Closing the AI-Loop: A Review of Human-Guided Machine Learning Approaches

Abstract

The integration of human feedback into AI models (Human-in-the-Loop, HITL) represents a central research field that is gaining increasing importance. While classical AI approaches primarily rely on historical data, HITL enables the incorporation of expert knowledge and user feedback into the training and decision-making process. This paper systematically examines the methods of Active Learning, Interactive Machine Learning, Reinforcement Learning, and Contextual Bandits. The aim is to highlight their respective strengths and weaknesses, to identify practical fields of application, and to discuss key challenges. Finally, an outlook on future developments in the field of human-centered AI systems is provided.

Keywords: Human-in-the-loop, Artificial Intelligence, Machine Learning

DOI: 10.54941/ahfe1007707