Tutorials

The following tutorials have been accepted for presentation at PRIMA 2026.

Tutorial 1: Consensus Formation with AI Agents: Discussion Support, Consensus Building, and Automated Negotiation

Instructors:
Takayuki Ito, Kyoto University
Shun Okuhara, Mie University

Keywords:
AI agents; consensus formation; automated negotiation; LLM agents; collective intelligence

Abstract:

How can autonomous AI agents help people reach agreements when preferences are linguistic, uncertain, and structurally complex? This tutorial connects two complementary strands of multi-agent systems research: AI-agent-based discussion and consensus support, and automated negotiation over complex utility spaces.

The first part examines how facilitation agents can elicit, organize, and mediate opinions in human and crowd-scale discussions. It also considers how LLM-based agents extend these capabilities through natural-language interaction while introducing risks such as hallucinated preferences, polarization, sycophancy, and premature consensus.

The second part focuses on automated negotiation involving multiple interdependent issues, nonlinear utilities, private preferences, and hard or soft constraints. Participants will learn how agents represent preferences, generate and evaluate offers, search agreement spaces, and balance individual utility with social welfare.

By integrating discussion support and automated negotiation, the tutorial provides practical concepts and evaluation criteria for designing consensus-support systems that are effective, explainable, robust, and appropriately governed by humans.

Target Audience:

The tutorial is intended for researchers, graduate students, system builders, and practitioners working in multi-agent systems, agentic AI, automated negotiation, collective decision-making, human-agent interaction, governance, and autonomous workflows.

The expected level is Beginner to Intermediate. Familiarity with basic AI concepts is helpful, but no prior knowledge of game theory, automated negotiation, fuzzy logic, or LLM orchestration is required. Necessary concepts will be introduced from first principles.

Outline:

Part I: Introduction

Consensus, deliberation, and negotiation
Differences among consensus, voting, negotiation, debate, and coordination
Classical multi-agent systems and current agentic AI

Part II: Discussion and Consensus Support based on AI Agents

Automated facilitation
Argument and discussion structures
Opinion elicitation and organization
Crowd-scale consensus support
LLM-based discussion agents
Human-agent collaboration
Risks, evaluation, and human oversight

Part III: Automated Negotiation among Agents under Complex Design Utilities

Multi-issue negotiation
Nonlinear utilities and interdependent issues
Hard and soft constraints
Preference and constraint representation
Offer generation, bidding, and mediation
Constraint relaxation and compromise
Search and optimization
Social welfare, fairness, and explainability

Part IV: Integration and Future Directions

Integration of discussion support and automated negotiation
Evaluation of efficiency, fairness, robustness, and human control
Case Study: AGEWEC 2026—Human-Governed Evaluation of Agentic Workflows

- Evaluation and human oversight of autonomous agent workflows

- Lessons for consensus-support system design; deployment considerations

Open research questions
Q&A

The tutorial follows the four-part structure described in the proposal.

Biography:

Takayuki Ito is a Professor of Social Informatics at Kyoto University. His research interests include multi-agent systems, automated negotiation, consensus building, collective and crowd intelligence, group decision support, computational mechanism design, and LLM agents. He has served as Program Co-Chair of AAMAS 2013 and PRIMA 2009 and General Chair of PRIMA 2014, and is a board member of IFAAMAS. He has also presented tutorials at major international conferences, including AAMAS.

Presenter Profile:
https://www.agent.soc.i.kyoto-u.ac.jp/~ito/

Presenter Photo:

Shun Okuhara is a Senior lecturer in the Graduate School of Engineering at Mie University. His research focuses on automated negotiation, multi-agent systems, consensus building, LLM-based decision support, explainable AI, and human-AI collaboration. His work includes automated negotiation involving multiple interdependent issues, constraint relaxation, negotiation-order protocols, and constraint-based approaches to efficient multi-issue negotiation.

Presenter Profile:
https://shunokuhara.github.io/website/

Presenter Photo:

Tutorial 2: Generative AI for Automated Negotiation

Instructors:

Yasser Mohammad, NEC Corporation / Assiut University

Keywords:

Automated Negotiation; Generative AI; Multi-Agent Systems

Abstract:

Automated negotiation is becoming increasingly important as autonomous AI agents interact across organizational and institutional boundaries. At the same time, generative AI has transformed the field: modern LLMs can negotiate in natural language, but they can also be exploitable, behaviorally biased, and difficult to evaluate.

This beginner-friendly tutorial introduces the foundations of automated negotiation, including negotiation protocols, the Nash bargaining problem, the BOA agent architecture, and multi-metric evaluation. It then presents a four-mode design space for combining generative models with negotiation strategies and surveys recent work on bilateral and multi-party negotiation, strategic reasoning, Diplomacy, persuasion, coaching, and human-AI negotiation.

A practical demonstration using the open-source NegMAS platform and negmas-llm library shows how generative negotiators can be implemented and evaluated. The tutorial concludes with open research challenges including robustness, theory-of-mind, fairness, and reliable agent-to-agent negotiation.

Target Audience:

The tutorial is intended for postgraduate students and researchers in multi-agent systems, game theory, simulation, generative AI, machine learning, and related applications.

The expected level is Beginner / Introductory. Familiarity with negotiation or LLM-based agents is helpful but not required. Basic Python knowledge is useful for participants who wish to follow the demonstration, although the tutorial can also be followed without programming.

Outline:

Part I: Foundations of Automated Negotiation

Automated negotiation problems and protocols
Alternating Offers Protocol and multilateral negotiation
Nash bargaining and fairness solutions
BOA architecture
Multi-metric evaluation of negotiation quality

Part II: Generative AI as a Negotiator

LLMs as negotiating agents
Prompting, personas, tool use, memory, and reflection
Opponent modeling
Four approaches to combining generation and negotiation strategy
Fluency, controllability, optimality, and exploitability

Part III: Generative AI for Negotiation

Capabilities and limitations of LLM negotiators
Strategic and game-theoretic reasoning
Multi-agent and multi-party negotiation
Commons dilemmas and cooperation
Diplomacy
Persuasion and negotiation coaching
Human-AI negotiation
Fairness and theory-of-mind

Part IV: Building Generative Negotiators with NegMAS and negmas-llm

Introduction to NegMAS and negmas-llm
End-to-end LLM negotiation
LLMs for natural-language and structured-offer translation
LLMs for bidding and acceptance decisions
Classical negotiators as tools for LLM agents
Evaluation, robustness, and exploitability

Part V: Open Problems and Research Directions

Robustness and exploitability
Theory-of-mind
Fairness and faithfulness
Agent-to-agent commerce
Concurrent negotiation
Hybrid generative and game-theoretic approaches
Protocol design, benchmarks, and competitions

The five-part structure follows the original tutorial proposal while omitting detailed minute-by-minute timing for the webpage.

Biography:

Yasser Mohammad is a Senior Research Scientist at NEC Knowledge Science Laboratories and a Professor of Intelligent Systems at Assiut University. He received his Ph.D. in Intelligence Science and Technology from Kyoto University in 2009. His research spans multi-agent systems, automated negotiation, machine learning, and time-series analysis, with recent work focusing on generative AI and reinforcement learning for automated negotiation.

He is the creator of NegMAS, an open-source platform for automated negotiation research, and has organized the Supply Chain Management League of the International Automated Negotiating Agents Competition (ANAC) since 2019. He has authored over a hundred publications, received multiple best paper awards including at IEEE ICA 2023, and presented tutorials at AAMAS, AAAI, PAKDD, PRIMA, AJCAI, and IEEE ICA. His experience with NegMAS and ANAC is also documented in the submitted proposal.

Presenter Profile:
https://yasserfarouk.github.io

Presenter Photo:

Tutorial 3: Model Reconciliation: Definitions, Approaches, and Applications

Instructors:

Tran Cao Son, New Mexico State University
William Yeoh, Washington University in St. Louis

Keywords:

Model Reconciliation; Human-AI Alignment; Explainable AI; Explainable Planning; Human-AI Collaboration

Abstract:

Explainable AI has become increasingly important for understanding why AI systems make particular decisions and for supporting effective interactions between humans and autonomous agents. Model reconciliation provides a formal approach for explaining differences between the knowledge or models held by different agents.

This tutorial introduces the Model Reconciliation Problem (MRP), its basic formalization, and its origins in explainable AI planning. It then presents computational approaches for solving MRPs, including their theoretical foundations and implementations using planning, SAT, and logic programming techniques.

The tutorial also examines advanced applications of model reconciliation to dialogue, negotiation, multi-agent diagnosis, and human-AI collaboration. It concludes by discussing emerging research directions, including human-AI alignment, explainability, and the use of model reconciliation for interactions involving LLM-based agents.

Target Audience:

The tutorial is intended for researchers and students interested in explainable AI, human-AI alignment, human-AI collaboration, planning, and multi-agent systems.

The expected level is Beginner to Intermediate. Some familiarity with SAT, logic programming, heuristic search, or related topics is helpful but not required.

Outline:

Part I: Introduction to Model Reconciliation

Introduction and motivation
Model reconciliation in planning
Basic definitions and formalization of the Model Reconciliation Problem
Relationship to explainable AI

Part II: Foundations

Planning
SAT-based representations and reasoning
Logic programming
Heuristic search
Formal foundations of model reconciliation

Part III: Approaches to Solving Model Reconciliation Problems

Computational approaches to MRPs
Algorithms and representations
Computing reconciliation solutions
Presenter-led demonstrations

Part IV: Advanced Topics and Applications

Advanced model reconciliation techniques
Dialogue
Negotiation
Multi-agent diagnosis
Logic program update
Human-AI collaboration and alignment

Part V: Future Research Directions

Explainability and human-AI alignment
Model reconciliation in multi-agent systems
LLM-based agents
Emerging research challenges

The original proposal contains approximately 220 minutes of content covering these topics; the webpage outline presents the same material thematically rather than as a detailed timetable.

Biography:

Tran Cao Son is Professor and Department Head of the Computer Science Department at New Mexico State University. His research interests include logic programming, reasoning about actions, and explainable AI. He has presented tutorials at IJCAI, AAAI, and ICAPS, and delivered “Model Reconciliation and Its Applications in Explainable AI” at the Autumn School on Logic Programming at ICLP/LPNMR in 2024.

Presenter Profile:
https://www.cs.nmsu.edu/~tson

Presenter Photo:

William Yeoh is a faculty member in Computer Science & Engineering at Washington University in St. Louis. His research interests include explainable AI planning, multi-agent constrained optimization, and human-AI collaboration. He has presented tutorials on multi-agent constrained optimization at AAAI and on goal recognition at ICAPS.

Presenter Profile:
https://wyeoh.github.io/

Presenter Photo: