AI, the Box, and the Black
https://lieber.westpoint.edu/ai-box-black/
Publish Date: 2026-06-01 14:30:00
Source Domain: lieber.westpoint.edu
- AI Discriminatory Nature: The article poses whether AI processes that are not explainable can be regarded as discriminatory under international law, particularly concerning the legality of AI-enabled attacks in terms of proportionality.
- Explainability Requirement: It questions if testing results are sufficient to justify AI deployment, highlighting concerns about systems that produce acceptable outcomes without the ability to explain their methods.
- Complex AI Systems: The article considers the impact of employing multiple inter-linked AI systems, questioning whether their combined output can be explained and whether this affects their legal acceptability.
- Trust Issues with Unexplainable AI: It explores whether decision-makers can trust and rely on AI systems whose operations are unexplainable, and what the role of trust is in assessing AI system lawfulness.
- Accountability in AI Use: It discusses the implications for accountability when AI systems are not fully trustworthy and suggests that accountability may hinge on understanding and explaining how AI processes work.
- Autonomous vs. Human Decision-Making: The article poses whether the fallibility of humans when making decisions should affect the deployment and trust in AI systems that support or replace human judgment.
- Autonomy Constraints: It raises the prospect of autonomous systems only being used if their processes are fully explainable, conflicting with the development of fully autonomous weapon capabilities.
- AI Testing and Weapon Review Requirements: The article examines the necessity for rigorous testing and measurement of trustworthiness in AI systems before deployment and the role of explainability in weapon reviews.
