As artificial intelligence becomes more capable of reading records, checking citations, drafting memos, and helping people navigate legal information, courts face a critical question: how can judges use AI responsibly without delegating the human judgment at the center of their role?
In this episode of AI and the Future of Law, Jen Leonard and Bridget McCormack are joined by Justice Jerod Tufte of the North Dakota Supreme Court for a conversation about AI, judicial workflows, public legal data, access to justice, and the future of courts. Drawing on his background in computer engineering and law, Justice Tufte explains how he uses AI tools to support legal analysis while preserving accountability, independent judgment, and professional responsibility.
Justice Tufte discusses his work building an open-source model context protocol, or MCP, for North Dakota’s legal materials, and why reliable legal data could help courts, lawyers, and self-represented litigants use AI more safely and effectively. Throughout the episode, Justice Tufte emphasizes that AI should not replace judicial responsibility. Instead, courts need trusted intermediaries, reliable data, clear guardrails, and hands-on experimentation so judges can understand how these tools work and use them well.
Key Takeaways
● Judges can use AI to support legal analysis, but they must preserve independent human judgment.
● AI tools can assist with records, citations, bench memos, and issue spotting when grounded in reliable source material.
● Courts need guardrails to prevent overdelegation and ensure judges remain accountable for their decisions.
● Judges and courts need practical experience with AI, not just theoretical training.
Final Thoughts
This episode offers a practical look at how courts can engage with AI without compromising human judgment, accountability, or public trust. Justice Tufte’s perspective shows both the promise and the responsibility of bringing AI into court systems.
Transcript
Intro + AI Aha!
Jen Leonard: Hello, everyone, and welcome to AI and the Future of Law, the podcast where we explore all of the rapidly changing dimensions of artificial intelligence and consider what those developments might mean for legal professionals. I’m your co-host, Jen Leonard, founder of Creative Lawyers, here as always with Bridget McCormack, president and CEO of the American Arbitration Association.
And we are delighted today to be joined by a forward thinker from the bench, Justice Jerod Tufte from the North Dakota Supreme Court.
We’re really excited to explore all the different ways that Justice Tufte is using AI, sharing what he’s learning from AI, and thinking about the future of judicial engagement with society in a world that is augmented by AI.
Welcome to the podcast, Justice Tufte.
Justice Jerod Tufte: Thank you, Jen and Bridget. I’m glad to be on and looking forward to our conversation.
Jen Leonard: Excellent. As you may know, we start every episode with our AI Aha! segment because we love to hear the different ways that our guests are using AI.
One of the things we hear most frequently is that other people are inspired by the AI Aha! moments of our guests. So we’re curious, Justice, how have you been using AI in your personal or professional life in a way that you find particularly interesting?
Justice Jerod Tufte: It surprises me that I still have these kinds of moments almost every week. Because I knew this question was coming, I thought about something that hit me recently outside the work context.
I’m back in marathon training. I’ve had a few months off. I’ve been a runner for a long time, so I have a lot of devices and gadgets that gather data and such.
I built a little tool that pulls all the Garmin data and gives it to Claude so that it has the whole context and can be a running coach and give me a review of how things are going. That’s all pretty basic. A lot of these platforms are building in AI assistants.
But a couple of weeks ago, it looked at a long run that I had done and noticed that my heart rate and my pace were drifting apart by about 9% in the second half. It was using a lot of really detailed metrics. Taken at face value, something like that would say your aerobic fitness is really not in a good spot to be running marathons.
But what Claude did next was the Aha! moment. It said, “Wait a minute. This is on a trail. It’s in the heat. There are a few hundred feet of climbing in this. You’re running this with a maximum heart rate, so every time your heart rate hits this limit, you’re going to be backing off. So this isn’t necessarily a fitness problem that we see in this deviation in the second half.”
Based on my own looking at that after I got Claude’s input, and then how the next couple of runs went, I think Claude was right.
The part that struck me is that I have standard instructions that tell Claude to push back and to label its assessments with how confident it is. Tell me why it might be wrong or what the other argument is. And that’s exactly what it did here.
To see that in a fitness context or a mundane context, rather than the usual “I ask a legal question and it gives me the other lawyer’s argument,” was interesting. This is, “I ask a running coach a question, and it gives me the, ‘Yeah, I might be wrong if this or that or the other thing was different.’”
I love to see that sort of interaction with these AI tools in the professional context, but when it catches you in a different context, it’s just sort of, wow. This can even evaluate whether my run on Tuesday was a good run.
Bridget McCormack: Just this morning, Strava gave me my monthly statistics, and I was particularly proud of them because I had cycled a lot, and I thought it was going to look really good. I fed them to Claude and asked for specific feedback, and Claude was not happy with my performance in June.
I’m a little bummed out. I wish I hadn’t asked, but it’s kind of amazing. I did learn a few things that I will actually be able to address while I still have time and the weather is still good enough to ride faster and better.
Jen Leonard: It’s usually so sycophantic.
Bridget McCormack: Exactly. I think those standard instructions, which I hope we’ll get to hear about a little, are really interesting. I love the idea of always asking it to tell you the other side. I didn’t, and it really was like, “I have to level with you. This is not great. You really could probably do better.”
I was like, “Oh, okay. I also worked a lot in June, for the record. I want you to know that.”
Jen Leonard: It should have taken that into account.
AI in Judicial Workflows and Guardrails
Jen Leonard: We wanted to kick off our conversation by looking backward to your training, Justice. You have an unusual background for a Supreme Court justice. You trained as a computer engineer before you became a lawyer and then a judge.
As we start to talk to you about AI, we’re curious how your dual backgrounds in engineering and legal analysis help you see things differently from the bench.
Justice Jerod Tufte: Going into legal education from a science and engineering background, the thing that strikes you first is that, whereas in engineering you can get clear, decisive answers, in law you generally get qualified answers with hedges. “It depends” is the standard lawyer answer to everything.
The idea that you have statutory codes that, even if they’re well crafted, don’t give you the answer to every problem — that is a shift. The analytical background, though, even in a domain that does have clear answers, still applies and is quite helpful for the vast majority of legal questions.
The way of focusing on precision and communication helps a lot.
I had gotten away from writing a lot of computer code in the last several years, although I still did a little bit here and there. With these new AI tools that can write code so well for you, if you have enough of a conceptual idea of how computer code works, that’s about all you need anymore in order to build your own tools for whatever problem you might have in front of you.
It seems like the real core value for a lot of these is wading through just mountains of data. Of course, that’s what we have. My court has about 20,000 opinions over its history. To be able to sift through those and find insights using these tools — that’s where the unique combination comes in.
Having enough conceptual knowledge of how the tools work in order to steer them is important in a way that nontechnical people sometimes don’t have. They don’t understand what’s going on behind the hood, so they don’t quite see the failure modes.
Bridget McCormack: I had the pleasure of hearing Justice Tufte present to a number of judges and justices from other state supreme courts and from some of the federal courts at a fellowship, and talk about what he’s built with these tools for his own workflows.
Would you describe a little bit how you’re using AI tools for your workflows in your chambers, and what benefits you’re seeing as a result?
Justice Jerod Tufte: Anybody who’s worked with a law clerk in a judicial setting, or with a brand-new associate or even a summer associate at a law firm, can think about and interact with these tools as if it’s a junior lawyer.
If you outline tasks in the way you would to a junior lawyer who has basic legal training but maybe doesn’t know a lot of the experiential knowledge that you’d have in a particular domain, you can, not 100%, but largely get the same kind of work product you can get out of young, inexperienced lawyers.
In my workflow, part of it might be: we’ve got a record with 900 documents in it, and I need to know how one party’s arguments changed over the course of a case that had two different appeals to my court. I need to know something that’s just hard to find in hundreds of pages of transcript. You can get those kinds of answers if you specify them.
But that’s really simple stuff. What you can do if you specify detailed workflows is get high-quality bench memos that check every legal citation in the briefs on appeal, check every fact against a voluminous PDF record, and give you additional cases that the parties missed. They can really do everything that you’d expect a high-quality law clerk to do in a memo evaluating an appeal.
I started doing that kind of work with the addition of having an editing tool — a very experienced lawyer editing. You specify, as you would with a checklist for anybody in your office, “These are the things that we have to check every single time before something leaves my office.”
By and large, my experience is that the way I’ve specified these things, with the addition that I require it to access the full text of any legal authority that it refers to in the output — if it can’t get it from its own database or from my court’s website, it’s not allowed to cite it to me.
That really seems to control the hallucination problem, because it’s not going on its training data. It’s based on specific documents that it has accessed and made available, and then it gives me the links so I can check it myself.
Jen Leonard: That really leads into a question that we hear almost constantly when we do presentations to legal audiences, which is: if AI is increasingly capable of doing the things that you’re describing — reading the record, checking citations, drafting the memo — how does the next generation of judges and clerks and new lawyers build a muscle to be able to understand the underlying architecture involved?
You said some of the discipline is built against losing or atrophying that muscle, but I’m curious how you’re thinking about what that means to be built against that lessening of the muscle.
Justice Jerod Tufte: I think that’s not just our profession that’s struggling with that. Engineers, op-ed writers, reporters, and almost every knowledge-worker field are struggling with this.
How do you get enough reps to learn the muscle memory, to learn the experience, to learn how to steer these tools? If you start out by steering tools, you need to have some intuition.
I do a couple of things that help. I wouldn’t say I have the whole answer for how we’re going to bring up the next generation of lawyers and get them the right experience.
Certainly some of it is that you have to do a few of them yourself and do the struggle. Until you struggle with working your way through a legal research question or a math problem or any other intellectual task, it doesn’t really sink in in the same way. Then you don’t know how to spot errors in the tool’s outputs.
While you could have these tools — Claude is the one I use — make recommendations for how to dispose of an appeal, I expressly tell it never to do that.
In my process and the way I run my office, you have to come to your own conclusion about what the dispositive issue is first. You have to have your own sense of the case from looking at the basic materials first. Then you can compare it to what Claude thinks about what the strongest authorities are, who’s got the stronger argument, or what the counterargument is.
I think that reduces the amount of atrophy. If I just let Claude make a recommendation and say, “That sounds right. Maybe I’ll save myself the next four hours reading briefs and checking citations,” humans are tempted to take the shortcut. I think we all are.
So you have to build in habits that discourage you from taking those shortcuts.
Bridget McCormack: Do your clerks use the tools you’ve built? Do you have rules about how they can use them and when they can use them? How do you train your own clerks?
Justice Jerod Tufte: I think that’s something that my colleagues on the bench who aren’t excited about using these tools themselves need to be aware of: their law clerks and staff attorneys will be using these tools, and we need to have oversight over appropriate and inappropriate uses.
Ultimately, when you put your name on the bottom of a court filing or a judicial opinion, that’s your responsibility. You need to have checked everything.
If you trust your staff attorney to have checked some of the routine standard-of-review citations that you’ve already read 20 times in your career, that might be okay for some of those cases. But if you worry that your staff attorney might not have checked Claude’s output, that’s a major black eye for our profession when people file things that they clearly haven’t read themselves.
So yes, I do have guardrails on what the lawyers who work for me are allowed to do.
I have the unique advantage of having written the tools. If they use my approved tools, I know that there are lots of guardrails in place.
My colleagues that I’ve shared some of this with get excited about what they can see that I’m able to do. I think there’s also a little bit of a fear of keeping up with people who are closer to the cutting edge of deploying these tools. They recognize that they need to do some more learning and adoption of some of these tools in the very near future.
Jen Leonard: Justice, in addition to making the argument that a judge’s signature means that they have verified the accuracy of everything in the opinion, are there other arguments that you make to skeptical or reluctant judges who aren’t yet using AI?
Justice Jerod Tufte: Part of it is the systemic concern.
What we used to see in the past is that, in particular, self-represented litigants would get the door closed on them because they couldn’t figure out the procedural rules to put together a filing that could get past the clerk’s office and into the merits-evaluation part of the court process.
These AI tools are getting really pretty good at getting past those early gateways, so the courts have to do more on the merits decisions. I’ve seen some preliminary data on this, and I think it’s only going to get worse.
We’ve got a flood of additional litigation coming at us as the costs to litigate go down, as people are able to do more without lawyers to help them, and as lawyers are able to take lower-dollar cases that they couldn’t have done economically before.
That’s my warning message to my colleagues on the bench: you need to be ready to have similar assistance to what the litigants have in order to handle the volume that’s coming our way.
That’s a little bit of a longer-term warning to my colleagues. I think they already see some of the early signs of that increased litigation volume coming our way.
We can hope that a lot more of that goes to the arbitrators and to the alternative dispute resolution folks, but a lot of it is going to end up in the courts.
Bridget McCormack: We’re seeing lots of early evidence of that. There’s that PACER study of increased filings. And like you, I’m hearing it anecdotally from my friends on state supreme courts, who all say they’re seeing an increase in filings not only from self-represented parties, but from lawyers as well.
Lawyers not only can take cases that they couldn’t have made work economically before, but they also are filing more pleadings within each case because it’s easier to do that.
You and your colleagues, I presume, are just going to continue to see more and more work. The work is not going to go down when legal information is abundant.
Public Legal Data, MCPs, and Access to Justice
Bridget McCormack: You’ve been pretty eloquent about access to the state’s primary law, which is how regular people know what the law expects of them and what the law can do for them, is a public good.
You’ve built a primary-law MCP of all of your state’s primary law that allows access to it as a reliable source. Can you talk a little bit about what that database is and does, and why you think it’s important?
Justice Jerod Tufte: An MCP, in case some people listening aren’t aware of that, is a model context protocol. It’s basically a database that has an interface for AI tools.
We went through a transition from one case management system to a new one a couple of years ago, so we learned a little bit about how good our data was in transporting it from one system to another. We still had gaps. We had transcription errors. There’s always messiness in any system run by humans. Mistakes get made. Records get lost.
In building these tools, I realized there were inefficiencies in having Claude go search the web for cases that I needed to use, or even pointing it at my court’s website, where we’ve had opinions reasonably reliably for 20 or 25 years. We often need to look at law that goes back much further than that.
I tried to build, to start with, the court opinions on the court I sit on. It seems like that would be something I’d have pretty ready access to. I’m a member of that court, and I should just be able to walk down to the clerk’s office to get ahold of where all the good data is.
But you publish in PDFs. You try to extract text from PDFs. Anybody who’s tried to copy and paste out of a PDF has seen that it’s very inconsistent. We’ve had different periods where we published in different formats.
If you’re trying to use an AI tool to automatically check a quote word for word from your citation in your draft to a court opinion, it can only do that if it has clean data to compare against.
I started with the court opinions. I went through everything we had. I looked at CourtListener, which is a public nonprofit that has a pretty good set of court data, also not perfect. You can think about Westlaw and Lexis.
Ultimately, because I’m on the court, this intellectual property came from my court. Most of it is not a copyright issue for me to get a set of. The publishers have their intellectual property, the headnotes and certain things that are their work, that I have to be very careful not to copy into my database if I cross-check against an opinion that we published through them.
For the cases, that amounts to about 20,000 opinions that are at about a 98% quality level at this point. I’m continuing to use Claude to audit these and proofread the opinions.
Interestingly, we have to preserve the court’s own errors in our original published opinions. You can’t just have Claude proofread something because occasionally we miss heading number five in a list of seven items. Humans, right? To maintain fidelity, you have to proofread with the idea of checking against the image.
That expanded to statutes, administrative code, court rules, attorney general opinions — everything that’s published by my state in public format and properly belongs to the public. They shouldn’t have to struggle to get access to that.
It’s really useful to control the hallucination problem. If you can give Claude a quick database query to get legal answers and then have citation cross-checking and cross-linking — if a case cites another case, or a case cites a statute — you have all of those links built into the background. Claude and other AI tools are really good for building those kinds of enhancements to databases.
Ultimately, it’s the public’s work product through their tax dollars. They’ve paid us to do this, and their representatives make these laws. If they need to interact with the courts, I think we need to be responsible for publishing our decisions in a way that’s accessible to the people, in the way that they want to use the law that they’re bound to follow.
It’s a philosophy and technology problem, and it has really been fun to realize some of the challenges with recordkeeping and all the work we have to do in order to make this stuff function for the people who need it.
Bridget McCormack: It’s pretty amazing. I guess it was never really possible before. Like you said, Claude is really good at helping build that database and adding the functionality that allows people to use it.
I assume it’s open source. Anybody in your state can access the database and use it?
Justice Jerod Tufte: Yes. There are two issues there.
I publish on GitHub, which is a public place where programmers and other folks publish their code. Both the code I use to extract all this public data and put it into the database is public, and the current status of the database is public.
Somebody can download the gigabyte or so of information. Then they’d have to run their own copy on their computer.
I do have a public instance that’s out there, but that’s out of my own pocket. I don’t have the capacity to make it available for everybody, and it’s not quite validated at the level that I want to stand behind.
My goal is to get it reliable enough so the court can publish this on the court’s website and do that instead of or in addition to the PDFs that we publish.
Bridget McCormack: Such a great service. Do you find other colleagues in other states wanting to build the same for their states? Do you have copycats yet? I want to help you get copycats.
Justice Jerod Tufte: Yes. I would love to see more states make their court rules and their opinions accessible through MCP and other similar services.
I don’t have takers on that yet. It’s a really tedious, time-intensive process to gather and then validate that data so that you can rely on it.
But I do have one colleague on another state supreme court. I told him about this. He also has a little bit of a computer background from decades ago, so he’s got a little bit of the “knows enough to get in trouble” kind of computer programming level.
The fun thing is you can just say, “Claude, I heard about this public repo on GitHub. Go look at it and help walk me through how to make it work for my state.”
You don’t have to really know anything about computer programming. It will step by step talk you through downloading it, installing it, and installing the auxiliary programs you need.
Then you can have a back and forth about, “Okay, this is really built with North Dakota in mind. I need it to work for Iowa or Michigan or Minnesota or some other state.”
It will go out and do the research and find where it can get the court rules, the statutes, and the opinions. Depending on how your state publishes things, it might be a bigger lift to have it mass download all those and put them into a database. Or it might be something where, given a citation, you can program a web address that just gets you that case. A few courts publish their opinions in ways that allow you to build a URL out of a citation and just grab the case. Those are a lot easier to work with.
In one afternoon, this colleague on another supreme court walked through with Claude, downloaded my code, adapted it to his state, and had it writing bench memos and editing drafts in four or five hours. He’s continued to incrementally improve it and has suggested some improvements upstream to my code, so he can share what he’s learning and doing, too.
Bridget, if you have some connections, and we could get a few more, I think this could really take off.
Bridget McCormack: I have some ideas, so answer my call when I call tomorrow.
Jen Leonard: Justice, you talked a little bit about the floodgates starting to open around litigation as the barrier to filing lowers. You’ve also talked about AI as partly an access-to-justice issue.
In an ideal world, how would a self-represented litigant in North Dakota be using AI tools to establish their own rights and responsibilities in society? What are the positive use cases that you’re thinking about or seeing in your courts?
Justice Jerod Tufte: We have seen a number of self-represented litigants filing very high-quality briefs that are on the level of what lawyers are filing.
But what comes to mind from your question is how people can do basic legal research to understand: I’m thinking about writing a lease with my cousin for my lake house, or I have this property dispute with my neighbor. What do we need to know? Who’s likely to win on these facts?
One would hope that if you could get reliable advice that’s grounded in your law, and not just Claude’s training data or Gemini’s training data, or 50-state law that might or might not be accurate in your jurisdiction — if it can be anchored to a jurisdiction through tools like the MCP that we talked about — then people probably can avoid court altogether, or maybe short-circuit some of these disputes.
As anybody knows, you go to a law office with a potential dispute and your lawyer says, “There’s a 99% chance you’re going to lose this,” and you don’t spend the time and effort and heartache going through that dispute at all.
The greater certainty these tools can give us in terms of what the really not-close questions are will help us focus our courts and intensive, expensive processes on the cases that really are closer to a jump ball, that really do need human judgment applied to them carefully and through often very challenging evidentiary processes.
That’s one of the things that trips people up. They can’t use these AI tools with them when they’re on their feet in court to figure out how to get evidence introduced. But at least their filings can be greatly improved if they’re anchored to the correct law.
Bridget McCormack: It’s such an opportunity to provide a service that was impossible before the technology.
We were never going to lawyer our way out of the access-to-justice crisis. That is just a reality and has been for a number of decades now. Most people can’t afford lawyers for their civil justice problems.
But we know from all the procedural justice studies that sometimes all they need is to understand why they’re not going to win what they think is their legal case. When they understand, they will move on and they will continue to trust the judicial system.
There’s a real opportunity with leaders like you on state supreme courts to really grow trust in your institution. It’s really, really commendable.
Jen Leonard: Even before AI, Bridget, you and I used to talk about the helplessness that you would feel if somebody understandably came up to you knowing you were a lawyer and asked you a legal question. It doesn’t make any rational sense to people why you can’t opine on it or support them.
As Bridget said, the economics frequently just don’t support people getting help. It feels like such an exciting moment to give access to people, no matter who they are and what they’re trying to do legally, to be able to help themselves.
Justice Jerod Tufte: I think that reaction you described lawyers having is what our deputy clerks have in the clerk’s office when people come in and ask them for help. Paralegals who don’t have a law license, but maybe work in the legal field, or in a law office, or for a state — there are regulatory things that we can do.
Arizona and Texas and Utah and other states are doing things with limited licensure. I think these AI tools are going to be able to augment those people, too, in order to really deliver some of those legal services in areas where it just doesn’t work economically for lawyers to put several hours into something.
Bridget McCormack: Not every legal problem needs a lawyer anyway, just like not every medical problem needs an MD. The PAs that I see at urgent care are amazing. They’re perfect for all of my needs.
We could probably have lawyer PAs that could handle a lot of information-giving on lots of issues and grow trust in the institution. It’s pretty exciting.
Advice for Courts in the AI Era
Bridget McCormack: I want to hear from you what advice you would give colleagues on other courts about the moment we’re in, the window we’re in, where I think we still don’t know exactly where all of this lands for the future of the legal profession, and certainly not for the future of courts, except the increased filings, which seems pretty clear that we know is going to continue to happen.
If you could talk to colleagues on other courts, what would your advice to them be about how they should think about this technology and this moment? What are the opportunities? What are the risks?
Justice Jerod Tufte: Anybody who’s been to legal education programming has heard endlessly about the risks. That mostly seems to scare people off from even inquiring about the upside of this risk-benefit calculation.
I start with this: the people in our states — governors and the people who put individuals into judicial office, whether that’s through election or appointment or some combination — didn’t appoint you to run an AI tool to decide cases. They appointed you to exercise human judgment.
I don’t think people are going to trust courts if they think that we’re overdelegating to AI tools any more than they trust courts when they hear that they’re overdelegating to staff attorneys or other non-judges to handle something and then just signing it without reading it.
You see a scandal like that, where a judge delegated judicial power to somebody who’s not a judge. The accountability question has to be front and center. If you are assigned to a case and you put your name on a decision, it’s yours. You read enough and heard enough to own that decision and be accountable to the people who gave you the power of that office.
I tend to see people overcorrecting and thinking, “I don’t want the risk. I don’t understand this tool. I’m not comfortable with some of the automation in other parts of life.” If you’ve driven in a Tesla or another car that does the driving for you, a lot of people are really not comfortable with that either.
But they need to find somebody in their system. Judges can’t talk to just everybody about a lot of these problems. Mostly you can talk to other judges and lawyers in your system. Every court needs somebody who understands the technology well enough to be that trusted intermediary: How can we safely use some of these things?
That means sometimes people like me end up taking on a second or third job in some ways, trying to help out a little, because I don’t think the courts function unless we are able to keep pace with the cases and the volume.
A big problem beyond the cost issue we’ve talked about is time to decision — how long it takes courts to issue a decision in some cases, in too many cases. Time is money, especially in business litigation. If you’re tied up without a decision, sometimes you’re willing to accept almost any tradeoff to get that decision quicker, even if it means a higher chance that you don’t get the outcome you want.
That’s important for us to remember, too.
Being responsible to the voters in our states, as the rules of professional conduct for lawyers require us to stay up to speed on technology and have that minimum competence, lawyers and judges have that same obligation.
It’s hard. It’s uncomfortable. We have to learn new things. But fortunately, judges typically come from one area of practice, and then they get on the bench and suddenly have to learn all kinds of other areas of practice that they didn’t have a lot of background in.
Bridget, you’ve been through that yourself as well. You have to be comfortable not being the expert, but knowing enough to know your limits and where to go for reliable answers.
Jen Leonard: One of the things that enthuses us is that we frequently hear from other judges that they listen to the podcast and they’re inspired by the guests.
Maybe we can help you in your second or third job to scale what you’re doing for the judiciary and ask you for one piece of advice for a judge who’s listening to this, who hasn’t touched AI or worked with it at all.
What do you think would be the best way for them to start the morning after they hear this podcast?
Justice Jerod Tufte: If you’re a judge right now and you’ve got some cases that have been decided a year or more ago that you can use as practice, where you’re not worried about confidential information and the decision has been made, you can’t really hurt anything.
But you’ve got real, live, actual documents to feed into whether it’s NotebookLM or Claude or one of the various tools that lets you load in a huge set of documents for context.
Just ask it questions like you would ask an attorney who’s working with you on the case. Struggle through deciding that case and use it as a sounding board to test out ideas, to identify issues, and to check questions like jurisdiction, which courts have to raise on their own and lawyers don’t always brief adequately.
You have to experiment. You have to do it. You can’t just read about what it means to do it. You have to do it yourself.
I find that you can’t really get the experience with contrived or fake documents. You have to use a real set of briefs, motion papers, evidentiary transcripts, or other things.
If you’re a judge, you’ve got access to closed cases that are still public records that you can use for this sort of thing.
Bridget McCormack: Perfect.
Justice Jerod Tufte: I’ve really enjoyed how you’ve interacted with folks, and I hope some people get something out of what we had to say today.
Jen Leonard: Oh, they absolutely will, Justice.
We have not had many guests from the bench, so we’re really grateful to you for sharing your perspective.
We’re grateful to everybody out there listening to AI and the Future of Law. It’s been a pleasure bringing this conversation with Justice Tufte to all of you.
We look forward to seeing you on the next episode. And until then, be well.