Implementing Ethical AI in National Defense: Challenges and Solutions
June 12, 2023
Artificial intelligence (AI) has become an integral part of our daily lives. Capable of making independent decisions at a rapid pace. However, concerns related to data bias, vulnerability, and explainability have highlighted the need for ethical AI implementations.
In response to these challenges, Northrop Grumman is collaborating with U.S. Government organizations to develop policies and guidelines to ensure the safety, security, and ethical use of AI models in the Department of Defense (DoD).
Implementing AI Ethics
The AI Principles Project
The DoD’s Defense Innovation Board (DIB) has established the AI Principles Project to address the ethical challenges associated with AI development for national defense.
These principles include responsibility, equity, traceability, reliability, and governability. In addition, AI software development should be auditable and robust against threats to operationalize these principles effectively.
The Operationalization Challenge
Dr. Bruce Swett, Chief AI Architect at Northrop Grumman, emphasizes the challenge of operationalizing AI ethics. It is critical to integrate ethical decisions into AI systems before oversights or flaws occur to prevent negative or catastrophic mission outcomes.
The nature of AI, constantly evolving and being upgraded, blurs the line between development and operations, making the task of developing secure and ethical AI complex.
The Fluid Environment and Multidisciplinary Approach
Updating AI models with new data to enhance performance can introduce biases, vulnerabilities, or instabilities that must be tested for safe and ethical use.
Dr. Amanda Muller, a technical fellow and systems engineer at Northrop Grumman, highlights the need for a multidisciplinary approach that considers technology, policy, and governance simultaneously.
Understanding the problem from multiple perspectives is essential in this constantly evolving environment.
The Integration of AI Security and Ethics
The challenges of secure and ethical AI design go beyond the traditional DevOps framework. Hostile actors not only expose AI implementations to learning experiences, but also when AI implementations go live.
Robustness against adversarial AI attacks becomes crucial for DoD applications, as attackers may possess their own AI tools and capabilities. Protecting AI data and models throughout the lifecycle, from development to deployment and sustainment, is critical in securing AI for DoD use.
Understanding Context and Human Involvement
AI excels at specific tasks but struggles with understanding context. While AI operates within its application, it lacks the ability to grasp the bigger picture. Contextual understanding often relies on human intelligence.
Dr. Vern Boyle, Vice President of Advanced Processing Solutions at Northrop Grumman, emphasizes the importance of involving humans in the system. He also emphasizes the configuring interactions to allow humans to contribute their unique capabilities.
Developing Justified Confidence
Dr. Swett raises a fundamental ethical question for AI developers:
Does an AI model meet DoD requirements, and how can justified confidence in its capabilities be developed?
The Department of Defense will provide its customers with verifiable proof that AI models and capabilities can be used safely and ethically in mission-critical applications. Through an integrated approach that includes policies, testing, and processes for AI governance.
Conclusion
As AI becomes increasingly pervasive in national defense, addressing the challenges of implementing ethical AI is paramount.
We can achieve secure and ethical AI, by operationalizing AI ethics, integrating AI security into the DevSecOps framework, understanding the complexities of context, and involving humans in the system.
Northrop Grumman’s collaboration with U.S. Government organizations exemplifies the commitment to developing auditable and trustworthy AI models for the future of defense applications.
