This article first appeared in Forum, The Edge Malaysia Weekly on October 26, 2026 - November 1, 2026

On Aug 18, in the final months of his time as Master of the Rolls and Head of Civil Justice in England and Wales, Sir Geoffrey Vos gave the John Lehane Memorial Lecture at the Supreme Court of New South Wales. It deserves careful reading in every common law jurisdiction, ours included.

His starting point was blunt. For lawyers and judges, he argued, adopting artificial intelligence (AI) is no longer optional. Access to justice demands it, and a justice system that drifts into irrelevance, he warned, would be the gravest threat of all to the rule of law.

His second proposition was the striking one. Sooner or later, he believes, AI will cross from informing judicial decisions to making them, first in narrow fields and not everywhere at once.

Given the choice between a machine’s answer in minutes and a human judgment months or years away, many litigants, he thinks, may take the machine.

The earliest candidates, he suggested, are largely formulaic decisions, such as assessing damages in personal injury and medical negligence claims.

And then he drew the line. The rule of law itself, he argued, requires us to resist machine-made judgments imposed on parties without their free and fully informed consent.

In his view, which he acknowledges others dispute, no machine is ever likely to qualify as an “independent and impartial tribunal”, and dislodging the human judge would take legislation, nothing less.

On Oct 9, the judiciary of England and Wales put a version of that line in writing. A position statement signed by Sir Colin Birss, its lead judge for AI and the judge appointed to succeed Sir Geoffrey, accepts that AI may support judicial work but insists that decisions belong to judges, and that “AI tools must not determine the outcome of a case”.

The direction of travel is unmistakable, and it makes one distinction indispensable.

May the machine assist? And may the machine decide? Nearly everything confused in the public conversation about AI and the courts comes from running these two questions together.

The answer this article defends is simple to state. The machine may assist the judge. It must not, however, become the judge.

Malaysia was not a bystander in this story. We were an early mover.

At the Opening of the Legal Year in January 2020, the courts of Sabah and Sarawak launched an AI sentencing tool, developed with Sarawak Information Systems, to analyse sentencing data and trends for selected offences.

On Feb 19, 2020, the Magistrate’s Court in Kota Kinabalu became the first court in Malaysia to use it, in drug possession cases.

Not every lawyer welcomed it. Hamid Ismail, whose client in Sabah was sentenced with its help, told the Thomson Reuters Foundation in 2022 that nothing in the Criminal Procedure Code provides for AI in the courts.

Objections of that kind are valuable, because they force the point of principle into the open. The judiciary’s answer has been consistent and it is the right one. The system’s output is a recommendation, not a sentence, and the judge is not bound by it. The machine proposes, the judge disposes.

Chief Justice Tun Wan Ahmad Farid Wan Salleh restated the position to reporters in December 2025. AI can help the courts see sentencing trends, he said, but judges are not bound by what it suggests, and the final decision is theirs. In his words, “the human aspect of AI is lacking”. It is a phrase that could serve as the epigraph for this entire debate.

In Malaysia, the question of the machine as judge is not merely one of policy. It has a constitutional floor.

That floor was tested in an unlikely setting, a dispute over compensation for compulsorily acquired land. In Semenyih Jaya Sdn Bhd v Pentadbir Tanah Daerah Langat [2017] 3 MLJ 561, the statute made the opinion of two lay assessors on compensation binding on the High Court judge. The judge could not reach an independent assessment. His function had been reduced to endorsement.

“Wherefore now stands the judge?” the Federal Court asked. “It would appear that he sits by the sideline and dutifully anoints the assessors’ decision.”

The court’s answer was unanimous. Judicial power resides in the judiciary, and the provision was struck down. The assessors could advise, but their opinion did not bind the judge, and the decision was for the judge alone.

Semenyih Jaya did not concern AI. Our argument is that its reasoning applies equally when the proposed substitute is an algorithm.

If parliament cannot make a human assessor’s opinion binding on a judge, a law requiring a judge to accept an algorithm’s determination would, by parity of reasoning, be constitutionally untenable.

The analysis does not change because the assessor has become an algorithm. An AI system holds no judicial office, takes no oath and bears no legal responsibility for its reasons or its result. The machine, not unlike the assessors, may inform the exercise of judicial power. It cannot, however, exercise that power in place of the judge.

The machine as assistant is an entirely different matter, and the gains on offer are real. It can summarise voluminous records, help with translation, transcribe hearings, support case management and analyse sentencing patterns of the kind the chief justice described.

That last use serves consistency and parity in sentencing, which are themselves aspects of justice. Used this way, AI is efficiency in the service of judgment, not a substitute for it.

But the assistant’s chair has a gravitational pull towards the bench, and Sir Geoffrey acknowledged it with unusual candour.

Present a judge with a thousand machine recommendations, he warned, and “economic and time pressures will inevitably lead to judicial approval” of them. Hamid, in that 2022 interview, made the point closer to home. A young magistrate, he said, may treat the recommendation as though it came from a senior judge and simply adopt it.

That risk has a name, automation bias, and it is the subtlest danger of all. A judge who defers to the machine by habit becomes the sideline judge by practice rather than by statute. No constitutional amendment is required for that erosion. Only fatigue.

If our argument is right, the constitution closes the front door. Only judicial vigilance guards the back.

This is also why the seemingly modest proposal, to let AI decide only the “simple” cases, deserves more resistance than it usually receives. The pressure will come. Sir Geoffrey himself thinks people are likely, for reasons of cost, to come to accept machines resolving small disputes, at least at first instance.

But deciding what is simple is in itself a judgement call. A parking dispute is simple until it is the small trader’s livelihood. And a system in which one class of citizen retains access to a judge while another is confined to machine determination risks creating a two-tier system of justice, whatever efficiencies it delivers.

None of this forbids parties from choosing the machine for themselves. In November 2025, the American Arbitration Association launched an AI arbitrator for two-party, documents-only construction disputes, for parties who opt in.

The system proposes the award, and a human arbitrator approves and issues it. Sir Geoffrey saw nothing in principle to stop commercially sophisticated parties from choosing such a process.

But adjudication by a court exercising the judicial power of the federation is not a mere service selected from a menu. Its decisions determine liberty, status, property and constitutional rights, and its precedents bind persons who never consented to the technology. Consensual machine-made resolution and non-consensual machine-made judgment are different in kind, not merely in degree.

On the constitutional argument above, consent has a further limit in Malaysia. Parties may, within the limits of the law, agree to take a dispute to an algorithm outside the courts. They cannot, by agreement, require a court to hand its judicial function to one.

Malaysia’s proposed AI Governance Bill is intended to govern those who develop and deploy AI systems across the economy.

The judiciary, however, is a separate branch of government, and judicial independence requires it to keep control of how AI assists its work, within the constitution and the general law.

Judges in England and Wales have worked under published guidance on AI since December 2023, most recently updated in October 2025. It requires anything an AI tool provides to be checked before it is relied upon, and holds judges personally responsible for material produced in their name.

It also requires judges to read the underlying documents. An AI summary cannot replace their own engagement with the evidence. In Evans v HMRC [2025] UKFTT 1112 (TC), a tax tribunal judge disclosed that he had used AI, principally to summarise documents. He checked the summaries, used no AI for legal research and kept the weighing of the arguments for himself. That combination of disclosure, verification and responsibility is instructive.

Malaysia was an early mover in practice. How our own courts frame their rules for the generative tools now arriving is, rightly, for the judiciary to decide. The experience in England and Wales suggests only that the questions are the same everywhere. What may the tool do? Who checks it? And who answers for the result?

Minutes may be enough to produce an answer. A judgment is more than an answer. It is a reasoned decision, given by a judge bound by oath and office, which can be read, tested and, where the law allows, appealed.

The important task of the next decade is to take everything the machine offers the courts, from speed to wider access to justice, and to make sure the judge still stands where the constitution placed him, at the centre of the judicial function and never at its sideline.

Tan Sri Zainun Ali is a former judge of the Federal Court of Malaysia, a member of the Judicial Appointments Commission, pro-chancellor of Universiti Malaya and chair of the Financial Lines Forum. J J Chan is a barrister-at-law of the Honourable Society of Gray’s Inn, London; an advocate and solicitor of the High Court of Malaya; an adjunct professor at the Faculty of Law, Universiti Malaya; and founder of the Financial Lines Forum. He specialises in financial lines, medical negligence, corporate and directors’ liability, insurance/reinsurance and AI governance risk. He also writes regularly on constitutional and public-interest matters.

Save by subscribing to us for your print and/or digital copy.

P/S: The Edge is also available on Apple's App Store and Android's Google Play.