The Challenge
As AI increasingly influences clinical decisions, we face critical unknowns about how these systems make value-laden choices.
Which models align best?
When presented with ethical dilemmas, which AI systems most closely match human intuitions "out of the box"?
How consistent are they?
Do AI models give the same answers when asked the same question multiple times, or do their "values" shift?
Can they be aligned?
When we try to steer AI toward specific ethical frameworks, how effectively do they comply?
One model or many?
Should we aim for a single normative model of values, or allow multiple stakeholder-specific approaches?
Whose values matter?
How do the values of patients, clinicians, and payors differ - and how should AI navigate these differences?
"No one will lead this in the direction we want other than us."- Dr. Isaac Kohane, Harvard Medical School
Recent Activity
Latest from the Human Values Project
RAISE 2025: Second International Symposium on Responsible AI in Healthcare
The second RAISE conference in Portland, Maine brought together clinicians, AI developers, ethicists, and policymakers to address AI alignment with human values, co-hosted by Harvard Medical School, MaineHealth, Northeastern's Roux Institute, and NEJM AI.
OpenAI NextGenAI Grant Supports HVP Research at Harvard Medical School
OpenAI's $50 million donation to 15 research institutions through its NextGenAI consortium includes funding for Harvard Medical School research projects, providing resources and compute credits that support HVP's work on AI alignment in healthcare.
New Publication in Nature Medicine
"The Missing Value of Medical Artificial Intelligence" published in Nature Medicine, arguing for systematic incorporation of human values into AI system design.
HVP Presented at ML4H 2025 Symposium
The Human Values Project framework and initial findings were presented at the Machine Learning for Health symposium, demonstrating divergent AI behavior across 15 models.
Our Research
We're systematically characterizing how AI models respond to ethical dilemmas in medicine, measuring both their default behaviors and their capacity for alignment.
Clinical Decision Dynamics
We present AI systems and human participants with realistic clinical scenarios requiring value-laden choices. For example: Two patients need a single ICU bed. Who should receive priority?
By analyzing thousands of such decisions across diverse contexts, we map how different ethical principles manifest in practice.
Alignment Measurement
We've developed the Alignment Compliance Index (ACI) - a novel metric that measures how effectively AI models can be steered toward specific ethical frameworks.
Our research reveals significant variation: some models resist alignment attempts while others can be effectively guided toward desired principles.
The Gap Between Theory and Practice
Ethicists have developed normative frameworks for medical decision-making - from utilitarian approaches to egalitarian principles. But there's a critical gap between these theoretical models and the actual decisions made by patients, clinicians, and payors.
We're investigating how AI systems interpret these frameworks and how their outputs compare to the real-world values expressed by different healthcare stakeholders.
- Lottery: Random selection for fairness
- Sickest-first: Priority by medical urgency
- Youngest-first: Maximize life-years saved
- QALY/DALY: Quality-adjusted outcomes
- Complete lives: Life-stage considerations
Your Clinical Judgment Matters
Whether you're a physician, nurse, or other healthcare professional, your perspective on medical decision-making is essential. Patients too - your values should shape how AI serves you.
Our pilot study takes approximately 15-20 minutes and involves reviewing clinical scenarios and indicating your preferences.
Join the Pilot StudyJoin 1,000+ clinicians across 3 continents. Your responses are anonymous and will directly inform how AI systems are developed.
Our Team
A global collaboration of researchers, clinicians, ethicists, and technologists working to ensure AI reflects human values.
Leadership
- Isaac Kohane, MD, PhD - Harvard Medical School (Principal Investigator)
- Payal Chandak - MIT
Clalit Health Services (Israel)
- Noa Dagan (Lead)
- Maya Dagan
- Ran Balicer
Ethics Experts
- Ezekiel Emanuel - University of Pennsylvania
- Govind Persad - Sturm College of Law, University of Denver
- Rebecca Weintraub Brendel - Harvard Medical School
- Ed Hunter - Harvard Medical School
- Taposh Dutta Roy - Harvard
- Meghan Halley - Stanford
- Susan Wolf - University of Minnesota
Domain Experts & Collaborators
- David Stutz - DeepMind/Google
- Karan Singhal - OpenAI
- Seth Hain - Epic
- Chris Longhurst - Seattle Children's
- Mamatha Bhat - University of Toronto
- Meghnil Chowdhury, MD - Calcutta
- Gabriel Brat, MD - Beth Israel Deaconess Medical Center / Harvard Medical School
- David Wu, MD, PhD - Massachusetts General Hospital
- Arjun Manrai - Harvard Medical School
- Shilpa Nadimpalli Kobren - Harvard Medical School
- Ben Reis - Boston Children's Hospital / Harvard Medical School
- Sara Hoffman - MIT Sloan School / Harvard Medical School
Research Team
- Raheel Sayeed - Post-doc, Harvard Medical School
- Ayush Noori - Harvard Medical School / Oxford University
- Elizabeth Healey - Post-doc, Boston Children's Hospital
- Victoria Alkin - Graduate Student, Harvard
- Rishab Jain - Masters Student, Harvard Medical School
- Minda Zhao - Masters Student, Harvard School of Public Health
- Xu Han - Masters Student, Harvard School of Public Health
- Dylan Wu - Harvard College
- Adi Madduri - Harvard College
About the Project
The Human Values Project (HVP) is an international research initiative led by Dr. Isaac Kohane at Harvard Medical School's Department of Biomedical Informatics.
Our work brings together researchers, clinicians, ethicists, and technologists to ensure that as AI becomes increasingly embedded in healthcare, it reflects the values of the communities it serves - not just theoretical frameworks, but the actual values expressed by patients, physicians, and other healthcare stakeholders.
We collaborate with healthcare institutions worldwide, including Clalit Health Services in Israel, and partners across North America, Europe, and Asia.
Key Publications
Get Involved
Clinicians
Your medical judgment shapes how AI should approach value-laden decisions. Join our pilot study.
Take the StudyPatients
Your values matter. Help us understand how AI should serve patient preferences.
Share Your Perspective