A conversation with Pietro Ortolani on agentic AI and the future of the human mediator
Last time, we talked with Amy’s co-editors about their new Cambridge Handbook of AI and Civil Dispute Resolution. It was a lively conversation, but we ran out of time before we could take up a topic that is on everyone’s mind right now: agentic AI. So, we hosted this conversation with Pietro to pick it back up.
A Buzzword with a Real Idea Inside
Pietro began by admitting that “agentic AI” is more buzzword than term of art, and that is part of its appeal. It makes a good headline, especially when the story is about a robot that goes rogue. The underlying idea is simpler. An agentic system can act in the world with enough autonomy that you can hand it a task, and it will carry it out for you. Pietro’s example was ordinary: buy me some sports gear, sign me up for a gym, and put the first workout on my calendar. He was careful to add that “agent” here does not mean agent in the legal sense.
Amy noted that this is no longer abstract. Her phone recently asked whether she wanted an AI agent to make purchases for her. She said no. But the offer shows how quickly these tools are arriving in everyday life.
A Book That Was Already About Agents
The Handbook never used the word “agentic,” but Pietro pointed out that it was largely about AI systems interacting within dispute resolution procedures, which is close to what an agent is. Its focus was on how humans and machines interact in civil justice, not on AI acting with complete independence.
Pietro stressed that this was a normative choice, not a technical prediction. The authors do not claim AI will never be able to act independently. As he put it, nobody knows whether something surprising will happen in five years or five days. What the contributors share is a conviction that humans should keep contributing to civil justice, even though they reach it by different routes. Some start from human rights and due process: if I am being judged, I have a right to a human judge. Others start from a philosophical observation: however advanced the models become, they do not experience reality the way a person does, so they cannot replicate what a mediator, arbitrator, or judge does, including the emotional intelligence involved.
Because no one can reliably predict where AI is headed, Pietro argued that the useful question is not what AI will look like in five years. It is what principles should guide us in dealing with whatever it becomes. Amy added that the Handbook, which even includes a chapter on the “Terminator judge,” tries to lay foundations that hold up whatever comes next.
The Uncomfortable Question
Then Bill asked the question that has been nagging him. What if we are wrong? We often say humans will always be better at empathy, cultural humility, and building trust in real time. But people once believed no computer would beat the best human chess players, and that a human paired with the machine would at least stay competitive. For a while, that seemed right. Then the machine simply pulled ahead. What if mediation turns out to be the same?
No one claimed to have a tidy answer, and we appreciated that. Three ideas stayed with us.
First, “better” has to be defined. Amy pointed out that the claim that AI could be better than a human mediator means little until we say what we are measuring. She is working on benchmarking that compares AI and humans, including on soft skills. Parties judge a process by more than its outcome. They care whether they felt heard and whether the process felt fair. In her work with JusticeTech on an online mediation platform, she and her colleagues keep returning to the point that parties should be able to choose a human, whether over Zoom or in person. Asynchronous or automated communication alone may not leave people feeling heard.
Second, this is not only an efficiency question. Pietro said that even if AI became very good at what it cannot do today, whether humans should keep a role would still not reduce to measurable efficiency or technical feasibility. It would be a philosophical, ethical, and political question, because it means handing a fundamental function of dispute resolution, with something of the character of sovereign power, to technology that may be supplied by a private company and that operates by a different logic than a human institution. His hope is for a deep, multidisciplinary debate, not a decision made because something is cheaper and technically possible. Amy agreed and added that psychologists and sociologists need to be at the table along with lawyers.
Third, being capable does not mean being chosen. Pietro reminded us of the pandemic, when companies bet heavily on virtual reality and the metaverse, convinced people would never want to meet face to face again. The technology worked well enough. People still preferred a real conversation with a friend to a virtual beer in a headset. Dispute resolution may turn out the same way, with all the usual differences.
What This Means for the Everyday Mediator
Amy also raised a practical concern, drawn from an earlier conversation with Sarah Cole. If we hand too much of this work to AI, new mediators may never develop the soft skills that make a mediator valuable, such as managing communication, building trust, and being trustworthy. Those are the very skills the market prizes most. Deskilling is a risk even if the technology never surpasses us.
So here is what we are taking away. Ask what “better” means before accepting that a tool is better. Keep a human option available to parties. Keep practicing the skills you would not want to lose. And if the day comes when a machine can do more of what we do, make that decision deliberately, with many voices in the room.
Watch the full conversation with Pietro in the video below and tell us what you think.
Author’s Note: Claude developed the first draft of this column. Amy and Bill edited.
