From Data to Decisions: Assessing the Feasibility and Rationality of Human-in-the-Loop for AI Value Alignment

The Human-in-the-Loop (HITL) is a very important approach in machine learning, it tries to integrate human suggestions/judgements with the results from machine learning, and help it to achieve balance between efficiency and quality in data processing. It should be used in the whole process of R&D of AI, especially in the moral AI. This paper highlights the application of the HITL for AI value alignment. We analyze the technical feasibility and rationality of HITL in AI alignment, particularly in the context of Autonomous Vehicles (AVs), and discuss its potential future applications.