1] laid out by the Defense Innovation Board were one of the first publications from the Department of Defense to outline the expectations for AI enabled systems and technologies. This document served as the first guidance for developing agencies and was reviewed to understand the requirements for these new systems. As engineers, the desire is to view the Ethical Principles as evaluation criteria and identify the means by which a system can be qualified against the language laid out by the Defense Innovation Board. One of the first parallels that was identified with this document was the concept of an assurance case. The Ethical Principles do not explicitly lay out any requirements but are more so suggestions or guidelines so the question was how to demonstrate adherence or fulfillment. To this extent the Materiel Release process [2], the process the U.S. Army follows to deploy and field a system, was reviewed as a means to demonstrate fulfillment through the requirements and documentation dictated by that process. This paper demonstrates how the processes and procedures followed by the U.S. Army, in pursuit of mitigating risks for fielding, also in turn fulfill the intent of the Ethical Principles. Upon further review of the principles, it can be observed as design best practices to ensure the development of trusted and assured products. The Materiel Release process is proposed as an assurance case for the adherence to the Ethical Principles of AI. The MR process and associated feeder processes are here compared to the language embodied in the Ethical Principles to the extent that the application of the same rigorous processes done for traditional systems may be applied to AI enabled systems - and in some cases adapted - to ensure justified confidence in the delivered product. Systems that have gone through a Materiel Release review can thus also be said to have demonstrated, as a byproduct, adherence to the Ethical Principles of AI.