Ad/Marketing Communication
This legal article/report forms part of my ongoing legal commentary on the use of artificial intelligence within the justice system. It supports my work in teaching, lecturing, and writing about AI and the law and is published to promote my practice. Not legal advice. Not Direct/Public Access. All instructions via clerks at Doughty Street Chambers. This legal article concerns AI Law.
Introduction
How useful is AI Legal Advice? As many readers of this blog will recognise, I believe it is important to observe how senior members of the judiciary in England and Wales are discussing AI and its potential impact on the justice system. On 16 July 2025, Sir Geoffrey Vos, Master of the Rolls and Head of Civil Justice, provided evidence to the House of Lords Constitution Committee during their inquiry into the rule of law. His evidence included notable remarks on the role of artificial intelligence (AI) and ChatGPT within civil justice, as well as the evolving role of lawyers in advisory work
The complete video can be watched here:
Parliament Live TV
The transcript can be read here:
Judiciary UK Transcript
Early Legal Advice, AI and Legal Aid
Lord Beith questioned Sir Geoffrey Vos on methods to encourage the public to seek early legal advice. In his response, Sir Geoffrey Vos highlighted the negative impacts resulting from significant reductions in Legal Aid for civil matters:
“Let me say something about early legal advice. You probably know that the Legal Aid system in civil was reduced very significantly a few years ago, and that has caused problems. There are less lawyers in the County Court than there used to be, and when there are lawyers, they tend to be in a particular kind of case which is funded by industry, say, for example, personal injury cases generally, because of funding mechanisms, have lawyers, not always but sometimes. But a lot of other cases, small claims, don’t frequently have them, and the problem that occurs with needing to go to the judge for that half-hour hearing to be told that you’re claiming for twice as much as you did should be dealt with by a lawyer if it isn’t obvious to the parties. You can’t get Legal Aid funding for that….”
Can AI assist with these problems? Sir Geoffrey Vos shared a personal experience regarding AI providing legal answers:
“…What I envision and what I’ve suggested and what I’m hoping will be picked up by government in a discernible period of time is the provision of legal advice online so that you will, first of all, probably use a legal advice chatbot, an AI-driven tool, which by the way works incredibly well. I’ve tried them myself, and I had a legal problem myself, and I went to a solicitor and I got the answer. Then I thought, ‘Gosh, I’m talking all the time about AI, I think I should just see whether the same answer comes off my ChatGPT.’ And I went into ChatGPT, I asked the question, I got the same answer, but for rather less money…”
With AI, You Must Ask The Right Questions
Regular AI users understand the importance of a good “prompt,” i.e. the question posed to AI to obtain a useful response. Sir Geoffrey Vos emphasised that these prompts are crucial:
“…So you can have such a chatbot, but many people won’t ask the right questions of a chatbot. In many cases, it’s not actually the answer that’s lacking, it’s the question, and although you can train chatbots to get to the right question, sometimes people just are not able to understand how to converse, and those people will need to be transmitted to legal advice. Often, I believe in half an hour, talking on a screen to a real human being, these problems will be resolved before even you get to court, and I think the best use of Legal Aid money for civil justice would be to provide such a service. It’s not there yet, but I know it’s being considered…”
Are Those Who Don’t Understand The Technology Left Behind?
Lord Griffiths raised concerns about individuals struggling to cope with technological advancements, describing himself and others as “Luddites.” He highlighted an effective model implemented in Wales:
“280,000 people… have been helped, with 1.1 million social welfare problems since January 2020.”
Lord Griffiths cautioned against assuming younger people inherently have advanced digital literacy, especially beyond their interests. Sir Geoffrey Vos acknowledged these concerns, emphasising judicial awareness of digital exclusion:
“We have done our own work on that in the Civil Justice Council and we are very concerned to make sure that we are not running away with ourselves in digitising stuff and leaving people behind.”
Nevertheless, Sir Geoffrey Vos pushed back on the assertion that younger individuals lack the capability or motivation to utilise AI seriously:
“I think they are perfectly capable of using ChatGPT extremely effectively, and probably more so than Google and do, and even use it to see how they can protect, vindicate their legal rights.”
He stressed his job as Head of Civil Justice was to protect the rights of all, not to select people we think are worthy of being protected. The role of judges remains paramount:
“…So I am concerned to ensure that we digitise as much as we can because it provides a quicker, more efficient, more effective and at more proportionate cost resolution of people’s disputes. But where that is not just, where that is not appropriate, then we should take other steps. You know, I have not come here and said, “I want to abolish judges, I want to abolish courts, I want to have a situation where people can never see a judge.” Quite the reverse. Judges are essential to the system that I’m seeking to create. But what I do want is a quicker system because speed of resolution will create massive advantages for our economy. Speed and fairness. Justice. That’s why I went into this job…”
AI in Judicial Decision-Making: What about AI Judges?
Baroness Hamwee queried whether Sir Geoffrey Vos distinguished between administrative and substantive judicial roles for AI. He clarified that this distinction wasn’t straightforward:
“I think that would be a little oversimplified. I think there isn’t such a bright line distinction, and even if there were, you would quickly cross the line into using it for substantive purposes. I think the use of AI is unbelievably complex in the law. I think it is actually less complex in some other areas. It is perfectly obvious that if you’re an engineer and you can design, do the calculations for your bridge by using AI, then you will learn from experience whether you need to check them or whether they are reliable, and if they’re reliable, what you need to do before you build the bridge to make sure that it doesn’t fall down.”
However, he noted:
“But in justice, it’s rather different. Now, we can use AI as lawyers to get an answer to the problem, “Can I defend the possession claim brought by my landlord because the house has got mould?” We can put that in into ChatGPT and it’ll give you a generic answer then you can give it more facts and probably if it’s trained on legal materials, it will give you quite a good answer. Even legal advice which will be obtained by AI is in a different category from legal decision-making.”
Regarding AI in judicial decision making:
“Now non-legal decision-making is used an awful lot already so I’m sure our pensions, our benefits are calculated using an AI programme already and nobody much objects to that, but again when you come to judicial decision-making I think people would be very surprised to hear that decisions were ever made by an AI programme instead of a judge. Now I think this throws up a number of philosophical problems and also legal problems. The first legal problem is whether, if ever, and I emphasise it has not happened and is not proposed to happen, a decision were ever taken by a machine rather than a judge, would that be compliant with Article 6 of the European Convention on Human Rights, which requires the state to provide a decision by an independent and impartial tribunal.”
“Is a machine ever capable of being an independent and impartial tribunal? Discuss.”
In his concluding remarks, Sir Geoffrey Vos posed the above question, adding:
“….That’s the first question for homework. Now, so when you say, “Well, is there a difference between substantive and administrative?” there really is such a difference, but in different fields, it’s a different concept.”
Comment
These comments continue the positive approach towards AI that appears to have been adopted by senior members of the judiciary in England and Wales, reflecting their belief that AI may provide improved access to justice for those currently underserved. For those interested in previous comments from senior members of the judiciary please see:
AI Can Provide ‘Reasonably Accurate’ Legal Advice, Dedicated tools in Employment and Property Law, Says Master of the Rolls—But Is That Enough to Replace Lawyers?”
Lady Carr on AI in Legal Education and Judicial Practice: Preparing Lawyers for the Future
Sir Geoffrey Vos’s reflections on the consequences of reductions in Legal Aid are particularly significant. Personally, I agree that the justice system has suffered as a result, leaving many without the representation they deserve. For my part, I would welcome the restoration of Legal Aid to its former, more comprehensive state.
Nevertheless, I’m afraid I do have reservations about relying on AI chatbots to bridge this gap, concerns I intend to address in greater detail in a future post. For now, I suggest that before any such step is taken, we must carefully consider essential issues including: who controls the chatbot, how biases will be managed, risks of hallucinations, vulnerabilities to hacking, deceptive behaviour and inherent limitations or inaccuracies. All these factors must be thoroughly evaluated before deploying AI legal assistance to the public.
I am also curious whether any members of the senior judiciary have voiced serious concerns about the direction we are heading with AI’s potential mass adoption, particularly in substituting significant elements of Legal Aid/Legal Advice, and whether they doubt AI’s capacity to fill this crucial role. If anyone is aware of such concerns expressed by senior judiciary members, I would greatly appreciate seeing them.
I found Sir Geoffrey Vos’s personal story intriguing, though I would advise readers, especially those without legal training, to approach it with caution and in the context it was given. I would not suggest that ChatGPT is a viable replacement for professional legal advice solely because it offered the correct answer in this instance. In my experience, no AI models currently available, including those I regularly use, which arguably are among the best, are yet reliable substitutes for comprehensive legal advice. The risks include not only the well-documented hallucination issues but also AI’s tendency to provide pleasing rather than accurate responses, occasionally contradicting well-informed legal analysis.
Therefore, users should approach AI-generated legal advice with extreme caution, given its propensity for inaccuracies and false information, something I actively monitor and document here.
Remember also that ChatGPT offers various models depending on subscription levels. It would indeed be impressive if the free version consistently provided accurate legal answers, although the specific prompt used is unknown. In my own experience, the free version tends to be more superficial and less effective at addressing complex legal questions, whereas advanced models such as GPT-3o Pro perform considerably better, albeit at a cost that may be prohibitively expensive for many users.
Finally, the big question for those considering AI Judges: “Is a machine ever capable of being an independent and impartial tribunal? Discuss.”
I have discussed these issues in previous articles on AI Judges and whether AI could ever claim Human Rights. Some jurisdictions have begun to explore AI Judges and so this questions is highly relevant. My position evolves regularly as I interact with new models and observe advancements in technology. Currently, my inclination is towards answering ‘no’: machines are not yet capable of being truly independent and impartial tribunals. However, one could reasonably counter with:
“But is a human ever truly capable of being a completely independent and impartial tribunal? Discuss.”
What do you think? Is it possible? If you’ve found this article useful please consider subscribing to my Substack newsletter, where I regularly share broader legal commentary. Many of you regularly read these articles, which is great, but comparatively few subscribe, so your support would be appreciated. Subscribe here.
