Some worry that as AI-systems become more capable and more integrated in society, we will lose control over those systems. The loss of control, some worry, may yield bad or even catastrophic outcomes. Some think that we can solve this problem in a way that avoids having to have control, simply by ensuring that the systems are value-aligned. Here is argued that value-alignment cannot solve the control problem, unless the value-alignment includes a notion of control. That is, any value-alignment must, in order to successfully avoid the control problem, provide humans with sufficient control.  

This presenter was invited by the ESDiT program.