Investigating AI/ML Models that Learn the Preferences and Constraints of Human Decision-Makers to Develop Interactive or Tools that Suggest Optimized Solutions While Allowing for Human Override and Incorporating Expert Judgment

Abstract

The growing adoption of Artificial Intelligence and Machine Learning (AI/ML) systems in very complicated and sensitive decision-making areas have had a big impact on the issues of transparency, trust, and ethical accountability. The optimization-based AI systems were mainly focusing on maximizing the computational efficiency and performance which led to less acceptance, as human preferences, contextual constraints, and domain-specific expert judgment were just barely being considered in the design of the system. As a consequence, the study investigated AI/ML models that learned and adapted to the preferences and constraints of the human decision-makers in order to create interactive decision-support tools that produced optimized recommendations while still allowing human override. A mixed-method conceptual–empirical approach was proposed that combined preference learning methods, constraint-aware optimization models, explainable AI components, and expert-in-the-loop mechanisms in order to foster collaborative decision-making. The results obtained from the study showed that human-centered AI systems considerably improved decision quality, user trust, and system adaptivity, especially in high-stakes fields where ethical issues, accountability, and contextual judgment were of utmost importance for effective decision-making.