Powered by
MOMENTUM
MEDIA
Ahead of his appearance at this week’s Australian Law Forum, a BigLaw partner unpacked lawyers’ biggest blind spots with AI use, stressed the importance of maintaining audit trails, and reflected on the place for both creators and verifiers.
Later this week, Dalvin Chien – a partner in information and communications technology and digital law at national firm Mills Oakley – will speak at the Australian Law Forum, on a panel discussion called, “AI accountability: Preserving legal integrity”, which will explore how firms are defining, operationalising, and enforcing verification standards to ensure evidentiary accuracy, ethical compliance, and client trust in a hybrid human-machine environment.
In conversation with Lawyers Weekly ahead of that panel session, Chien discussed what he sees as the biggest pattern failures when lawyers are working with AI-generated content.
Firstly, he said, lawyers tend to treat AI outputs as a finished draft: “You address this by basically treating AI as flawed research. That is, you treat AI output with a high degree of constructive scepticism.”
Another blind spot is what Chien called the “seduction of good legal writing”, noting that after being asked to put together something that is practical and addresses the client’s problems, lawyers will use AI that will “spit out brilliant and eloquent prose”.
“The problem underlying that brilliance is terrible legal reasoning,” he said.
“You address this by basically realising that AI frequently struggles with analogical analysis. It struggles with ambiguity. And, you also deal with this by exercising your own judgement by checking what AI does,” the outputs for which he said is the final blind spot to be on top of.
“We have a world of protected data that constrains what we do, be that private information, proprietary information, or confidential information. To the extent we put that information into an AI algorithm, and we don’t discern what that algorithm does with that information, we could be non-compliant. It’s a significant risk,” he said.
On the question of the need to maintain proper audit trails to avoid reputation risk when AI is being used for client matters, Chien said that audit trails involve identifying what goes into the AI algorithm, and also identifying what version of the AI algorithm is being used.
Lawyers should resist the temptation to use AI to check AI, he stressed: “You create a significantly unhelpful pyramid scheme if you did that on disclosure.”
Law firms, Chien said, should just be upfront to the extent they’re using AI: “We should be upfront with our clients. This is the AI that we’re using. This is how we’re using it.”
Given the range of protected information at play, lawyers should be transparent about what information is being used, and obtain consent when possible. Reputation risk looms large, he said, and with this in mind, what will be really damaging is if a client informs you that one of your lawyers is using AI, but you’re not aware of that fact. Shadow AI, he said, is a big concern. “You should eliminate it by basically being permissive if you can, but also controlling what is used.”
Elsewhere, Chien said he sees a place for both creators and verifiers, in the context of lawyers’ evolving roles in the age of AI – given the need for effective governance and accountability structures.
This said, there is “no doubt” that different skill sets will be required for both roles, he added.
“For a creator, for example, they’re no longer working through a black piece of paper. Generative AI gives you advanced precedent creation capabilities. Creators need to get familiar with prompt engineering. The quality of AI outputs depends almost entirely on the input. Lawyers and teams must write structured context-rich prompts using framework techniques such as assigning persona, defining constraints and specifying, for example, the exact jurisdiction procedural posture,” he said.
“Creators also need to combat cognitive atrophy, which means muscle memory, or training your muscle to the point of exhaustion, which then creates new muscle.”
In the legal context, Chien said, this means that lawyers who now no longer need to draft from scratch risk losing the muscle memory of legal reasoning. Lawyers need to hone their skills through active criticism and auditing of the work that they generate with AI, he said.
“Finally, creators need to focus on legal architecture and strategy: with the time saved on drafting, you really need to put that time and effort into crafting a formula and strategy.”
The Australian Law Forum is being held on Thursday, 30 July, at Royal Randwick in Sydney. To view the full agenda, click here. To view the full speaker line-up, click here. To secure your ticket for the day, click here.
Jerome Doraisamy is the managing editor of professional services (including Lawyers Weekly, HR Leader, Accountants Daily, and Accounting Times). He is also the author of The Wellness Doctrines book series, an admitted solicitor in New South Wales, and a board director of the Minds Count Foundation.
You can email Jerome at:
