How to judge slop

Connectez-vous pour écouter l'article

B.C. Supreme Court Justice David Masuhara may have been the first judge in Canada to encounter AI-generated fake citations. As he wrote at the time, “generative AI is still no substitute for the professional expertise that the justice system requires of lawyers.”

Two years later, the phenomenon certainly hasn’t gone away. So, there could hardly be a better guest for a discussion about the frequency of hallucinated citations, strategies for identifying them, and ways of dealing with lawyers and self-represented litigants who rely on them in court.

Check out these CBA resources designed to help lawyers use AI responsibly: 

• AI Academy
• AI In Practice

Verdicts & Voices is a legal current affairs podcast presented by the Canadian Bar Association. With her retinue of expert guests, host Alison Crawford keeps listeners up to date on news, views, and stories about the law and the justice system in Canada.

Views expressed are not necessarily those of the CBA.

Transcript

Welcome back to Verdicts & Voices, your weekly legal affairs podcast from the Canadian Bar Association. My name is Alison Crawford and I'm your host. In one of our earliest episodes, I spoke to AI expert and University of Ottawa law professor Amy Salyzyn about the very first reports of generative AI hallucinated case law and citations turning up in Canada's courts.

One things that’s interesting is when we had these first cases pop up, in the United States in particular, about six months after Chat GPT was released, there was certainly a dominant narrative that this would burn out, that once you have a lawyer on the front page of the New York Times, which the lawyer was, for doing this wrong, the message would get out and people would know not to do it. And we’ve seen almost the opposite happen. The person who’s tracking this internationally – there’s someone who’s keeping a database – feels that it’s actually increasing rather than decreasing.

Canadian legal tech company CourtReady is tracking data here in Canada. And as of April 7th, its live database had logged 247 fictitious citations. The information suggests AI submission happens most at the Federal Court of Canada, followed by courts in Ontario and British Columbia. CourtReady also tracks how judges are responding once they detect these fake cases and citations.

Joining me now to share his thoughtful perspective is BC Supreme Court Justice David Masuhara. Welcome to the show, Your Honour.

David Masuhara

Good morning and thank you very much.

Alison

It's widely thought that you were the first Canadian judge to have detected AI hallucinations in a case that got a lot of media attention in February 2024. Can you tell me what that was like for you to spot this and figure out what it was?

David Masuhara

Well, it was spotted because counsel had actually raised, initially, the issue and it came to the attention of the court because some of those issues were itemized in the materials that were presented. And thereafter, obviously it was still early days, and the concern was with respect to where is this all going to go and some early intervention, I think, required some comment.

Alison

And comment you did. In your decision, you said, competence in the selection and use of any technology tools, including those powered by AI, is critical. So, you know, as you know yourself, the use of generative AI, you know, such as chat GPT, has only accelerated since then. What does that mean for you as a judge?

David Masuhara

I think there's a couple of aspects of that. The first is, in court, just being cautious about the presentations being provided to the court by council. And in many cases now, in the majority of these cases that are identifying difficulties, are from those who are self-represented.

And with respect to self-represented people, that's the, you know, they are not trained and they don't have sort of the legal professional obligations of what they present to the court. And I think the concern really is just what are they presenting to the court, beyond just the discipline that would be required of a professional.

And with respect to the lawyers that appear, that's concerning in so far as how much confidence can we place in counsel, counsel that you may not know, may have seen for the first time, and the materials that they're handing up to the court with respect to whatever position they're advancing.

Alison

Is there a way that you now have had to change how you approach your work because of your concerns about AI?

David Masuhara

I think one of the things that is a quick check for the courts that I use is asking for lists of their documents, one, and two, asking particularly for the case law that they're handing up, so that there's a complete set of decisions as opposed to just simply the name of the case and the citation. Because the problem that we see is the fake cases are fake names, fake citations. And if they actually have to go to the second stage of actually finding the case and producing the entire decision, I think that goes some distance in terms of verifying the source.

Alison

Yeah. Now, many courts have something online where lawyers and self-reps have to certify that they've verified all of their sources before submitting them to the courts. Is that something that exists at your court as well?

David Masuhara

No, we don't have that directive that's been issued from our court, but I do see the benefit of that type of directive coming from various other courts.

Alison

How many more cases have you had where you've detected AI? Like, hallucinations.

David Masuhara

I've had two further cases. One from an in-person, and fortunately, as I mentioned, there was a lawyer on the other side who was able to identify that the 20 cases that were identified in a polished submission were AI-generated because they couldn't find the cases.

Alison

All 20?

David Masuhara

Except for one.

Alison

Oh my goodness.

David Masuhara

It was a judicial review. And the one case that was accurate was that one decision, of course, that kicked off the whole reason for the judicial review. The other one was just a little bit more odd in that the council had prepared for final closing, and there was a real difference between the delivery of that final closing from the person that I had observed throughout the case. And a question was raised and it didn't affect the outcome because obviously it was a submission that was being made, and somebody was making some points that just seemed quite unaligned with the personality that I had observed throughout the case.

Alison

So, how much extra work is this creating for judges? Because once you detect this, you have to deal with it.

David Masuhara

It's hard to say because sometimes you can deal with it fairly quickly, where you can just point it out and say this doesn't appear to be real. I think where it becomes an issue that takes some time is the fact that we have to approach these things a bit more cautiously now. It's taking more time in terms of thinking, okay, does this make sense? Where's the source for this? I haven't heard this line of argument before. Is this a new principle? Is this a novel new principle? And where is it coming from? And sometimes the arguments are, oh no, this is well developed. But then you have to dig a little deeper into the validity of the argument.

Alison

Yeah, and A.I. is getting more sophisticated as time goes on.

David Masuhara

That's true. Yes.

Alison

When it comes to consequences for using AI, generative AI, that turns out to be hallucinations, I know there was—some judges are fining even self-representatives. There was a Quebec Superior Court judge who fined a self-rep $5,000. What do you feel should happen to lawyers and self-reps who are submitting hallucinated material?

David Masuhara

Well, that's an individualized process, as you know, Alison, and each case has to be considered in its own circumstances. I think for in-persons, some are very sophisticated, and I've seen arguments and presentations that are very polished. So, it's not as though they don't understand the requirements in the truth-seeking function. I think the same standard would apply then if they are highly sophisticated as it would to a trained lawyer with respect to some of these problems and deficiencies. Financial penalties, I think, are now becoming common with respect to penalizing people who are putting fake cases and things like that before the court.

Alison

What about for a lawyer, though, where clients and the public have a higher expectation that they're competent and that they are, you know, if they certify that they verified their sources? I mean, how should that be treated? I know you say it'll be different by every…

David Masuhara

No, but I think people who are lawyers who are verifying it in accordance with the requirements of their particular court or their particular jurisdiction in terms of professional responsibilities, are exposed and should be disciplined in accordance with the level of the divergence from the expected norm.

Alison

And who should be responsible for that? Is that something the courts need to do or law societies? Where do the consequences—where should they originate?

David Masuhara

Yeah, that's a good question. I think there's a dual role here. I think professional competency and responsibility are the jurisdiction of the legal profession’s regulator. But when things are happening before the courts, we also have to look at the implication of that type of deficient conduct and its effect on the process of the court. The court is obviously concerned with the administration of justice and the truth-seeking function. And so to the extent that this is impairing those important aspects, the courts do have the rules in order to express its displeasure with the conduct.

Alison

Is there anything else you think the courts could be doing, in addition to perhaps using that certification process, which people are flouting anyways? So what other measures do you think could be taken?

David Masuhara

Well, I think the courts are exercising its approach in the time-honoured approach of observing and so incrementally, I think the courts can still, the leadership of the courts can obviously continue to encourage the profession to seek further training, and to express its concerns regarding the integrity of the court system.

Alison

Yeah. Last month, the Quebec newspaper La Presse reported on a judgment that was made at the Quebec Superior Court that included fake Supreme Court of Canada case law. That would be the first case that we know of of a judge citing AI slop. What is the process for dealing with something like that?

David Masuhara

That gave me some pause when I heard about that case. The way that I think that that would normally be dealt with is, of course, through the leadership of the individual court that's involved. The Chief Justice obviously would be involved. And I think the other step may well be that judges are under the jurisdiction of the Canadian Judicial Council and therefore whatever steps are taken there may occur. And that happens to judges, you know, not just in AI course, in all other realms of professional conduct.

Alison

What are your greatest concerns about where this AI creep is going to take our justice system?

David Masuhara

Yeah, I'm glad you said the word creep. You know, it's something that is not sort of a… We talk about technology being disruptive and causing things to be turned upside down. I think the real concern more is, as you said, the word creep. And it's something that, in some earlier work that I had done, was to identify the general problem, sort of, I would say, existential consequences, when we undertake some of the technology so-called benefits, things such as... And this is probably apparent in terms of why we see some of these continuing problems with lawyers. And as you've identified, there are cases now where members of the judiciary have gotten into circumstances where hallucinations have entered their cases and decisions. And it's the passivity that sometimes is one of the consequences when you start adopting wholesale certain technological processes. That then leads to decreased agency in terms of our own processes and a sense of decreased responsibility because we've handed over some of the controls over our own processes and it lessens our culpability for consequences, negative or positive.

Other aspects are sometimes becoming increasingly ignorant of our understanding of how our processes work through the delegation of the tests. Detachment, of course. Our dependency can result in what I call de-skilling. We forget how to perform a task or become less capable of doing it. And we can lose our motivation to increase our knowledge and skill. And so that's the kind of way I've been looking at the impact and consequence of AI upon the judiciary. And I know we have guidelines and things that talk about the bright line of the difference between decision making and the human in the loop, etc. But it's the creep of some of these new technologies that begin to find their way and become embedded in our own processes to the point where, just as I've mentioned, these various consequences can occur. And I think we need to be looking out into the future of not just sort of looking at bright lines, but saying, what is the nature of, what is our courts going to look like into the future, you know, five, 10, 15 years out?

Alison

Yeah. What you painted there with the de-skilling, that kind of scares the hell out of me.

David Masuhara

Yes, that is a real problem. That has an impact then on the critical aspects of integrity, public confidence, all of those things that go into the judiciary in exercising its truth-seeking function. And it could be that some of these things can enhance the productivity and efficiency. But as you mentioned, there is a certain cost that comes with doing these things and we need to identify that and to recognize how that might impact our judicial function going forward.

Alison

Well, you've given us an awful lot to think about. Justice Masahara, it's been a privilege to have you on the show to share your experiences and your expertise. Thank you so much.

David Masuhara

Thank you very much. It's been a pleasure.

Alison

Justice David Masahara has been a member of the BC Supreme Court since 2002. He has also served as chair of the Joint Courts Technology Committee, especially during the pandemic.

And that is a wrap for this episode of Verdicts & Voices. This weekly legal affairs podcast is produced by the Canadian Bar Association without any artificial intelligence. So any mistakes are mine. I'm your host, Alison Crawford. See you next week.