AI Aligns Heart Transplant Policies with Human Values

Itai Zilberstein, Ioannis Anagnostides, Zachary W Sollie, Arman Kilic, Tuomas Sandholm· September 1, 2026 View original

Key takeaways

  • Directly eliciting preferences over outcomes improves AI alignment with human values.
  • A novel two-phase algorithm efficiently learns utility functions from stakeholder input.
  • Applying this method to heart transplant allocation significantly enhances policy alignment.
  • The approach offers a robust way to integrate ethical considerations into AI decision-making.

Who benefits

HealthcarePublic PolicyEthics & GovernanceSocial Services

Summary

This research introduces a novel preference elicitation algorithm that directly learns utility functions from stakeholder preferences over outcomes, rather than comparing algorithmic decisions. Applied to heart transplant allocation, the method optimizes policies to better align with human values, significantly outperforming the status quo.

This paper presents an innovative algorithm for eliciting human preferences to better align AI systems with societal values. Unlike traditional methods that ask stakeholders to compare specific algorithmic decisions, this approach directly gathers preferences over desired outcomes to construct a utility function for policy optimization. The algorithm operates in two phases: an initial phase uses pairwise comparisons to quickly narrow down the space of possible attribute weights, followed by a second phase that provably converges to the user's true utility function. The researchers applied this technique to the complex problem of heart transplant allocation, where policies must balance multiple competing objectives such as post-transplant success, waitlist mortality, geographic factors, and equity. A user study was conducted to aggregate a community-aligned utility function. Policies optimized using this new method achieved a near-optimal competitive ratio of 0.95, a substantial improvement over the status quo policy's ratio of 0.54, demonstrating a significantly better alignment with human values.

Why it matters

Professionals in healthcare, public policy, and AI ethics can use this method to design AI systems that make decisions in sensitive areas more ethically and effectively, reflecting community values.

How to implement this in your domain

  1. 1Adopt outcome-based preference elicitation for designing AI policies in critical domains.
  2. 2Utilize the two-phase algorithm to efficiently learn stakeholder utility functions.
  3. 3Conduct user studies to gather and aggregate community preferences for policy optimization.
  4. 4Apply the optimized policies to improve fairness and ethical alignment in decision-making systems.

Original post by Itai Zilberstein, Ioannis Anagnostides, Zachary W Sollie, Arman Kilic, Tuomas Sandholm

"arXiv:2608.28620v1 Announce Type: new Abstract: Preference elicitation is essential for aligning AI systems with human values. Prior approaches (e.g., for organ allocation) often ask stakeholders to compare the decisions of an algorithm (e.g., patient A vs. patient B). Such a dec…"

View on X

Originally posted by Itai Zilberstein, Ioannis Anagnostides, Zachary W Sollie, Arman Kilic, Tuomas Sandholm on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses