The American Arbitration Association just launched its AI Arbitrator system for construction disputes, and it's already processing its first case. Bridget McCormack, former Michigan Supreme Court Chief Justice and now AAA president, told The Verge the system uses around 20 AI agents to parse claims, organize evidence, and draft awards—all with a human arbitrator signing off at the end. It's a limited rollout focused on documents-only construction cases, but McCormack predicts we'll look back in 10 to 30 years and wonder why we ever trusted humans to decide so many legal disputes. That's a bold claim for a system that's handled exactly one case.
The century-old nonprofit arbitration giant just rolled out what might be the legal profession's most consequential automation experiment yet. But with just one case in the system and trust in both courts and corporations at historic lows, the AI Arbitrator faces a credibility test before it can prove whether machines really can deliver fairer, faster justice.
McCormack came to the AAA in February 2023, right after ChatGPT's November 2022 launch sent shockwaves through every industry. She spent her six-week break before starting the job diving into large language models, convinced they'd reshape dispute resolution. Her first move at AAA? Give everyone—engineers, caseworkers, legal staff, marketers—enterprise ChatGPT licenses. "With any general-purpose technology, you need the domain experts to figure out where it's going to impact them," she explained.
The AI Arbitrator works like this: parties submit complaints, pleadings, and evidence through a web-based case management system. A series of agents then parses the claims, identifies the elements of each claim, maps evidence to those claims, and contextualizes everything within the relevant legal framework. Then it circles back to both parties: "Here's my understanding. Did I get that right?" If they say no, the agents go back to work. This loop continues until both sides confirm they've been heard and understood.
That feedback mechanism is the secret weapon, McCormack argues. Human judges rarely have the time or resources to check in mid-process and confirm they've grasped every argument. Appellate courts routinely ignore entire issues in briefs. "I reviewed 2,000 applications a year when I sat on the Michigan Supreme Court," McCormack said, "and I can't tell you how many times the intermediate appellate court didn't even rule on issue number three. They just ignored it."
Once parties are satisfied, reasoning agents take over. They analyze the dispute summary, draft an award, and hand it to a human arbitrator—one of AAA's construction specialists—who reviews the work, makes any necessary changes, and issues the final decision. That human-in-the-loop design is AAA's guardrail against hallucinations, which McCormack acknowledges as "a huge issue." The system is "governed, trained and grounded" in construction arbitration case law, not just a raw LLM spitting out answers.
Why start with construction? AAA does hundreds of thousands of cases a year, and construction disputes are common, documents-heavy, and happen in an industry already comfortable with AI. "Construction sites are being infiltrated with AI that's making what they do significantly faster and better," McCormack noted. Plus, AAA had a deep library of historical construction cases to train the agents on. The partnership with McKinsey's Quantum Black AI team ran about 10 months—McKinsey engineers led for the first half, AAA's team drove the second half after the MVP shipped.
But the system's early-stage reality check is stark: one case. That's it. Both parties must agree to use AI arbitration, and since no contracts included it before the November launch, adoption depends on word of mouth and new contract clauses trickling into the market. McCormack is optimistic that businesses will start adding AI arbitration clauses once they see the speed and cost benefits. Early interest is coming from companies that want to use it as an "early case evaluation tool"—plug in your evidence, get an AI-generated assessment, decide whether to proceed.
The bigger question is trust. Americans' confidence in the judicial system hit a record low in 2024, according to PBS. McCormack argues that 92% of people can't afford lawyers, most disputes never get resolved, and human judges make plenty of errors—she cited DNA exoneration data showing 3 to 5 percent wrongful convictions. "If you're landing planes, that's not a great number," she said. "And I think the criminal justice system should be more like landing planes."
Still, McCormack drew a hard line: criminal cases and disputes against the government must stay in public courtrooms with publicly appointed judges. "Those have to happen in courtrooms," she said. Everything else—B2B construction disputes, payer-provider healthcare billing fights, supplier disagreements—could benefit from AI arbitration if it's transparent, audited, and shows its work.
That's a big if. Arbitration already has an image problem. Most consumers experience it as a clause buried in terms of service agreements they never read, stripping away their right to sue in public court. Disney famously tried to force a wrongful death lawsuit into arbitration because the plaintiff had signed up for Disney Plus. McCormack counters that AAA requires businesses to file their consumer contracts and satisfy due-process protocols before using AAA arbitration—a safeguard not all for-profit arbitration providers enforce.
She also believes AI can de-bias dispute resolution more easily than human judges. "You can de-bias the data set a lot easier than you can de-bias a human," she said. When you benchmark and audit an AI arbitration system and show your work publicly, "you can either convince the public that it's treating both people fairly or you won't. That's for us to show."
The roadmap from here is slow and cautious. AAA's engineering team will build out support for new dispute types—documents-only cases in energy, healthcare billing, supplier contracts—over the next couple of years. McCormack expects that within 15 years, businesses won't bother with slow, expensive, human-led processes for routine B2B disputes. But she also acknowledged the legal profession's training model is "broken" for an AI-driven future, and no one's rushing to build the 2.0 version.
One wildcard: agentic commerce. Walmart already uses AI agents to negotiate and execute contracts, and some estimates suggest 40% of B2B contracts could be agent-to-agent by late 2027. When those agents make mistakes—and they will—who arbitrates the fallout? McCormack wants a seat at that table. "Your agentic commerce is only as good as your process for fixing it when it breaks," she said.
For now, the AI Arbitrator is a live experiment with a sample size of one. If it works—if parties feel heard, if the reasoning is transparent, if the outcomes are fair—McCormack's vision of AI judges handling routine disputes could reshape access to justice. If it doesn't, the backlash could set legal AI back for years. Either way, this is the first real-world test of whether machines can judge us better than we judge ourselves.
The AI Arbitrator is live, but it's barely breathing—one case, a handful of interested businesses, and years of iterative rollout ahead. McCormack's confidence that we'll look back and marvel at human judges handling routine disputes hinges on this system proving it can be transparent, fair, and trustworthy at scale. The legal profession has resisted disruption for 250 years; it won't roll over for AI without a fight. But if AAA can show that machines make fewer mistakes, cost less, and give people a genuine sense of being heard, the arbitration industry—and maybe the courts—will have no choice but to follow. The question isn't whether AI will reshape dispute resolution. It's whether the first movers get it right before the backlash hits.
