Trust, Traceability, and the Future of AI in Medico-Legal Practice: Insights from Tom Davies

AI is already in the room. The question is whether it's being used well or not.

In this episode of Medico Legal Mastery, host Jess Marshall sat down with Tom Davies - solicitor, medico legal technologist - to examine what artificial intelligence actually looks like in practice, why the trust problem is everything, and what the courts are trying to figure out before the technology runs ahead of them.

Why AI is becoming essential in medico-legal practice

Here's a thing the industry doesn't say out loud: asking an expert to read 800 pages before they walk into an examination room - and then retain all of it - is an unreasonable ask.

"It is a really tall order to ask even an extremely intelligent, high focused person to slog through that much information - and we pretend like they're able to retain it all."

AI changes that equation. But only if it's built correctly.

Trust and traceability: the foundation of responsible AI

The trust layer is everything.

The concern most experts raise first is hallucination - the well-documented tendency of AI models to produce confident-sounding information that isn't accurate.

Tom’s central argument is that the value of AI in medico legal isn't in getting answers. It's in being able to trace those answers back to the source.

"It's being able to make sure that the inferences you're getting out of the tools you're using are traceable back to the source material. That's the big challenge - and one of the big achievements."

Without that traceability, you have a black box. And a black box is useless to an expert who needs to stand behind their opinion in cross-examination. With it, you have a tool that doesn't replace the expert's judgment - it feeds it, clearly and verifiably.

Using AI to identify patterns and conflicting evidence

Tom describes the moment the penny dropped for a sceptical expert he was demonstrating the technology to. Summarising documents, building an index, splitting a PDF bundle - useful, but not revelatory. Then he asked it something different.

"Which of these prior medical reports disagree with each other?"

That was the moment. Not just less to read. Not just organised information. But a tool actively showing the expert where the tension sits - which reports are in conflict, where the evidence pulls in different directions - so the expert can decide what that means for their opinion.

"It shows you what rocks to look under. And then you can dig further down, and understand how one part of the corpus relates to another - where they disagree, where they do agree."

The expert's judgment about what that means stays exactly where it should be: with the expert.

AI and medico-legal report writing

The report-writing side of medico legal carries its own friction - and AI is just as useful there, perhaps more visibly so.

Dictation has been a pain point for as long as medico legal has existed. Machine transcription has existed for a decade. The problem has never been getting a verbatim record of what was said. The problem has been turning what was said into what was meant to appear on the page.

"We're now able, with good structure, to parse that in a way analogous to what a transcriptionist actually does - to put what the expert wants on the page, from the dictation they give. That's the sort of thing that, once I show people, they never want to go back."

The human review loop stays. The speed of getting to that loop changes dramatically.

There's a framing problem with any conversation about AI in medico legal. It's easy to hear "using AI to process the brief" as code for "not doing the work." Tom Davies pushes back on that framing directly.

"If you can hone in on what's actually relevant to what you're being asked to do in that 800-page file, that's not cutting corners so much as it is just being precise."

He's equally clear about the risk of getting the intent wrong. AI makes it easier to do what you instruct it to do. If the goal is speed at the expense of accuracy, you'll get something sloppy. If the goal is effectiveness - giving the court the most useful, most defensible opinion possible - you're on firm ground.

"If you instruct them to do something sloppy, you'll end up with something sloppy. If you come at it from an angle of wanting to make yourself more effective in putting what matters onto the page - you put yourself on much firmer ground."

Court guidelines and the future of AI in expert evidence

Dr Anna Manolopoulos, a guest from episode 5, put a question to Tom that goes to the heart of the legitimacy debate: how does this technology sit with the existing court guidelines?

The answer is more nuanced than the headlines suggest. New South Wales issued a practice direction in early 2025 that prohibits AI from generating the content of an expert's opinion - but it carves out significant exceptions for transcription, formatting, and getting the expert's opinion onto the page. Victoria has taken a lighter-touch approach: disclose the use of AI, but your evidence remains your evidence.

"The opinion should always be the expert's. I think that's what they were really driving at."

Tom's read is that the courts - including New South Wales - will eventually converge on what he calls the common-sense view: it's your name on it, you have to own it, and how you organise the information that forms the basis of your opinion is largely your business. The analogy he draws is to practice managers and assistants who already perform organisational tasks for experts without anyone suggesting that compromises independence.

"It is already getting to a point where it's too useful to deny - and too useful to everyone - for any one party to say, 'no, stop, you just can't use this."

The future of AI-assisted expert practice

Tom's description of where this ends up is worth sitting with. An AI system that has learned - over years - how a specific expert works. That understands their preferences without being told explicitly. That presents the expert, when the examination is over, with exactly what they need to form a complete, concise, defensible opinion. And that provides a clear audit trail back to the

source material at every step.

"It's a system that understands how the expert wants to work - better than the expert does themselves."

That sounds confronting. He acknowledges that. But his point is that it's also freeing - because it means the expert's attention can stay entirely on the application of their expertise, rather than on the machinery of how that expertise gets transmitted.

Listen to the full interview with Tom Davies here, or find Medico-Legal Mastery wherever you get your podcasts.