The U.S. Cyber Command Legal Conference in April 2026 highlighted critical legal and ethical questions regarding artificial intelligence (AI) employment in cyber warfare, particularly concerning "black box" un-explainability. A central concern is whether AI processes whose outcomes cannot be explained can be deemed discriminating, violating Additional Protocol I, Article 51(4) on indiscriminate attacks or the proportionality rule.
The article questions if testing alone suffices when AI methods remain unknown, potentially yielding correct answers for incorrect reasons. It also addresses the legal acceptability of multi-agent AI systems if their composite output is impenetrable. Commanders face implications for responsibility if they authorize missions using less than fully trustworthy AI. The author suggests trustworthiness is a matter of degree, necessitating national policy to define thresholds for military AI applications. Unlike human errors, AI mis-performance may be untraceable, hindering accountability and learning. Rigorous testing must replicate intended use and identify all factors causing AI mis-performance.
No comments:
Post a Comment