Optimizing Ethical Risk Reduction for Medical Intelligent Systems with Constraint Programming

TL;DR

MISRO addresses the challenge of reducing ethical risk in Medical Intelligent Systems under the EU AI Act by formulating a constrained optimization that jointly quantifies risks, maps them to trustworthy AI requirements, and enforces a reference criticality. Implemented in MiniZinc, MISRO supports three risk-quantification forms (linear, bilinear, quadratic) and is solved with CP, MIP, and SAT backends to compare performance, expressiveness, and scalability. The study shows that CP (notably Chuffed) achieves the best overall runtime and robustness, including higher-quality solutions under time pressure, especially for nonlinear risk formulations. The results suggest CP as a practical core of an automated, adaptive risk-management workflow for MIS, with future work to integrate interactive decision-support and EU Act-aligned governance. $Q^* = \arg\max_Q (\min_j Q_j),\quad C_{Q^*} \le C_{ref}$, and the framework enables dynamic constraint updates in real time as the MIS evolves.

Abstract

Medical Intelligent Systems (MIS) are increasingly integrated into healthcare workflows, offering significant benefits but also raising critical safety and ethical concerns. According to the European Union AI Act, most MIS will be classified as high-risk systems, requiring a formal risk management process to ensure compliance with the ethical requirements of trustworthy AI. In this context, we focus on risk reduction optimization problems, which aim to reduce risks with ethical considerations by finding the best balanced assignment of risk assessment values according to their coverage of trustworthy AI ethical requirements. We formalize this problem as a constrained optimization task and investigate three resolution paradigms: Mixed Integer Programming (MIP), Satisfiability (SAT), and Constraint Programming(CP).Our contributions include the mathematical formulation of this optimization problem, its modeling with the Minizinc constraint modeling language, and a comparative experimental study that analyzes the performance, expressiveness, and scalability of each approach to solving. From the identified limits of the methodology, we draw some perspectives of this work regarding the integration of the Minizinc model into a complete trustworthy AI ethical risk management process for MIS.

Figures (5)

• Figure 1: MIS Risk Management Process. Each lifecycle phase includes a risk evaluation cycle that updates risk quantification by solving the ${\tt MISRO\ }$problem.
• Figure 2: MiniZinc implementation of the unified ${\tt MISRO}_m$odel (${\tt misro.mzn}$).
• Figure 3: Average CPU solving times (in sec) for the 81 instances per mode: ${\tt MISRO}_1$, ${\tt MISRO}_2$, and ${\tt MISRO}_3$. Results for ${\tt Chuffed}$ (CP), ${\tt HiGHS}$ (MIP), and ${\tt PicatSAT}$ (SAT).
• Figure 4: Solver performances on 7 timeout-prone hard instances.
• Figure 5: Solver performances comparison across three quantification functions.

Theorems & Definitions (6)

• Definition 1: Medical Intelligent System
• Definition 2: Reference Criticality
• Definition 3: Risk Quantification
• Definition 4: Mitigation Action
• Definition 5: Risk–Ethical Requirement Matrix
• Definition 6: Calculated Risk-Based Criticality