NeLI PodJudge Scott Schlegel on AI in Chambers, Court Innovation and Human Judgment
Listen on Apple Podcasts Listen on SpotifyAbout this episode
Judge Scott Schlegel of the Louisiana Fifth Circuit Court of Appeal opens Season 3 of the NeLI Pod. His road to the bench ran from physical therapy plans to financial advising, law school at night and felony prosecution, before he cashed out part of his 401(k) to run for judge. He tells Daniel Gold and Brandon Mack how he modernized his trial court, how he uses generative AI in chambers with friction built in, why the existing rules already cover AI misuse, what judges are doing about hallucinated citations, and the one decision he would never hand to a machine.
What we covered
- Why Judge Schlegel left financial advising and took the risk of running for judge.
- Fixing court workflows one at a time, from e-filing baskets to an online hearing calendar.
- Why sometimes the right answer is no technology at all.
- Where he draws the line on AI in judging and why the social contract requires a human decision-maker.
- How he builds staged prompts with friction so judgment stays with the judge.
- Why Rule 11 and the rules of professional conduct already cover AI misuse, and the case for AI CLE requirements.
- Building a court-run AI model on the court's own servers.
- What judges are doing about AI hallucinations in orders, including peer groups for judges.
- Using AI to rewrite public-facing court forms and videos for self-represented litigants.
Full transcript
Transcript of the published episode, edited for transcription errors, with speaker labels added by Gold Standard Leadership.
This is NeLI Pod, the official podcast of the National eDiscovery Leadership Institute. Our guest today is the Honorable Scott Schlegel of the Louisiana Fifth Circuit Court of Appeal. And his road to the bench started with a plan to become a physical therapist. And he moved into finance and became a financial advisor, went to law school at night, and then prosecuted felony cases. And then he quit his job, cashed out part of his 401k, and ran for judge while his son and his wife were both in school. That story explains a judge who will question a process and take a measured risk to fix the system in front of him. From there, we move on to his trial court work, his framework for AI in chambers, the national guidance he co-authored, and a court building AI on its own servers.
Judge Schlegel, thank you so much for joining us.
Judge, welcome to NeLI Pod. Thanks so much for joining us.
Thanks for having me. Sounds like it's going to be a lot of fun.
It is. And Judge, I want to start off probably where some of your interviews that you have with some other outlets may not necessarily start. And when Brandon and I were doing our research, one of the interesting things that came up during that research is that you didn't have the traditional arc for a lawyer turning judge, etc. You actually start out aiming at physical therapy. And then you built a career as a financial advisor. So I'm curious, what pulled you away from physical therapy and what did financial advising teach you about people that law school never could?
Well, first, I commend you for your deep research, because I think I've mentioned the PT thing maybe once in my entire interview. So I don't know how you found that one. So I'll be I'll be interested to figure that one out. But, yeah, I mean, look, back in high school, I actually did sports medicine training. I was a sports medicine trainer, sophomore, junior year of high school. And the doctor who was over the sports medicine training actually allowed us to do real things. Helped with taking x-rays. We helped with the actual wrapping. It wasn't just kind of ball boying. It was true sports medicine. So it was a wonderful thing that I was very interested in, and I thought I was going to go into sports medicine training.
So, you know, I went to LSU undergrad and started that journey, and then I got to anatomy, and I took a big dub. On anatomy because it wasn't working out. I couldn't pronounce the words. I couldn't memorize the stuff. So I had to pivot very quickly. I found that I wasn't cut out for physical therapy. And then I lost nine hours and got a finance degree from LSU. So that's what started the journey to financial advising.
It's interesting. When I graduated college, I had a broadcasting degree, interestingly enough, and I found out that freelance video work after I graduated was more free than Lance. And I said to myself, well, I'm going to take the LSATs and I'm either going to get into law school. Or I'm going to do financial advising. And Syracuse Law, they were foolish enough to accept me, so that explains that.
Yeah, well, I did the reverse. I did financial advising and then took the LSAT.
Didn't you go to law school at night, though? Is that correct?
I did, yes, yes. So a week after graduation, my wife and I got married, so we've been together for over 30 years, and I was financial advised. You know, I had my Series 7, 66, 24 and 53. So I was also a supervisor. But that was if you remember in 99 when the market went south for 10 years straight. So I was a 20 something, you know, asking folks to trust me with their retirement funds. And everybody was literally losing, you know, 30, 40, 50 percent of their retirement accounts. And my wife looked at me and said, you need to go to law school. So I said, yes, dear. I went and took the LSAT and started to go to law school at night while I was doing the financial advisor by day.
I would take classes in the morning, go to work, take classes at night. I was trying to finish in three. Normally it's a four year track. I was trying to finish in three. And then my third year, I decided to just go ahead and. Finish out in three-fold. So I resigned as a financial advisor, finished out my third year, and then went to work at a big downtown law firm doing product liability defense as my first law career.
When you talk about the move from attorney to judge, you've described that process as, you know, I quit my job, cash out some 401k, and I ran for this position. And you did it at a time where I believe your wife was in school, your son was in school. There wasn't a whole lot to sort of fall back on if it didn't work. When you had that conversation at home about this is what I think I want to do as a next step in my career, what were you thinking about? What made that risk worth taking for you at that point?
Yeah, I mean, anybody who knows me, you know, faith is everything to me. And I just felt it was a calling to actually run for judge. You know, I was working as a prosecutor trying all the horrible, you know, terrible murders and child rapists. And, you know, I just saw the inefficiencies of the system and it rubbed me the wrong way. And I truly felt called to serve my community in the capacity as a judge because I thought the only people who can actually change the justice system are those in charge of the justice system. And so I talked to my wife again. We've been together since we were 17 years old, you know. We will be together forever, and she supported the decision. As long as we're together, whatever happens, happens.
It's all good. If I lose, I lose. If I win, I win. That doesn't mean go jump off a cliff with that. I don't want anybody to listen to this podcast and go, well, might as well just throw my entire career away. That was not the case. It was a long, thought-out. Process. You know, can we do this? Well, you know, it's my wife with me on this. And we made the decision that was OK. And, you know, I joke because two years earlier, she had quit her job as a pharmaceutical rep and because she felt called to go back to the seminary to get her license or master's degree in marriage and family counseling. So I said, hey, babe, I feel called to I'm going to do this, too.
You know, and that's why she was in school and my son was in school. And so it was it was incredible. And thankfully, it worked out. And that was back in 2013. And every single day I used the opportunity to change one little thing about the justice system each and every day. And, you know, it's a it's that whole you can tell me no to no today, but I'll outlast you. And so, you know, the folks that succeed are the ones that stay the longest and continue to work at it. You know, there's Michael Jordan, not Michael Jordan, because he's the greatest, you know. Free throw shooter of all time. It's because he worked at his trade every single day.
You know, it's interesting that you bring up the concept of change and changes within the court system and changes that you want to push through or have in your career. When you think about the way things change within courts, a lot of times there's risk to making change. We have a system, we've done it this way, we see that it works, it could work better, or maybe this change doesn't actually work out. And a lot of times... When a change with the court doesn't work out, the people who pay that cost are litigants and staff, and there are ripple effects to it. At the same time, the way you just described, you're someone who took a big risk on making the change to get into the court system.
Do you think you learned something through that process in your personal life that helps you assess the risks that you're taking when you're making changes now in the position?
Yeah, of course. Again, nothing I do is willy-nilly. I don't want anybody to take away that I have an idea, I run with it. I have an idea. I think about it. I talk about it. I always tell everybody, really understand the workflows first and foremost. If you're not an expert in workflows, it doesn't matter. Build your partnerships. Go to the people that you work with, whether if you're in the criminal side, whether it's the DA, the sheriff, the probation officers, the clerks, and then pick your tech stack. What I've learned over the years, again, is yes, it is extremely difficult, and yes, there are consequences to every decision, but there are many more negative consequences in my mind with indecision and not making a choice.
If there are 100 problems, I always go for the 70% rule. I want to win 70% of the time. The other 30% will... We'll come up with them. But this whole notion of that's how we've always done it. Well, great. There might be a real reason for that. Defend it. Tell me the why. If you can't tell me the why, I'm coming for it. If you can tell me the why, I got 99 other problems. I'll go. I'll go work on those. And so, you know, from the basics and you've obviously done your deep dive from the basics. When I got to the bench, you know, we couldn't do e-signatures by law. We can only sign warrants. Well, why aren't we doing that for civil and criminal?
Why aren't we building e-baskets? Why are we passing literally paper files from the bottom floor to the floors above? And you have to wait at the litigant for three, four, five days, maybe a couple of weeks because the clerk didn't put it in the right wire basket. And so I started on that process from day one. Met with the partners, met with the clerk of court, met with a state senator, had the legislation drafted, passed, and we're off to the races. Then we can build our e-baskets. Then we can build notifications from the e-baskets telling the judge there's something in your e-basket. You might want to go take a look and sign it. So simple little basic things that exist in the private sector that we're all used to conveniences.
You know, I understand my role. You understand your role. I'm not diminishing any of that. It's reduced the barriers. So that we can then make the choices that we need to make. And none of that, in my mind, is dangerous for the public. It's more dangerous not to do those types of things. You know, so I could go on and on with these examples from the calendaring online to the building a website to the text and email reminders with Zoom links before Zoom was a word. I was doing that in 2014 before the pandemic even was a thought. And so it's just build your business better. It's not a business. The justice system, there are different complications. I get that. But look at your business and run it better.
You know, doing research and talking to the person are always two very different things.
And it's listening to you speak about this. There's a lot I want to dig into. And you brought up this last example. I want to talk about it as well because I did see that too. You talked about Michael Jordan a moment ago about there's a concept. That you have to build the reps to be a great leader, right? You have to build the reps. It's not, it is the ones that people don't see. It's the quiet ones. It's the ones where you don't get recognition for it, but you do it because it's the job, because you want to get better, because you want to do better. And what I'm hearing from you is that this is something that, and you're a faith-built person, this is something that has to be part of the job.
It has to be. You have to train yourself to do those reps. And where does that come from with you though? It's like, that's not something that every single person has, but when you do those quiet reps that nobody sees, you do make those incremental 1% changes. That is the reason why Michael Jordan was, you know, the best basketball player, et cetera. So what is it about, you know, you as a person that really drives you to want to do those things?
Again, a true calling and a true belief that. You're giving an opportunity in any, whether you're a podcaster, a lawyer, you know, whatever, you're given an opportunity to use and utilize that position to the best that you can in any way that you can. And so, you know, first and foremost, you know, I don't want to I don't want to again, I hate I don't hate broad stroking. I hate broad stroking in the sense of like judges should all be, you know, dealing with all the legal technology and become admitted. No, that's not what I'm saying. You know that. Judges are jurists. Jurists are not technologists. Technologists are not jurists. So it's you're only in control of what you're capable of doing as well. You know, not all of us have the same competency levels.
And again, I'm not I'm trying to convey a message that do what you can do in your world today. And whether that's getting rid of the paper. Great. Some people need their paper. I get that. Paper less doesn't mean no paper. But that's pick one. Part of your workflow that you understand very well and can easily spot and see an inconsistency or something that doesn't really flow well and fix it. You don't have to be a technologist to fix the wire basket problem. Get rid of the dang wire basket. It doesn't help. Or get rid of the two other people of the past to the wire basket. So truly understand your workflow. And so for me, personally, I truly feel that I've been given this opportunity to help the justice system.
And it's not just my backyard. Everything I build is for the entire justice system, whether that's my state, my parish or your county or your state or the country. And that's why I speak so much about it. That's why I write so much about it, because we're all in this together and we should be sharing these ideas with each other. And some of my ideas are awful. And I want a judge or a lawyer to write me or call me and say, Schlegel, that is the worst idea I've ever heard. Please do not do that. And I've had some fails, some big fails. I wanted to put a hologram in my machine. I got the budgeting. I got everything, but I could not get the plus one in en banc when I was at the trial court.
They weren't going for it. I don't know why. It would have been amazing, but it didn't work out for me. I love that. I love that.
Well, before, like you mentioned before, Zoom and before AI entered the picture, the story that I read was that a lawyer in your trial court wanted a date that had to call the minute clerk or send a runner, and the date lived in this big red book. You bought a $150 online calendar, and lawyers picked their own dates and sent reminders, and you have said that half your docket fell off because the lawyers were sending letters saying that they had resolved it. So take us back.
That's correct.
What broke and what was that one thing that you wanted to fix? And curious, because we will migrate and Brandon is going to get us there, about what AI still depends on through this system of yours.
Yeah. Again, when I first took the bench, there was my minute clerk, my civil minute clerk, and she had this big red book. It is a physical book. And for the first month or so. I would have to go to my minute clerk to see what hearings were set and what was coming up. I didn't understand this. It was not functioning in my brain again because I came from the private sector. And even, you know, I was also in the public sector, but I came from the private sector and I was like, I don't understand this book. And so I literally went and purchased $150 Acuity scheduling calendar online, did all my research because I was using Squarespace to build personal websites too. And I literally took that big red book.
I literally took that big red book from my minute clerk and I was bad mouth as long as you can imagine. And just I can't believe I mean, that is that is sacrosanct. You don't touch a minute clerk's big red book. And so I literally took it and said, all right, well, I'm going to build a Web site for my division. So I built a whole Web site for my division. It was onlinecourt.com. I built it for my division, and I literally put all the orders that you need on there, all of the different jury instructions you need on there. I then put an online calendar and said, I can handle whatever. We did criminal, civil, and domestic. So I had 100 on my criminal docket on a Monday.
I had, you know, on a Tuesday, I had all these other criminal and civil cases and a mixture. So I said, look, I know I can handle. 25 pretrials at 9, 10 motions at 10 because lawyers sitting in court wasting their time and wasting their clients' money was ridiculous to me. And so I literally put my calendar online and said, call opposing counsel, make sure the date works for you, put it on there, dump it on your Outlook calendar. You still have to sign the rule to show calls, but I'm giving you the availability. If you can schedule a hair appointment on a dentist online, why can't you schedule a hearing? And it would pull them off. So at 25, went to 24, went to 23.
So you always knew if you could get a hearing with me or not. And then because I had the text and email reminders built in, it cut down on the motions to compel because they realized, do we need to come see Judge Schlegel for that? No, no, you do not. You can figure life out. And a trigger for a lawyer is go pick your file up. They'd pick the file up, call each other, send the letter. It also helped with criminal cases because how many cases can I handle at 9 a.m. if I have 100 set at 9 a.m.? One. So why do I have a courtroom full of 100 people? Not only is it a security issue, but it's an issue for all the people in the courtroom.
I can't handle all those cases, so I started to tell criminal defendants they couldn't pick online, but they could pick while in court. You're ordered to appear on this date. What time would you like to come? 9, 10, or 11? It doesn't matter to me. I'm going to be on this bench until 100 cases are resolved. So you tell me what time you can come. They would then pick the time, and then I would order them to appear at that time, put it in the online calendar, and then it would send them text and email reminders, as many as I wanted. And it would include the name and the phone number of all the public defenders in our courtroom. It would include links to that website that I just built that if you want to fill out a plea form, it's right here.
Get with your attorney beforehand. And then we cut down on failures to appear so that people aren't getting arrested and sitting in jail for days for possession of cocaine when they truly forgot. Some people are running. I'm coming for you. Keep your head down. But not everybody is running. And so as that matured. We then would start saying, all right, if it's a low level offense, I'll allow you to appear by Zoom. If it's just a hey, how you doing? Do you have your discovery? Because it's no big deal. Crimes of violence or sex crimes, not doing that. You're coming physically. But again, you're I understand the workflow. I am an expert in the justice system. I have done this so many times that I understand where the tech matters and where it should be placed and where it has no.
Basis for being placed into that part of the workflow. Sometimes the answer is no tech. Don't touch tech. And so that's really what it is. And, you know, most people get worried about this because they think it's some weird voodoo science as opposed to, you know how to do this. You've got this. Just think through the workflows and think through your daily life and figure out, will that work there? Will it not? Should I? Shouldn't I? Do I have the money for it? Do I have the votes for it? Do I have the procurement for it? And then just start building.
Speaking of tech and workflows, one of the things you've written is that AI is a tool that belongs in your workflow, but not on the bench. When you think about the way that you use AI in your chambers, where do you draw that line? Could you give us an example of a task that you would trust to AI at this point and something that you would never see yourself giving over?
Yeah, the easiest example I always give is in the domestic arena. So we handled child custody issues all the time. And so I always tell everybody, listen, nobody ever wants to hear that Judge Schlegel, if he would have used this model, I would have had custody of my children. But he used that model. So I don't have custody of my children. But what they probably want to hear is that six month process. Was shrunk to two or three months because I used Gen AI to get ready for the hearing. And so, you know, there's this whole movement out there that because of Gen AI, SRL, self-represented litigants, are now able to file more and more pleadings. As a result of Gen AI, lawyers are able to file more and more lawsuits and more and more motions.
And because of that, judges and courts who were already understaffed and overworked are getting to a point of being swamped by all of these new pleadings. And there's this movement out there that says, and because of that, we should allow AI judges to be built to handle the low level, low cost, whatever you want to say, parts of the justice system. And I'm always pushed back and say, that's not the social contract. I love all of this technology. I think it's wonderful tools. But I never signed up and Brandon and Daniel, you never signed up for an LLM to rule over you. That is not the social contract. Hey, look, in 20 years, if they want to rewrite the social contract and everybody wants to roll with it, have fun.
But that's called arbitration and mediation. It's by consent. Do it all day, every day. I'm here for it. That's all good. But once you are sued and you are brought before the government, it is. Our responsibility to hear your case, ensure that you are heard and provide the justice that you demand. And that is why I think the Jenny line is the human judgment, the human decision making. And we can argue and debate over, well, you got it to do this. That's human judgment. Is it? And that's the great debate right now. So, again, by consent, do whatever you want. Once you've been brought before the court. That is a different ballgame. That is a social contract that I'm not willing and able or ready to rewrite.
How about we equip the judges that we have with the gen AI tools necessary to do their jobs and educate them so that we have a competent AI? Competent judiciary? So that I can handle this flood of new cases? That's what I would propose.
And one of the things that you've told judges is to treat. AI where they use it like it's a first year law clerk. You double check everything. Don't trust it with more than you would trust someone with that level of inexperience and so on. In your mind, what are things what are the things that law clerks still do that they will do better than tools are currently able to do for you?
Yeah. Again, I have a analogy or story for everything, Brandon. So you'll just have to you'll have to bear with me. Absolutely. So I got access to one of these tools early on that was tethered in actual case law. And so, you know, magical. I have like a three page rule set that I have developed to feed the Gen AI tool before it even begins. And a very important part of this is friction. You know, everybody is talking about convenience and getting rid of all the friction. That is not what we want. We want friction to remain where it belongs. And so the way I built these this instruction set is to. Stage, stage, stage, stage. You have to stay in control the whole time.
Make sure it's good before going on to the next stage. So all these things. So I built this whole process out. I tried it with a new case that came in just to get it to draft me a bench memorandum. I want the procedural history. I want the list of the witnesses and exhibits. I want the assignments of error verbatim. You know, the whole really solid bench memo. And what it produced was. Like, I was blown away at how incredible these tools were. I mean, it laid out the arguments of counsel. It laid out potential paths. It had case law that I could click and check. I could click within the record and check and make sure everything was correct. It was incredible. Fifteen minutes later, my clerk walked in and had not used any Gen AI tools and said, Judge, I think the matter is moot.
And she was right. The Gen AI tool. Understood the instructions, my law clerk understood the assignment. And so the point is, as smart as it is, as intelligent, and I'm not trying to anthropomorphize it, as intelligent and smart as it is, it doesn't see what an experienced lawyer or judge can see. So keep it in its proper place, and you will find that all these tools are amazing. You know, I always liken it to when I was a first-year associate, and that partner came to me, well, The partner called me to come to him and said, hey, Scott, I've got this issue I need you to research. Go prepare a memo for me and get it to me by tomorrow. So I went and I did all this research, and I was so proud of this memo.
I was like, Brandon, I'm going to make partner in three years. No chance. I mean, it's so good. I walk in and handed the partner, and what did the partner say? That's trash. You saw the one tree I pointed to. You need to see the forest around that tree. And you're young, so I expect you to see at least two or three or four of the trees. I don't expect you to see the full forest. And that's where we are with these Gen AI tools. First, it could only spot the one tree. Now it's spotting the three or four or five trees around it. It still can't see the forest. It still doesn't understand what you and I intuitively understand and can see because of our years of experience.
And so if you understand the strengths and weaknesses of these tools, you'll be good. If you think you're Max Verstappen and you jump in the F1 car, you're going to put it in a wall so fast because you are not ready for the strength and power of that vehicle. Pick the Ford Mustang. You'll get around the track. You'll fully get around the track and you'll have a blast doing it because you'll actually be able to handle that kind of power.
That last part, interestingly enough, at the NeLI conference, one of the panel speakers, Marisa Zalabak, also she's been on podcasts. And Marisa, she's an AI ethicist. And her quote that has stayed with me since is. I love your analogy, and she says something similar. She said with AI, it's like all these lawyers were handed a surfboard in the middle of a tsunami. And I see that. And, you know, it's a great transition. In November of 2023, when the federal Fifth Circuit weighed on AI certification rule, you wrote an open letter, Judge, arguing that orders banning Gen AI or requiring disclosure of its use were unnecessary. And you disclosed that you drafted that letter with ChatGPT4. And then in February 2025, you and five co-authors published Navigating AI in the Judiciary.
And it warned that. No known tool had fully resolved the hallucination problem and it flagged automation bias, confirmation bias, extrajudicial information that the parties never had a chance to address. So I'm wondering what makes education and guidance the better first move and which warning from those guidelines are still relevant right now?
Yeah. Education of regulation all day, every day. And, you know, I always like to tell everybody, just do me a favor, close your eyes for one minute and think about. The cat lawyer. And for those that those that are listening right now that may not potentially remember the cat lawyer, maybe just 30 seconds to inform them. When when nobody knew how to use Zoom, a lawyer appeared virtually as a cat and he could not figure out a way to turn his cat imagery off. And the judge was very patient and kind and waited. And side note, most confident lawyer who ever lived because the guy could not figure out how to turn the cat off. And he's like, Judge, I know I'm a cat, but I got this.
Let's keep going with the hearing. And so, you know, I use that as an example to say, listen, when that happened, we didn't run out and say you cannot use Zoom or any of that kind of stuff. What we started to say is education over regulation. Every CLE that we attended for the next three years was about. Please don't stand in front of an open window. Turn off your phone. Don't let people walk into your office. I don't want to see your ear. Okay, you're on mute. You're on mute. For the love of God, use your space bar. You're on mute. So we did that over and over and over again. We are lawyers. We know how to adapt. We know how to do this.
There's no reason to say you cannot use Gen AI because what that means is you can't turn your computer on because don't use Word. Don't use Westlaw. Don't use Lexis. Don't use Grammarly. Don't do anything. Just handwrite all of your pleadings and send it by horse and buggy because, God forbid, you get into a Tesla and then they're using AI tools. And the people who are doing the regulating, the judges and the legislators, they're not experts in this area because if you remember, the first order was you cannot use AI. And then they went, oh, there's a distinction between AI and Gen AI? Okay, you can't use Gen AI. Well, look. You can use Gen AI, but you need to tell me that you're using Gen AI.
And I always push back on that and say, OK, we were all lawyers. We all worked somewhere where a paralegal or a first year associate prepared a pleading for us that we looked at, thought it was great and signed it, maybe a few edits and then sign it. Right. Nobody that listening to this podcast ever did that. I know you're better than I am, but I certainly did that. I didn't run to the court and say, hey, Judge, Daniel actually wrote this. And then I signed it. Right. What I remember is the judge saying, that's your signature, Daniel. I own you. And you told me that everything in that pleading is true and correct to the best of your information. What do we need a new rule for?
All the rules exist. Now, look, some rules need to be modified for today's day and age, like a rule 11. And it's equivalent, you may not have one in your state or you might not have one in your court. Just adapt the Rule 11 equivalent just to make sure that everybody understands. If you use Gen AI, you're still responsible for everything that's pleading. That's fine. You're not changing the rule. You're just modernizing the rule.
It's funny you brought up Rule 11 because I wanted to ask you about this. So I give a lot of CLEs all over the country, and one of the things that I talk about in a lot of my CLEs, or almost all of them, is the fact that. The technology changes, but the rules don't. And the rules don't need to change. I don't think they need to change. And who am I in this voice of every barrister in this country, right? But I don't think that the rules need to change. We have Rule 1. We have Rule 3. We have Rule 5 in particular. But we also have Rule 11. And Brandon and I have talked about this before. To me, when I was in law school, Rule 11 scared the living daylights out of me, right?
We have this idea, this notion that we uphold ourselves. I like to say as officers of the court, and I feel like it holds more weight than us saying that we are lawyers or attorneys, because when we say officers of the court, it gives just a higher level of understanding that Rule 11 exists for a reason, right? Rule 5, in the Rules of Professional Conduct, for those that don't recall, it says what you said, Judge, which is that if... Anyone else is doing the work, the clerk, the intern, the associate, whoever is doing the work, you have to read it before you put your signature on it because the second you put your signature on it as an officer of the court, you are attesting to the fact that it's true, it's valid, and you're not wasting the court's resources and yours, right?
So with Rule 11, I mean, where is it falling short lately, though?
I mean, you look at – there's a researcher online, Damian Charleston. I think I pronounced his last name incorrectly. But he's got a list of, I think it's like 1,600. So over 2,000 now. Over 2,000 now. Thank you.
Over 2,000 lawyers who are smart, educated people, passed the bar exam, practicing law, and still doing this with all of the hallucinations or just making things up or copying and pasting. Now expert witnesses are doing it. We have case law to talk about that as well. What is happening? Like, what is wrong, Judge, that this is all happening now? Yeah, I've pivoted in my thought process, you know, and to be fair, some of those are SRLs, too. So hopefully it's not over 2000 lawyers doing it. But I've pivoted to say, OK, I don't think we need any more regulation, but I'm now. Arguing and asking Supreme Courts to mandate one-hour Gen AI CLE. We already have to do CLEs. This is the most powerful, dangerous, if you're not paying attention, technology out there.
So maybe we need to mandate one-hour CLE on Gen AI. It's not going to prevent it, but hopefully if people hear it enough times, just like we did with the Zoom, hey, make sure you check this. Hey, it's predictive. It's a predictive model. It's not authoritative. Why are we taking it as such? Check everything. You know, all the things that we're talking about right now, you know, unless you do this and think about it all the time, you're, you know, you don't understand the whole, you know, if you throw the ball left, that thing's going to go left. It's designed to please you. You know, it wants to hear you say good grok, good grok, good Claude. You know, it's designed for that. And a lot of people don't understand the basics like that.
You know, I have started to, you know, like you, I've given this talk a lot. And I've started to push some of the slide deck and, you know, delete things that I thought are no longer necessary. But it's amazing when I get there that the whole, you know, it's designed to please you. The way you ask a question is which way it's going to lean. It's like a dog on a bone. You can tell it to drop the bone, but 10 prompts later, it's still going back to the thing you told it to drop. So you got to start over and just get rid of that context window and start a new one. The basics still matter because, you know, people have heard it, but they don't really get it.
And so it's important in my estimation, you know, that we do the one hour CLE and then we can just talk about all of it.
You've written with Chief Judge Susan, forgive me if I'm mispronouncing your name, Shahardi. You've written about building your own AI inside your court, looking at your own couple of AI servers, some open weight models to use there, interviews and questionnaires with your staff before building any code around this. A lot of care in how you would build something that is native to your court. And then you talked about filing compliance and intake as the first task that you would want to hand off to that AI. When you think about the process or even just writing about the process, what do you think that building your own AI environment taught you that you couldn't get from just using a product off the shelf?
Yeah, I mean, we're still in the building process, so we're still going to be learning. But the process itself, I think you can learn the similar process, whether it's on-prem or off-prem. You know, just the whole like, again, start with the workflow, build your partnerships, you know, make sure that you're doing it right. I think one of the issues that the reasons that we have gone on-prem are, you know, not that we can't. Have a secure off-prem model. I don't want anybody to take away that you can't have secure off-prem. But we wanted to start thinking through the whole process of confidentiality, security, privacy. On-prem makes judges feel warm and fuzzy inside. It just does. I mean, that's the way it is. And so if we can start building a model that handles some of the...
The processes that don't require frontier model legal reasoning, you know, it's that, again, really good is good enough. And a lot of times, you know, a lot of people like I need the most powerful that whole same car analogy. I need the most powerful car that I can drive. You might in certain situations. But, you know, you need to learn first to how to, you know, take that, again, Ford GT around the track. And you're going to have fun and it's going to do just well and you're going to do great. And so we're trying to do that on the smaller tasks that don't require such powerful legal reasoning. And it'll do a great job at all of that. We also don't know where all these models are going.
You know, you hear all of these different issues with token prices, you know, that your subscription today is wonderful, but they lose billions of dollars a year in compute. You know, maybe they start charging you per token. And, you know, if I can change that into what just cost me power as opposed, you know, in my courthouse. As opposed to tokens, you know, maybe we don't lose everything. So I think at the end of the day, I think it's going to be both. We'll find out where this goes. You know, the models that we can load on our servers, we can't buy big enough servers that handle frontier model open way. You know, we're at maybe 30 billion parameter models that we've got. We can't load.
I tried to load an 80 billion and it was like, no, thank you. Not happening. You know, because it's not like the court has this kind of money to buy the servers that you come to your mind. And so, you know, they're powerful enough to do everything that I want. It's powerful enough to give a quick, you know, what I need from it. And it's good enough for some legal reasoning. And maybe it surprises me as the open weight models get better and better and better. You know, I know they release one every six months. So we'll see. It's part of the learning curve. And, you know, I want to share that experience with everybody so everybody else can hear that experience and go, well, maybe I'm going to go buy an AI server.
Who knows? It's a it's a it's a fun exercise. And I think we're learning a lot. And I think there'll be plenty of value to not only the court, but also the lawyers and litigants that practice before us. Because, you know, our hope is that you try to file something and it says, Brandon, you're at your pagination is a little off here or, you know, local rule blank says this. You can file it if you want. I'm just you know, it's essentially what a clerk would do after you file. Hey, Brandon, you need to amend this, buddy. It's not working out for you. So we have to do that on the front end, too, to help before you spend time, money, and effort. And at the end of the day, you're the lawyer, you're the litigant.
You want to push submit, go ahead. And then it gets the hands of the minute clerks and the central staff a report that they still have to verify. We're not saying, hey, AI, you get to take control. We're saying, hey, AI. Because of the amount of work that we have, it's your job to serve the clerk or to serve the central staff and help them do their jobs better and more efficiently.
I think I can infer from your answer why you started where you did or why you started with the processes that you did in this initial mapping out. Have you mapped out where you want to go next as far as where is it going next?
I mean, where is it? Where do I want it to go next? When it when it hits, I want it to do all the site checking immediately on the front end to ensure that there's no hallucinations in the case to make sure that, you know, all the things I can do now manually, I want to make and, you know, make it done on the front end. So obviously site checking. I want to also hyperlink all of the all of the pleadings to, you know, the actual cases so that I can just click on the click on the pleadings and get it straight to the cases. I wanted to create a bench memo, the one that I talked about that I do manually. I want it to be able to do that again on the front end so that it can just be part of the process.
And again, look where friction belongs and where friction can be taken back. You know, I've added a lot of friction to the process. Once you've done it 10 times, you don't need all that friction. You know, then you can understand you can understand it. But I always build friction in on the front end. And then you, the user, can decide where you want to remove friction once you, you know, once you become Max Verstappen. Hey, if your mic's for stopping, go for it. You know, let it rip. But you're not there yet.
Yeah. You know, we've talked a lot about what happens when attorneys make mistakes with JNAI or pro se litigants as well, right? But one of the things we haven't talked about is what happens when the judges, not to say that anyone on the bench ever would judge, but.
Oh, they do and they will.
Well, you actually said once that when lawyers make mistakes, they get sanctioned or the pleadings get dismissed. But when judges do it, it becomes law. That's a much different thing. And there's this ABA Journal article that basically carried that line with you, right? And the federal judges in Mississippi and New Jersey withdrew rulings after litigants pointed out errors. And then a Georgia appellate court overturned a divorce decree built on fake case law. So what has to happen in chambers? To prevent that outcome? Give us the behind the scenes or under the cloak, if you will, right? What are judges doing both on the state level as well as nationally to ensure that judges are practicing what they preach to the lawyers about Gen AI?
And you forgot the U.S. Fifth Circuit just removed a case from a federal judge in Mississippi who had submitted an order with hallucinations, made up facts, made up case law. That happened two weeks ago. So, you know, what are we doing? We're educating. And, you know, I also say it's not our jobs to be first. It's our jobs to get it right. And so when I educate judges, I always say, look. Start small. Start with again. Start. Do something that you've already done before so that when you can see it, you can see the behavior of the model and go, that's not right. That is right. Treat it like a first year associate. That's one of the reasons that we formed Jake, the Judicial AI Consortium, myself, federal judge, magistrate.
Maritza Braswell out of Colorado Springs and federal district court judge Xavier Rodriguez out of Texas. We formed the judicial Act Consortium. We have over 450 judges now in the country, both state and federal at all levels. Incredible. And we do what we call pop-ins every month to a month. And so judges will come and pop in for 40, 45 minutes and say. It's led by a judge. It's only judges. We're not recording. There's no vendors. There's no it's just us talking about issues so that we can all see it. We've talked about these orders before on requiring disclosure, not requiring disclosure. What are you doing? Why are you doing it? Some, you know, some had been issuing them and actually said, you know what, after that pop in, I actually rescinded the order because I didn't think, you know, I didn't think about it and thought that was a good idea to rescind it.
We're just talking about these issues. You had the prompt injection case out of Connecticut with Judge Spader. So he led one of them to talk about the prompt injection case and how to look out for those types of things. We talk about what tools we're using, what security and privacy you should have on, whether or not you should have your IT involved and where you should have them involved. I'm sure we'll have an on-prem, off-prem discussion. So it's just a way to bring light to the issues. And, you know, tell every judge, look, I'm not telling you to go use it to draft an opinion. Please don't do that. But you might want to start with a very simple legal research issue and a memorandum on a singular research issue that you know backwards and forwards just to see what happens and, you know, go from there.
And again, just because just because you're a judge doesn't mean that, you know, you're ready for the. Every issue of everything that a genii tool can produce, so learn it, whatever, and I tell everybody, use what you have. Don't go get free ChatGPT, or free Claude, or free Grok, or free perplexity for the love of God. Don't do that. Thank you, thank you, you know? Use what you have. If you have Westlaw, use CoCounsel. If you have Lexis, use Protégé. If you have Office 365, use your enterprise-level co-pilot. Use what's already in front of you and understand how they work and how they don't work. And then build from there. You can certainly buy enterprise-level Claude and Chetchy PT. The $20 version might be acceptable if you have those settings right for the right type of use case.
But again, I... I begged them not to use, you know, these other tools because they are so powerful and they look so good. The outputs are so convincing. You know, it's like it was a was a three year old when we got them. Then they became 15 year olds. Now they're PhD students where you're like, sounds pretty good. You know, I'm going to roll with it. You know, back when your teenager tells you something, you're 15, you're like, get out of here, kid. That's ridiculous. You know, now your kid's a Ph. D. Or now a doctor, maybe. And you're like, yeah, you're the medical expert. I'm not. Sounds good. I'm going to roll with that. I don't have enough time to fight you on this one.
You've also argued that some of the most promising uses you see for Gen AI in the core system. Are with public-facing court materials for the general public to use, things like handbooks or videos on what to expect when you come to court. What would you say are the things that most need that sort of Gen AI overhaul, and what's keeping courts from pursuing that?
Yeah, so if you go to any courthouse across the nation, you'll go and pull a... a handbook or some forms anywhere in the courthouse. And those forms haven't been touched in 20 years. You know, if you ever get picked for jury selection, you'll get that piece of paper that just tells you to appear at this date and time. And then when you get to the courthouse, you sit down and they show you a video of judges that are deceased now because it's so old and it's just the same folks telling you something. And those videos cost 30, 40 grand back in the day when they were produced. I mean, they're not cheap. And so, you know, I always tell everybody, here's another low level risk way to do it.
You know, the administration of justice side, take every form that you have just stuck in some folder somewhere and just put it in a gen AI tool. Again, you are the expert. Hey, revise this form. Make sure that it, you know, a third grader can understand it. Make sure that it, you know, has the amount of pagination that we want. Make sure that the font, it looks good for readability. Make sure that, you know, whatever. And then you can sit there with that form. In 30 minutes, you're going to have a much better form that's much more readable for everybody. Same with the videos. Once you finish that form or the jury selection process, I always say I could put a video online. Deep fake, you always say deep fake is a nasty word.
It is if you're using it improperly. But synthetic videos can be wildly useful for the public writ large because I can literally create a video that says, hey, Daniel, when you get to court, make sure that you bring a, you know, it's kind of cold in here. Make sure that you bring a jacket. Here are the phone numbers and the people that you need to see. Here's where you should park and how much it's going to cost you. I can literally put that video up. Any race, any sex, any language that I want, put a QR code on the back of that jury notice, and they can literally scan it, bring up the video, and now they're more comfortable when they get to the court.
And so this is that whole running your business better. I mean, if we have it in the private sector, why can't we do this? And everything I just said costs absolutely no money. I mean, literally, my running schtick well before Gen AI was, I can modernize the justice system for under $1,000 a year in any jurisdiction within 30 days, and I'll do it for you.
You know, we have spoken to a lot of judges, and I'll tell you that you are a fascinating judge because you are one of the more tech-forward judges that we have ever spoken to here on The NeLI Pod. And it fascinates me, like you were saying, even before JNAI, I mean, you were building the court's website. I mean, I've never heard a judge say that before. It's fascinating to me, and it explains a lot about who you are as an individual. And, you know, I know we only have time for one last question here. So I'm curious, you said something before I want to come back to. You were saying, and you've said this before publicly as well, that it's part of the social contract.
That you want a human being struggling through the difficulties of a case and making a decision, not a bot. And I love that. And I'm so glad that you brought it up on the air because it really is something that Brandon and I want to go over. And so let's talk about the one thing that you had touched on lightly earlier. Suppose a tool one day does perform a judicial act well, which act would you still defend as human only? And what do you want every listener to do this week?
After listening to this podcast with you. So you may know that I wrote a book, an e-book called The Weight of the Rogue, and it discusses all of these concepts and principles. And so I would ask you, Daniel and Brandon, how much do you think my robe weighs? How much does your robe weigh? How much do you think my robe weighs?
I'm going to go with 350 grams. Brandon?
Grams? He's going in grams?
I weigh all my food, Judge, so I weigh in grams. If we're going to go with grams, I'm going to get closer to like 900 grams.
Yeah, and what I would tell you is you have absolutely no idea how much that weighs until you put it on. And if you understand and appreciate the weight of the robe and understand what these decisions are, if you're not ready for that weight, it will crush you. And if you misuse it, you are going to hurt somebody. And so I would never. Ever turn that over to a bot because it cannot feel the weight of the road. And without that ability to feel that weight, you cannot make an appropriate decision.
Yeah, I love that. That's a great answer and something that, how can you imagine an AI ever truly understanding that? Something to think on. Well, Judge, it's been a fantastic conversation. We want to thank you again on behalf of both Daniel and I and all of NeLI Pod listeners for taking the time with us today and for all the insight you've given us. It's truly been a pleasure.
Thanks for having me.
For listeners, here's your assignment. Pick one workflow in your own practice or in your chambers, write down the results that you will measure, and run it for two weeks before you buy anything new. Then tell us what you measured. Links to AI in Chambers, Navigating AI in the Judiciary, and his October 2026 piece on building AI inside the courthouse are in the show notes at ediscoveryleadership.org. Keep the tool in the workflow. Keep judgment on the bench, and we'll see you next time on The NeLI Pod.
Daniel Gold is an attorney and a Principal at BDO USA, where he leads eDiscovery managed services. He co-hosts the NeLI Pod with Brandon Mack for the National eDiscovery Leadership Institute, hosts BDO's Legal Tech Talk and founded Gold Standard Leadership.
More about DanielGet the next one in your inbox
Daniel Gold writes weekly on leadership, legal technology and eDiscovery.
Subscribe on Substack