Representative Robots: Can AI Systems Act in Our Name?

Using AI systems to make decisions in the place of humans promises greater efficiency, but some authors raise a number of ethical worries about this. The undermining of responsibility, the removal of humans from meaningful participation, and a misalignment of values, it has been claimed, may result from a reliance on algorithmic decision-making. Most ways of avoiding these problems that have been proposed involve somehow ensuring that humans have direct control over the AI systems in question. But there is reason to think that there are limits to what this strategy can achieve. In this paper, I propose an alternative strategy. Building AI systems in a way that allows them to act in the name of humans, I suggest, would allow us to avoid the problems without the need for direct control. Drawing on recent work in democratic theory, I then claim that it is possible, in principle, for AI systems to act in our name. Achieving the benefits of AI without the costs that are often associated with their use may depend on our ability to create AI systems that function as our representatives.