How the AAA-ICDR Institute and Northwestern Law are combining interdisciplinary experimentation with ADR expertise to explore better mediation preparation for self-represented parties
The AAA-ICDR Institute™ was established as the “think and do tank” of the American Arbitration Association® (AAA®), with a mandate that extends beyond developing ideas to creating environments where they can be tested, challenged, and translated into practical learning. This approach is particularly useful for emerging questions involving AI, access to justice, and alternative dispute resolution (ADR), where technical capabilities are developing quickly, but their value depends heavily on how well they account for the people, processes, and institutional realities surrounding their use.
The Institute’s recent collaboration with Northwestern Pritzker School of Law’s Innovation Lab put that model into practice. An interdisciplinary team of law and computer science students explored whether an AI mediation coach could help self-represented parties prepare more effectively for mediation, initially focusing on employment disputes. Northwestern brought its structured approach to interdisciplinary innovation; the Institute provided sustained engagement with AAA mediation expertise and practice. Together, the teams could investigate not only whether a particular technology could be built, but what it should do, where its boundaries should lie, and what self-represented parties may actually need from it.
Professor Daniel Linna, Director of Law and Technology Initiatives at Northwestern Pritzker School of Law, describes the Innovation Lab’s methodology as a “people-process-data-technology approach to innovation,” incorporating Agile project management, design, and lean thinking, iterative development, and user testing. The partner organization is integral to that process. “This isn’t about the organization outsourcing a project,” Linna said. “The student team and organization work together closely to learn about the problem and iteratively develop solutions.”
For the Institute, that meant weekly engagement with the student team, practical guidance on mediation and pre-mediation statements, scenarios for prototype testing, and access to experienced AAA mediators Anne Jordan and Karen Layng. The Institute also worked with the AAA’s Mediation division and Employment teams to arrange opportunities for students, with the consent of parties and mediators, to observe active mediations involving self-represented parties. Those interactions allowed assumptions emerging through research and prototype development to be examined against mediation as it is actually practiced.
Designing AI Around Mediation
The initial premise behind the AI mediation coach was relatively straightforward. Self-represented parties may enter mediation without counsel to help them organize relevant facts, understand the process, clarify their objectives, evaluate alternatives, or prepare for difficult conversations. A structured AI tool could potentially provide some of that preparatory support without attempting to replace either counsel or the mediator.
As the project developed, however, the students encountered a more complex design problem. Their research had already identified emotion as an important component of conflict, but observing mediation deepened their understanding of how emotional readiness affects the process itself. Rather than seeing mediators move immediately into substantive negotiation, students observed the considerable work involved in explaining the process, establishing expectations and ground rules, listening, building trust, and helping parties feel heard before productive settlement discussions could begin.
David Gerchik, the first student on the team to observe a mediation, was particularly struck by how much time the mediator devoted at the outset to preparing participants for the process. The observations reinforced a direction already emerging in the students’ work: effective mediation preparation involves more than providing information, organizing evidence, or identifying settlement goals. The prototype incorporated emotional check-ins that encouraged users to reflect on what was driving their concerns and consider different perspectives as part of preparing to participate constructively.
This raised a more interesting question for the design of AI-supported mediation preparation. Trust, emotional readiness, feeling heard, and self-determination are integral to mediation, but they are not simply functions that can be automated. A useful mediation coach, therefore, needed to support the conditions for productive participation without attempting to reproduce the mediator’s role. That distinction began to define both the capabilities of the prototype and its limits.
Turning Mediation Principles into Product Design
Jordan’s and Layng's involvement brought those limits into sharper focus, particularly around self-determination. Their practitioner feedback consistently reinforced that the tool could help users prepare, reflect, organize their thinking, understand mediation, and consider their options, while avoiding legal advice, predictions about outcomes, or questions and guidance that might steer someone toward a particular settlement position or decision.
Self-determination became relevant not simply as an ethical boundary around the technology, but as an input into its design. It affected the kinds of questions the AI could appropriately ask and the distinction between helping someone make a better-informed decision and influencing what that decision should be. Practitioner expertise entered the development process while those choices were still being made, rather than functioning only as downstream validation of a finished product.
Privacy and mediation confidentiality created a similar opportunity to interrogate functionality before it became fixed. Features allowing users to describe disputes or upload documents raised questions about personally identifiable and other sensitive information entered into an AI system. Mediation confidentiality influenced what dispute information users should be encouraged to provide in the first place.
Those discussions resulted in practical changes to the prototype, including guidance to avoid entering unnecessary sensitive information and redact PII where appropriate, and greater transparency about the tool’s intended use. They did not resolve the broader privacy and confidentiality questions associated with AI-assisted mediation, but they changed how the students approached development. Capabilities that appeared useful in isolation had to be reconsidered in the context of real-time use.
What the Prototype Taught Us About AI Mediation
The final prototype combined a guided interview, mediation education, and AI-assisted coaching. Initially designed around employment disputes, it asked structured questions about a user’s employment history, underlying events, and desired outcomes. It helped organize relevant facts and identify issues, explained the process and mediator’s role, incorporated opportunities for reflection, and supported document creation. The AI functioned as a structured coach, guiding users through preparation without requiring them to know which questions to ask at the outset.
The development process also produced a more nuanced picture of what AI mediation coaching might usefully accomplish for self-represented parties. Information is necessary, but on its own, may not do much to prepare someone to participate effectively. Process understanding, realistic expectations, emotional readiness, perspective-taking, and the ability to articulate priorities all influence how a party enters mediation. At the same time, designing technology to support those capabilities requires particular attention to preserving the user’s agency and maintaining appropriate boundaries around legal advice and mediation decision-making.
As Linna observed, “The final prototype is an important deliverable, but a key takeaway is that the learning generated from a systematic approach to innovation is just as—if not more—important.” For the AAA, these collaborations can function as applied R&D and strategic learning, creating a structured environment in which emerging ideas can be investigated, challenged through interdisciplinary and practitioner perspectives, prototyped, and better understood before determining whether further development or investment is warranted. The knowledge generated through that process has value in its own right, including when the work reveals new questions or changes the understanding of the problem being explored.
The model has implications beyond a single AI mediation coach. Law schools increasingly combine legal scholarship and clinical expertise with computer science, design, data, and other disciplines capable of contributing meaningfully to justice-system innovation. Institutions like the AAA possess a different form of knowledge: accumulated experience with disputes, processes, practitioners, and practical constraints. Sustained collaborations give academic teams access to problems and environments that can make their work more consequential, and institutions benefit from a rigorous setting in which to investigate emerging ideas.
For the Institute, that intersection is where thinking and doing become mutually reinforcing. Applied research can shape experimentation, practice can challenge assumptions, and prototypes can generate knowledge even before there is a decision about whether they should become something more. The opportunity is not limited to identifying the next technology worth building; it is to develop better ways of investigating the questions that will shape how ADR evolves.