Highlights
- AI can dramatically boost court efficiency—without replacing judicial judgment
- Judges must follow the “bright line test”: If a clerk can’t do it, AI can’t do it
- A seven‑step framework ensures responsible, secure, and accurate AI use
In a recent American Bar Association webinar, three legal technology leaders provided judges with a critical framework for navigating AI adoption: embrace the efficiency, but never delegate the judgment.
The session, “From Bot to Bench: Managing Responsible Use of AI in the Judiciary,” brought together Hon. Samuel Thumma of the Arizona Court of Appeals, Hon. Robert Hilliard of Santa Rosa County Court in Florida, and Mike Dahn, Head of Westlaw Product Management at Thomson Reuters. Moderated by Hon. Heather Welch, the conversation delivered a practical roadmap for judicial chambers grappling with AI’s promise and peril, which we’ve outlined for you here.
Jump to ↓
The promise: Real-world results from courts worldwide
The foundational principle: Human in the loop
The bright line test for delegation
Understanding AI’s unique error profile
Seven-step workflow for responsible AI use
The optimistic view: Transforming access to justice
The promise: Real-world results from courts worldwide
The evidence is compelling: In Argentina, the virtual AI assistant Prometea has helped legal professionals process nearly 490 cases per month, compared to just 130 before its introduction—almost a 300% increase in productivity.
In Germany, the Frankfurt District Court’s AI system “Frauke” has significantly reduced processing time in air passenger rights cases by extracting case-specific data and expediting judgment drafting, allowing judges to focus on complex legal reasoning rather than repetitive documentation.
Perhaps most inspiring are the access-to-justice victories: Self-represented litigants are using AI tools to successfully navigate complex legal proceedings, with one California tenant negotiating a settlement and saving over $2,000 using AI-assisted legal research.
The foundational principle: Human in the loop
Judge Hilliard articulated the core principle through analogy: “It’s not a lawyer, it’s not a judge, but don’t panic. It’s a tool. And it’s a tool that, when like any power tool properly used, can enhance a good builder’s capabilities and if used improperly it can be very dangerous.”
Mike Dahn reinforced this: “AI makes mistakes very regularly, and so that’s why a human needs to be in the loop is to know what it does well and what it doesn’t do well and to correct mistakes.”
The bright line test for delegation
Judge Thumma offered what he called a “bright line test”: “If you wouldn’t let a clerk do it, don’t let AI do it.”
Judge Hilliard framed it another way: “If the task at hand requires judgment or discretion, or weighing evidence, credibility, weighing or applying law to facts, then AI is out. That remains the judge’s non-delegable duty.”
This principle draws a sharp distinction between judicial functions and administrative tasks. As the panelists emphasized, AI cannot perform constitutional duties like credibility determinations or applying law to facts. These remain exclusively within the judicial domain.
The corollary: “If it requires the judge’s brain, AI can’t do it. If it requires the judge’s time, then maybe AI can assist.”
Understanding AI’s unique error profile
To maximize benefits while managing risks, judges must understand AI’s limitations. Dahn provided a taxonomy of AI errors beyond hallucinations: fabricated case citations, incomplete discovery of relevant authorities, and misplaced focus on primary content while missing critical exceptions.
This led him to revise the common mantra: “I would alter that slightly to say don’t trust; verify.”
Seven-step workflow for responsible AI use
The panelists outlined a rigorous framework:
- Determine whether AI should be used. Only tasks requiring judicial time, not judicial judgment, are appropriate for AI assistance.
- Use the right tool. Choose enterprise-level legal platforms with zero data retention, encryption, and comprehensive security measures.
- Conduct preliminary judicial review. AI doesn’t get the first look at a case—the judge does.
- Upload material and compose prompts. Only share permissible information. Use neutral, specific prompts.
- Draft all substantive legal content yourself. Findings of fact, credibility determinations, and legal conclusions remain exclusively judicial functions.
- Verify everything. Check whether facts come from the record, whether citations exist and are accurately quoted, and whether summaries are balanced.
- Final judicialization. Perform a line-by-line review of the entire document, not just AI-touched parts.
The optimistic view: Transforming access to justice
Despite cautionary notes, Judge Thumma sees enormous potential in expanding access to justice for self-represented litigants through training, automation, smart chatbots, and case triage.
As Mike Dahn summarized: “This is not the law profession being replaced. They’re being equipped.”
Don’t let the complexity of AI adoption delay the benefits. Start with informed guidance built specifically for the judiciary—because when judges are equipped with the right tools and frameworks, justice moves faster without compromise.
Advanced AI built for judicial precision
Improve efficiency with secure, enterprise‑grade tools that reduce errors and enhance public service.
Discover what’s possible ↗
