AI Could Open the Door to Non-Lawyer Judges-But Human Judgment Still Decides
Singapore's courts are leaning into AI-summarising self-repped cases, modelling likely outcomes for motor accidents, and exploring low-cost transcription. The head of transformation and innovation, Justice Aidan Xu, believes this arc could reshape who sits on the bench.
His view: as AI matures, deep specialist legal knowledge may matter less for some judging roles. What will never be optional is human judgment-credibility assessments, fairness, and the final decision.
What That Could Look Like
Picture a family judge focused on counselling and guiding parties through a divorce, with AI handling precedent and statute lookups. Or an engineer sitting as judge in construction disputes, making sense of technical facts while AI supports the legal analysis.
Today, the law still requires seven years as a "qualified person" before one can be appointed a district judge. That requirement lives in the Legal Profession Act, which frames who is considered a qualified person in Singapore's legal system. See the Act for definitions and pathways: Legal Profession Act (Singapore Statutes Online).
What Singapore's Courts Are Testing Now
- AI summaries for Small Claims Tribunals: Built with Harvey AI, the tool drafts clean, side-by-side case summaries so self-represented parties and tribunal magistrates can grasp the core issues faster.
- Algorithms for motor accident claims: Rules that factor collision type, location, and damage to narrow likely outcomes-helping parties decide whether to negotiate or settle.
- Translation at scale: A translation tool already supports Small Claims matters; bulk translation is on the roadmap. Human interpreters remain essential for nuance and contested proceedings.
- AI-enabled transcription: Ongoing trials aim to offer accurate, low-cost transcripts to the public. Today's commercial services are priced in the thousands per hearing day-out of reach for many.
- Future exploration: Settlement proposals generated from case ranges (e.g., maintenance claims), presented as options parties can accept or refine.
The Hard Problems No Model Fixes Yet
Real courtrooms aren't controlled environments. Old rooms have tricky acoustics. Sensitive mics sometimes capture "hot mic" side conversations. And code-switching mid-sentence (English to Mandarin and back) can derail automatic recognition and translation.
There's also misuse risk. Some lawyers have cited hallucinated cases. The courts are training, guiding, and monitoring use, and individual judges are handling incidents. If needed, a broader policy response may follow.
Principles the Judiciary Is Emphasising
- AI is a tool, not a shortcut: No "click, paste, submit." Judges and officers remain responsible for the work product.
- Training and guidance over blanket disclosure: A single disclosure checkbox doesn't work when AI can touch everything from research to drafting.
- Tech isn't always the answer: Sometimes the fix is a simpler process, a rule change, or better information design-plus real human help at the counter.
Job Impact: Fewer Hours on Grind, More Time on Judgment
Expect less manual document review and boilerplate drafting by juniors. That's the immediate shift. The opportunity is to move talent toward judgment-heavy work: fact synthesis, client strategy, case theory, negotiation, and courtroom craft.
This raises a training problem: if juniors do less grunt work, how do they build judgment? The judiciary expects more simulations, role-playing, and targeted instruction. Firms will need to design new development paths.
Action Steps for Legal Teams
- Adopt an AI-use policy with verification requirements, source logging, and a "no unverified citations" rule.
- Stand up a fast-check protocol: case law validation, quotes to source, and facts to evidence before filing.
- Pilot AI summarisation on matters with self-represented litigants to improve clarity and cut hearing time.
- Budget for low-cost transcript solutions as they become available; set accuracy thresholds and fallback plans.
- Redesign junior workflows: more structured training, shadowing, and simulation to build advocacy and judgment.
- Prepare for multilingual hearings: mic discipline, language markers for transcripts, and interpreter-first practices where stakes are high.
- Create an incident playbook for AI errors (hallucinations, misquotes, privacy breaches) with escalation paths.
- Use internal audits: random spot checks of AI-assisted work, plus matter-level attestations before filing.
What to Watch in the Next 24 Months
- Standards for AI-assisted judging and drafting: explainability, transparency, and audit trails.
- Cost curves for transcription and translation-and how that affects access to justice.
- Regulatory experiments that could let domain experts adjudicate specific disputes with AI support.
- Firm-level training models that replace "learning by grind" with simulations and structured coaching.
Upskilling: A Practical Starting Point
If you're building capability fast-especially for trainees and first- to third-year lawyers-pick targeted, hands-on courses on prompt practice, verification, and workflow design. A curated path by job function can help:
AI courses by job role (Complete AI Training)
The Bottom Line
AI will change who does what in legal. It may even change who can be a judge. But it won't replace the human role at the heart of adjudication: weighing people, facts, and fairness-and owning the decision.
Your membership also unlocks:
