Towards Model-informed Precision Dosing with Expert-in-the-loop Machine Learning

Human oversight improves both accuracy and fairness: The paper demonstrates that incorporating expert feedback into machine learning models can help "remove algorithmic bias by introducing experts' feedback into the iterative model learning process," suggesting that human-in-the-loop approaches are more effective at catching and correcting unfair outcomes than fully automated systems.

Interpretability matters for high-stakes decisions: In the precision dosing application, the framework produces "interpretable rules from data" rather than black-box predictions, which is critical when ML systems inform medical decisions that directly affect patient safety and outcomes.

Expert involvement can reduce workload, not increase it: Rather than burdening experts with endless data annotation, the approach "may potentially lower experts' workload by replacing data annotation with rule representation editing," making responsible AI practices more practical and sustainable in real-world settings.

Human-in-the-loop is necessary given current AI limitations: The authors note that "ML models are not fully comprehensible yet" and therefore "we still need humans to be part of algorithmic decision-making processes," emphasizing that responsible AI requires acknowledging the current boundaries of what machines can do alone.

This approach addresses data scarcity challenges: The framework is designed to handle situations where "the cost of data annotation is high and the lack of appropriate data" exists, making responsible AI development feasible even when resources are limited.