Reframing AI Loss of Control: What It Is, How to Have It, How to Lose It

The paper presents a framework for understanding the concept of control in the context of AI, defining it as the "setting and getting of goals" and drawing on principles from cybernetics and control theory. It explores how AI can contribute to the loss of control and outlines necessary conditions for maintaining control, emphasizing that significant risks exist even with AI systems that do not possess superintelligence. This work is relevant for practitioners as it provides a structured approach to assessing and managing control risks associated with AI deployment.