To this day, one of my most popular and viral posts ever involved the 2017 launch of LegalBoard, a keyboard designed for lawyers by lawyer Brian Potts, who is now a partner at Husch Blackwell. Hits to that post went through the roof, far exceeding any post I’d ever before published. Now, Potts is back, […]
To this day, one of my most popular and viral posts ever involved the 2017 launch of LegalBoard, a keyboard designed for lawyers by lawyer Brian Potts, who is now a partner at Husch Blackwell. Hits to that post went through the roof, far exceeding any post I’d ever before published.
Now, Potts is back, teaming up with a Harvard Law School student and a computer scientist to launch what they say is the first end-to-end AI arbitration system, with the potential to bring about a significant shift in how commercial disputes are resolved.
Fortuna Arbitration, led by Harvard Law School student Kimo Gandall, computer scientist Kenny McLaren, and Potts, will debut its Arbitrus.ai platform on Feb. 3, offering automated arbitration services at what the company says will be half the cost of traditional arbitration.
The product, which the developers have tested on 100 hypothetical scenarios, achieved zero hallucinations and maintained complete issue coverage across all test cases, according to a detailed law review article authored by the founders along with federal judicial clerk Jack Kieffaber, “We Built Judge.ai. And You Should Buy It.”
Targeting Private Disputes
The platform aims to significantly reduce the cost of dispute resolution, while providing what the company argues will be more predictable and consistent outcomes than are achieved with human arbitrators.
“Arbitrus.ai is a better and more feasible manifestation of Judge.ai than Judge.ai itself,” the founders write in their article, referring to an earlier proposal for automated judicial decision-making published by Kieffaber last September.
Rather than attempting to replace public court systems, the Arbitrus.ai system is designed to operate within the framework of the Federal Arbitration Act and targets private contractual disputes.
“Nobody will ever throw a switch and spontaneously consolidate the federal judiciary into one robot,” the developers say in their article. “Rather, the machine will cut its teeth on the private market by automating arbitrations.”
Although I have not seen the product in action, it appears from the article that Arbitrus.ai represents one of the most ambitious attempts yet to automate core legal decision-making in the dispute resolution context.
Multi-Layered Approach
According to the law review article, Arbitrus.ai employs a multi-layered approach combining classical machine learning, human-driven reinforcement training, and large language models. The system processes cases through several stages:
- Initial processing and classification using voting ensemble models to determine basic outcomes.
- Legal reasoning and analysis using large language models.
- Response verification and checking through a proprietary agentic system called Ra.ai.
- Human review of outputs, at least during the initial deployment phase.
The company has developed a preliminary set of procedural rules governing everything from initial filing to appeals, with specific timeframes for each stage of the process. Cases typically proceed through notice, briefing, discovery, and hearing phases, with decisions promised within three days of the final hearing.
Cost Structure
According to the fee schedule set out in the procedural rules, the cost of using the service depends on the nature of the case and the amount in controversy.
For commercial cases, the schedule of fees is as follows:
- Claims under $75,000: $2,500 initial filing fee plus $2,500 closing fee.
- Claims $75,000-$150,000: $5,000 initial filing fee plus $5,000 closing fee.
- Claims $150,000-$500,000: $10,000 initial filing fee plus $5,000 closing fee.
- Claims over $500,000: $10,000 initial filing fee plus 1% of claim value (up to $250,000) closing fee.
For consumer cases, rates are much lower, with initial filing and closing fees as low as $225 for claims under $75,000.
Testing and Validation
In their article, the developers describe having extensively tested the product across various case types, including disputes involving contracts, insurance, intellectual property, employment, real estate, and corporate matters.
“In totality,” they reported, “Arbitrus.ai performed as expected: it did not hallucinate; it answered the issues in controversy, and almost always — with two narrow exceptions — grounded those answers in relevant case law.”
Based on the article, there are some potential drawbacks to using the platform as compared to a human-led arbitration. For one, the hearing is conducted entirely in writing, with no opportunity for cross-examination or oral arguments. For another, it appears the system is not designed for disputes with more than two parties.
Future Implications
In their article, the developers say that the launch of Arbitrus.ai potentially represents a significant step toward what they term the “Arbitration State” – a system where private automated dispute resolution could handle an increasing share of legal conflicts.
While the immediate focus is on commercial arbitration, the technology’s developers suggest it could eventually expand to handle a broader range of legal matters. According to the article, the company has already introduced Arbitrus.ai to large institutions in Latin America, including Colegio de Abogados De Lima.
While online dispute resolution platforms are nothing new, Arbitrus.ai appears poised to be a proof of concept for the extent to which AI can automate legal decision making.
