Using Human-Aware AI as a Framework to Achieve Intent Alignment

In this paper, we discuss the notion of intentalignment, a special case of the larger value-alignment problem. This version of the value-alignment problem is particularly relevant in scenarios where the human is instructing the robot to carry out a specific task. Under the intent-alignment problem, we would want the robot to follow the human instructions in such a way that it satisfies the human's true and possibly hidden intent behind the instruction. We will see how human-aware AI as a modeling framework allows us to not only capture the problem but also provide us with tools that will enable us to develop methods for intent alignment. We will also look at some existing intent alignment methods and discuss some open problems related to intent alignment.