Artificial Intelligence and War

Defense experts must move beyond wanting to add artificial intelligence (AI) to military capabilities generally and toward a more particular understanding of its strengths and weaknesses. AI algorithms have demonstrated a host of generic errors, which generally arise from flaws in the training database, the mechanism of the algorithm itself, or the user’s interpretation of the algorithm result. Tragedy beckons if the Department of Defense (DOD) uses AI for warfighting but fails to recognize these potential problems. This report examines these issues and how the DOD should develop tools to measure and mitigate them, particularly in sensitive uses, to assure military effectiveness and compliance with the law of war. 

This report was made possible through the support of the Smith Richardson Foundation. The author is particularly grateful to Marin Strmecki and Chris Griffin.