Are We In Control? Challenges to Governance of Autonomous Weapons Systems in the Age of Artificial Intelligence

Sammanfattning

The rise of AI is leading to paradigm shifts in every sector of modern society, including the military. One concerning change is the increase in the development, testing, and deployment of autonomous weapons systems (AWS), which utilise modern AI without adequate controlling regulation or governance. This thesis explores what it means to have control of AWS and whether it is possible to effectively govern AWS using our current normative framework of weapons governance. Using the theoretical framework of the Control Problem of AI alignment, I argue that it is impossible to ensure control over an autonomous AI system, and therefore, it is impossible to ensure meaningful human control of AWS. This problem is further compounded by the normative tradition of weapons governance, which has a shaky historical track record for governing new weapons technology during times of instability in world order, as norms in weapons governance tend to emerge through unregulated practice rather than through deliberation. I argue that due to the high risks of AWS slipping out of human control, AWS use cannot be governed and therefore should be banned preemptively.