The Algorithmic Black Box
Ashley S. Deeks · 2025
Abstract Chapter 2 introduces basic concepts associated with AI, including machine learning and deep neural networks, and explains why AI tools are often considered “black boxes.” It considers why we are seeing a significant escalation in the use of these tools in national security settings, especially by the United States and China, and what advantages they provide over basic automation. The chapter identifies the most common critiques of AI, including a lack of transparency about how the systems work, a lack of accountability for their use, and the embedding of biases in algorithms due to flawed training data. The chapter then describes what we know about the current state of AI/ML tools at work within U.S. government agencies. The chapter argues that algorithmic black boxes, which are opaque both to the democratic public and to their governmental users, complicate our pursuit of the public law values of legality, competency, accountability, and justification.