Judging AI
What it can do for the courts; and what it can’t
Judges are using artificial intelligence to improve their judgments, a senior member of the judiciary disclosed last week.
Sir Colin Birss, who as chancellor of the High Court is the senior chancery judge and has day-to-day responsibility for the business and property courts, described four ways in which judges were now using AI:
To identify inconsistencies. “Once I have finished writing a judgment,” he said, “I give it to the secure Copilot system on my computer and ask it to identify any internal inconsistencies. It is remarkably effective. What I choose to do with the proposals is up to me. I don’t always agree with the AI but it has been helpful and I have clarified wording in draft judgments as a result.”
To produce anonymised judgments where needed. After a judge has drafted a judgment in the normal way, AI can be asked to suggest redactions. Judges have reported that AI has picked up details that, when read together, can lead to “jigsaw identification” of the individuals concerned. These details can then be edited.
To generate transcripts. “This is very exciting and has potential to make a big difference in all our courts and tribunals,” Birss said.
For administrative tasks. “With the form of Copilot available to leadership judges like me,” he explained, “the ability to find things in emails and files has been transformed. No longer do I need to do word searches on old emails.”
Outlining these advantages in a lecture to the City of London Law Society, Birss stressed that judges must not use public AI systems unless these are known to be secure. But all judges in England and Wales already have access to a secure version of the Microsoft Copilot.
There is a separate secure system developed by HM Courts and Tribunals Service with the Ministry of Justice. Judges were reminded that they must take full personal responsibility for every judgment they deliver.
Turning to the use of AI by unrepresented litigants, Birss said that “quite often the litigant’s case is presented more clearly and coherently than I would have expected in similar circumstances in the past”. If these litigants had taken advice from a legal professional, they would have been entitled to claim legal professional privilege — meaning that communications between lawyer and client would generally be protected from disclosure.
What, though, if a litigant seeks legal advice from a public AI system? Birss thought the advice might not be privileged because these systems appeared not to be confidential.
Confidentiality has always been a requirement if privilege is to apply. However, privilege might still be available if the lawyer looked up the answer on a secure AI system.
Comment
It may seem like the judges are talking about AI all the time. Lord Briggs, a justice of the Supreme Court, said last week he foresaw AI leading to a “tsunami of new claims”. Addressing a conference in London on class actions, he questioned whether the court service had the resources to handle them. On the other hand, he thought, parties to anything other than the most minor cases would not want them to be decided by a machine.
The judge who has spoken most about AI is Sir Geoffrey Vos, who retires ias master of the rolls in the autumn. But Birss was the lead judge on AI until his appointment last November. He’s also thought to be the first senior judge to have included an AI-generated paragraph in a judgment. That was more than two-and-a-half years ago.
We can expect to hear a lot more from the judiciary about the unprecedented impact of AI in the months and years to come.
A part-time judge writes:
"The secure ejudiciary version of Copilot that we have access to as judges is not very adept, probably because they have naturally limited the information it handles and has access to. It is not, for instance, able to find sources of law as easily as the big commercial AI large language models which are paid for or public, but it is not without its uses
The thing that I have used it for, and I’m sure others do too, is for feedback after the fact. So I have asked AI large language models to identify bias, tone, consistency and themes including readability across all my published judgments (I help it via Bailii etc). It is edifying to see when, for instance, it tells you that things might be easily accessible for the public reader, or the use of scaffolding helps a lay person etc."
Many thanks; and important reflections on AL and advice privilege