No solvency – it is impossible to verify human control
Horowitz, 2019 - Professor of Political Science, University of Pennsylvania [Michael C. May 2“When Speed Kills: Autonomous Weapon Systems, Deterrence, and Stability” https://ssrn.com/abstract=3348356 Acc 12/27/20 TA]
Jervis argues that arms races occur due to a security dilemma when states have the ability to measure each other’s capabilities, but not their intentions.57 LAWS could be especially risky in this regard because of potential opacity about capabilities, in addition to the “normal” opacity that exists about intentions. First, it will be extremely difficult for states to credibly demonstrate autonomous weapon capabilities. The difference between a remotely-piloted system and an autonomous system is software, not hardware, meaning verification that a given country is operating an autonomous system at all would be difficult. Second, uncertainty about the technological trajectory of machine learning and specific military applications means that countries might have significant uncertainty about other countries’ capabilities. Thus, countries might invest a lot in artificial intelligence applications to military systems due to fear of what others are developing.