Hadi Hosseini

Associate Professor of Informatics and Intelligent Systems
Associate Professor of Economics (by courtesy)
President of IFAAMAS
Penn State University
Affiliate faculty: CS Theory Group · Rock Ethics Institute

Westgate Building, University Park, PA 16802

AI now makes decisions for and by a society of agents. My research asks which principles should govern those decisions when preferences conflict.

We combine the foundations of collective decision making: principled methods for aggregating diverse preferences (social choice), reasoning about strategic agents (game theory), and designing mechanisms that incentivize desirable behavior (mechanism design). We test these foundations in experiments with people, evaluate how AI agents reason and interact (AI behavioral science), and ask whether their decisions align with human values and ethical considerations (normative AI and moral philosophy).

NSF CAREER Award · Associate Editor, JAIR and JAAMAS · Associate Director, Penn State CAFE · Bio and honors

Principlesaxioms, theories, algorithms

Humansdiverse priorities and moral values

AI agentsinteractions, strategies, incentives

Research Focus

AI Alignment: aligning AI with diverse and conflicting human preferences, including at inference time, by modeling alignment as a game.
Normative Alignment: whether AI decisions and moral judgments reflect human values, where there is no single right answer.
Multiagent Coordination: what emerges when AI agents coordinate and compete, with each other and with people.
Fair Division and Matching Markets: what fairness and stability can be guaranteed when dividing resources or matching agents, and when they provably cannot.

I direct the Foundations of Artificial Intelligence Research (FAIR) Lab and am always looking for exceptional students. Please read this before emailing.

Research Highlights

Normative Alignment

Do AI systems share our values?

In organ allocation and resource division, language models diverge from people: they rarely hesitate, set aside responsibility, and fail to reduce inequality.

human subjectsfairness axiomsmoral judgment

The Judgment–Consequence Gap: LLM Moral Reasoning in Healthcare DecisionsAIES 2026
Who Gets the Kidney? Human-AI Alignment, Indecision, and Moral ValuesFAccT 2026
Distributive Fairness in Large Language ModelsNeurIPS 2025

AI Alignment

Can one model serve people who want different things?

We align models at inference time, without retraining, using Nash equilibria and maximin objectives from social choice.

Nash equilibriumgeneral preferencesmaximin

Maximin Inference-Time Alignment with Diverse RewardsPreprint
Inference-Time Nash AlignmentNeurIPS 2026

Multiagent Coordination

What happens when AI agents coordinate?

In games played by both people and LLMs, models converge on similar choices and struggle to stay diverse, even when diversity pays.

game theoryalgorithmic monocultureexperimental economics

Fair Division: Online Algorithms and Graph Structures

How can we ensure fairness in online or graph settings?

We introduce the first fairness models for groups in online matching and gig workers sharing delivery routes, design algorithms that guarantee fairness, and investigate when it conflicts with efficiency.

online algorithmsclass fairnessgraphs