How can we design AI systems that not only learn from human expertise but also empower human experts to learn from AI? In this talk, I share my recent work on Reciprocal Human–Machine Learning (RHML), which details how humans and algorithms can iteratively refine one another's knowledge, using text-classification tasks as a concrete example. I then show how human oversight can be preserved and strengthened throughout the AI lifecycle by focusing on continual human learning. By illustrating how humans can remain "in the loop and in control," I highlight both the technical and ethical design considerations that ensure automated decision-making remains aligned with broader organizational and societal values.
