Military AI Risks and Opportunities Series – 4 of 4

Executive Summary of the Series

What Western Voters Want

Citizens in democracies do not want to turn over life and death decisions to AI. Lone Star’s polling shows some significant differences among nations, but broadly, there is resistance to AI controlling killing machines. The results are consistent across other polls, with attitudes on completely autonomous cars aligning well with Lone Star’s polls on military AI.

What Western voters want is roughly the same thing military and diplomatic professionals call “Meaningful Human Control.”

Meaningful Human Control and the Law of Armed Conflict (LOAC)

The Responsible AI in the Military Domain Summit (REAIM) was held in 2023 in the Hague. Several Western nations have embraced the principles of the summit, and in particular, the concept of “Meaningful Human Control.”

This concept does not ban turning over control of military action to computers. Rather, it suggests each nation must ensure humans control the limits of computer actions within the bounds of LOAC.

This concept avoids parsing what is AI and what is not AI. And broadly, it seems to align with the Western values of responsibility, accountability, and the rule of law.

However, history shows full and faithful compliance with treaties, and LOAC is rare. Since the adversaries of the West do not share our values, we should not expect others to comply with the AI norms we hope for. Further, since AI is not a physical thing which can be inspected like a missile silo or nuclear reactor, cheating will be difficult to detect.

The Military AI Risk Matrix is Complex

The third paper of this series defines distinct risks which help frame how nations can seek advantage over each other in the AI domain. It is not a complete list. Since we should expect both friends and foes to cheat on norms such as Meaningful Human Control, the number of potential advantages and risks across the family of nations is a vast game space.

Since most of the West’s paradigms about AI are based on consumer technology, these paradigms themselves are a risk to the extent the military needs differ substantially from selling advertising. At the same time, commercial markets in capitalist economies will drive innovation faster than military markets can hope to achieve. Military customers must adapt to commercial markets while understanding a minority of military needs are truly unique.

Where from Here?

The West has begun to build consensus about the control of AI. Military, policy, and diplomatic leaders should consider the lessons of history and the unique attributes of AI technology to define the policies which best serve their own national interests.