Adam Unikowsky, an appellate litigator with nine appearance in the U.S. Supreme Court, argues that judicial law clerks could be replaced by AI. We discuss:
💻 “AI will make judges release more accurate decisions more quickly. This is good.”
💻 Judges already rely on clerk summaries, so if AI produces better summaries faster, that is good.
💻 AI is a mysterious black box, you say? Well, law clerks are already invisible to the public yet influence judicial decisions without any input from the litigants.
💻 True, law clerks are human—but they are still often wrong. “Is it really preferable that judges receive recommendations and draft opinions from ideological 26-year-olds?”
✍ A writing tip: “Unclear writing usually implies unclear thinking. If something is unclear, it’s probably because I haven’t really figured it out.”
👩⚖️ An an oral argument tip: Don’t read from your notes. Adam relates a story when the Supreme Court stopped an advocate by asking, “Counsel, are you reading this?”
Use this link to get a 25% lifetime discount on Casetext.
Other items discussed in the episode:
Adam Unikowsky 0:03
unclear writing usually implies unclear thinking when you're reading something. And it's not a well written oftentimes because you don't really understand your argument. Welcome to
the California appellate podcast, a discussion of timely trial tips and the latest cases and news coming from the California Court of Appeal, and the California Supreme Court. And now your hosts, Tim kowal and Jeff Lewis.
Jeff Lewis 0:24
Welcome, everyone. I am Jeff Lewis.
Tim Kowal 0:26
And I'm Tim kowal both Jeff and I are certified appellate specialists and as uncertified podcast co hosts we try to bring our audience of trial and appellate attorneys, some news and perspectives they can use in their practice. As always, if you find this podcast of use and helpful to your practice, please do recommend it to a colleague and a quick announcement.
Jeff Lewis 0:42
Our podcast is sponsored by casetext, casetext is a legal research tool that harnesses AI and a lightning fast interface to help lawyers find case authority fast casetext has also announced a new AI tool called co counsel, which we highly endorse listeners of the podcast will receive a 25% lifetime discount on casetext legal research tool available to them if they sign up at casetext.com/calp that's casetext.com/calp
Tim Kowal 1:08
and speaking of AI that's the topic on everyone's lips is what is AI going to do to legal industry is going to replace all US lawyers. And to talk more about that topic. today. We welcome to the podcast Adam unit kowski. Adam is a former law clerk to Justice Scalia, he's a prominent Supreme Court and constitutional litigator. He has had nine Supreme Court arguments under his belt is Adams, appellate and Supreme Court experience spans a wide range of industries and subject matters including securities litigation, Indian law, patents, civil procedure and constitutional law. With his depth of insight into the appellate and supreme court litigation process. Adam frequently collaborates with larger legal teams in developing strategies across multiple cases and courts. And in addition to his supreme court clerkship, Adam also was a clerk for Judge Ginsburg on the DC Circuit, and specifically bearing on our conversation today about AI in the legal field. Adam holds undergraduate and graduate degrees in electrical engineering and computer science. atomia kowski. Welcome to the podcast.
Adam Unikowsky 2:11
Thank you. I'm looking forward to participating.
Tim Kowal 2:13
Well, thank you very much for joining us. We did see you posted a tweet on Twitter some months ago about your substack blog post. I think the title we'll talk about a little bit later on the podcast the title, I think it's should AI replace law clerks. But before we get into that, let's talk a little bit more about you and your practice. Would you tell us a little bit more about your practice, Adam, and do not leave out your supreme court appearances?
Adam Unikowsky 2:38
Sure, well, I partner at General block and the appellate Supreme Court practice in Washington, DC. But really, I handle both called supreme court litigation, as well as trial litigation, litigation agencies, pretty much interested in anything. So any type of litigation with a legal angle, I'd love to do it. You know, for instance, this year, I've argued appeal as an arbitration case, and then a securities enforcement action case. I've argued cases and everything from federal Indian law, to constitutional law, criminal procedure to lots of other things. So yeah, I have interests across a wide range of legal issues. And you know, I love taking I love arguing the Courts of Appeals and the Supreme Court. That's what I love to do. There's just nothing else that gets me more excited. But of course, I love doing cases in federal district court to where you really have to be creative, and there's no record they have to use, you can make your own new arguments and not be accused of making a new argument. So that's pretty great tip. And as you said, He's I've argued several times in front of the Supreme Court, which is, of course, a real rush for any lawyer, and I'm doing another one in October, at least in the fall, probably October Runa. Interesting case about Article Three standing, so I'm really stoked about that.
Tim Kowal 3:38
Oh, that's fascinating. We'll have to watch for that, if you would send us a link to that case, as we can put it in the show notes and follow it. So speaking of clerkships and asked you this a few moments ago, just before we hit record about the value of clerkships. Now, you've clerked on both Supreme Court and an Intermediate Court of Appeals in the DC Circuit, it seems like part of the value of a clerkship is that you get an experience and a sense of how the court operates. And if that holds true, I wonder if you think a clerkship holds additional value or heightened value. If you're doing it a different level where your practice takes you most at the district level, or at the intermediate level or at the US Supreme Court level? I wonder if you can comment on that.
Adam Unikowsky 4:14
Yeah, I do think it helps, you know, having a sense of how the judges decide cases, gives you a sense of what you should put in your briefs to try to get them to decide the case in favor of you. I mean, it's probably the case that if you're litigating regularly in the courts of appeals for purely instrumental purposes, the appellate clerkship will offer you more information about the process behind the scenes. But of course, clerked for the Supreme Court is a really neat thing to do. I mean, you know, you're working on very significant cases, you're seeing the how things work from behind the scenes, and it's a great year.
Tim Kowal 4:42
Can you point to anything, any experiences, or behind the curtain conversations, glimpses that you've got at the intermediate level that maybe you wouldn't have gotten through your experience at the US Supreme Court?
Adam Unikowsky 4:53
Well, you know, I think one of the differences is that the Court of Appeals has mandatory jurisdiction over appeal. They can't just take the cases that seem to be good vehicles to resolve legal issues, they have to decide every case that reaches them. And so you know, a lot of cases tend to be very complicated, factually, with lots of twists and turns and sort of seeing how the judges deal with that and kind of pick out the important issues. I think it's helpful for an appellate lawyer that you really have to turn your brief into, like a story, you know, and crystallize the case into a couple of issues where else the judges eyes kind of glaze over. I think that just like understanding, especially since the judges hear a lot of appeals, like actually, the DC circuit where I clerk hears less than, for instance, the Ninth Circuit, you know, I mean, the California Court of Appeals, where the judges are probably hearing 510 more cases per sitting. And there's only so much information they can retain about a given case. And so like, recognizing that you will know a lot more about your case in the judges, and you've got to crystallize the case into an easily comprehensible story is like one of the things that you learn that's particular to clerking in an appellate court, I think, yeah.
Tim Kowal 5:54
Now, Adam, you're an appellate attorney, can you give us tell us, you know, the day in the life of your practice as an appellate attorney? Are you always in the Court of Appeal? Are you sometimes in the district courts or trial courts helping to make the record what's a typical day or week look like for you?
Adam Unikowsky 6:08
I'd say it's, it's actually probably less often in the courts of appeals than in other tribunals. I do a lot of cases in federal district court where as you say, I think in some ways, there's more creativity as a lawyer necessary, because you're not sticking with the record, you're making the record, you're figuring out your legal theories. And oftentimes, I'm asked to come in before lawsuits even been filed, you know, client will call and say, you know, there's this looming issue, and should we sue? Should we do something before we sue, you know, could we sue, and you know, that the public never learns about this representation. But that actually often involves some of the most creativity involved, because you have to figure out the legal issues, and then you can also figure out what you can do to avoid them. And so there's several degrees of freedom there. So, you know, depends on the day really exactly what I'll do.
Tim Kowal 6:47
Right? Do you have any preference being the appellant or the appellee?
Adam Unikowsky 6:51
Well, I like to be the appellee. Because then you want to say there is no abuse of discretion advocacy here. But you know, getting a decision from a district court flipped is very exciting, right? Like, clients love it, because they feel vindicated. You know, it's obviously harder to get a decision reversed than not most cases are affirmed. So I guess you're more likely to win if you're the appellee. But there's nothing quite like getting a decision flip like you feel like you've done your job. So
Tim Kowal 7:15
what are some of the things that you bring to the table as an appellate attorney, that most commonly kind of make the trial attorneys eyes light up in recognition? Yeah, this is why we wanted to bring an appellate perspective into the room. Did you see any things that commonly come up things that are commonly missed perspectives that maybe are commonly not seen until you bring an appellate attorney into the picture?
Adam Unikowsky 7:34
You know, I just think it's sometimes helps to have a fresh eye on some of the issues. I mean, you know, it's easy for trial lawyer, you've been locked into a case for several years to, you know, get upset about maybe certain factual rulings or some procedural rulings that may actually have been totally unfair in context, but which the appellate court isn't really going to care about at the end of the day, they'll say, no abuse of discretion, have, you know, and so I think what the appellate lawyer does is try to pick apart the issues of law that are reviewed de novo, it may not seem particularly salient to a trial lawyer who's battling over the facts and gauging emotions practice over discovery, but it's the kind of issue of an appellate court might be interested in.
Tim Kowal 8:11
Yeah, yeah, you have to start moving the case from day one, you know, asking the questions that you think a panel is going to ask, What about bringing developing a theme of the case? Do you find that when you come onto a case, the trial team already has a theme that you figure can carry forward onto the appeal? Or do you feel that you often have to redevelop or redesign a theme of the case for the appellate arguments?
Adam Unikowsky 8:31
Yeah, the trial teams usually have a theme, because, you know, they're always planning to bring the case to a jury. I mean, our firm as with many firms, like, you know, when you start the case, you're assuming you're going to try the case, and the case goes away, it goes away. But from day one, that's how you're laser focused. And of course, you have a jury that hears the case, they're going to fall asleep, unless you have a pretty clear theme. You know, unfortunately, sometimes the theme for the appeal may not be the same, like the justice and fairness of the situation, may be relevant to factual findings. And then on appeal, it's actually really the interpretation of the contract. So you know, you've got to just make sure to preserve your issues, and know what you're going to raise to the jury and know what you're preserving in the record.
Tim Kowal 9:08
Yeah. Let's talk briefly about do you have any favorite briefing tips, any patented you add a muta kowski tricks that you like to use commonly in your briefs?
Adam Unikowsky 9:17
I always like writing an intro, I mean, a punchy and show that sets forth my case, I like to do that. You know, first, I always try to be under the word limit by a lot if I can. I really don't, you know, like, because I find when I'm reading briefs, when I'm reading cases, like my eyes start to glaze over around like page 37. So my philosophy is like, as short as possible, and if you have to go to the word limit, fine. That's certainly that's important to me. And the other thing is clear, unclear writing usually implies unclear thinking, like when you're reading something, and it's not well written, oftentimes, because you don't really understand your argument and you're just like, putting it there and trying, you know, so like, for me, that's the most important thing. Like if something is unclear, it's probably because I haven't really figured it out. You know, and when I proofreading a brief or when I'm writing and reading over that's, that's what I tell myself anyway,
Tim Kowal 10:03
I'm writing that down. If something is unclear, it's probably because I haven't figured it out. I think that really rings true for me, I like to sometimes come back to my appellate briefs that I filed. I mean, in an ideal world, I'll have, you know, put it in the drawer for a couple of weeks before I file it and make sure it still makes sense. But sometimes it doesn't happen until a couple of months after I filed it. And I bring it out to see if it still makes sense. And every now and then you find something I don't know if that argument quite makes sense. And it doesn't make sense to me. Now, it probably wasn't exactly clear in my mind, then. And it probably has a very low percentage chance of persuading anybody, but it's unclear.
Adam Unikowsky 10:36
And the other thing is also collaboration is very important. Like I always assume I've missed something, even if I'm certain I didn't miss an argument. I'm assume I've done it, I've missed it. And I always like to collaborate with another person, you know, associated another partner. And for the associates listening out there, don't just say this is great. You found a typo. Like the purpose of the review is to tell me what I've missed. So I mean, collaboration is really essential to to getting good product.
Tim Kowal 10:58
And we covered Adams briefing tips. What about oral argument? You know, you've got nine Supreme court appearances under your belt Do you have what goes into making an effective oral argument? And do you still get butterflies and making oral argument appearances other than in the Supreme Court? Now, you know, you're you've been up against the big nine is now all clear sailing, when you're in an Intermediate Court of Appeals
Adam Unikowsky 11:20
is definitely not clear sailing in any court. Judge can can end your career if you did something bad. So no, I get I mean, even if I'm in state court, or if I'm in an agency, I mean, I had an argument in the Patent Trademark Appeal Board, excuse me, a few weeks ago, or a few months ago, where I was like stream butterflies. It was a zoom argument and like, in front of a judge deciding the case. So it's so nerve racking, the tips for me are just like, answer the questions directly. And like, don't try to get back to what you're going to say because they don't care. You've already read it. And so like you get a question, just don't filibuster answer directly. And if you never in your 10 minutes, get back to like the prepared remarks, don't care about them anyway.
Tim Kowal 11:59
I think that's a hard lesson for a lot of attorneys to get to, they've got their outline, and everything on there is gold, they have to cover them all.
Adam Unikowsky 12:05
So when I come to the podium, and I never bring any written materials, and no outline, because I worry, I'll use it as a crutch. I have my briefs there in case they're like, on page 46 of your brief, you said this, but other than the briefs that are on next to me, I don't bring a paper and by design, because I don't want to start reading. So
Jeff Lewis 12:20
even when you're in front of the United States Supreme Court, you walk up to the podium without prepared remarks,
Adam Unikowsky 12:26
yet no notes? Wow. Well, you know, it sounds better. You know, there's actually there's a funny argument from several years ago, where an attorney was clearly reading for like about a minute and then Justice Scalia, it's on Oh, yeah. But I forget what it's called, stops him. And he says, counsel, are you reading this? And there's this like, painful 22nd? Pause. And then Justice Breyer is like, sorry, I just just go on. So that was one of the most painful oral arguments are heard and I've striven to avoid? It's a strike with you. Okay. Yeah.
Tim Kowal 12:54
All right. Yeah. So yeah, the judges and justice hate read oral arguments that much that they will they'll stop the an advocate and mid sentence.
Adam Unikowsky 13:02
You don't want to do that. Yeah. And you're there to answer their questions. And like, if they see that you're stumbling, or stuttering, maybe they'll start asking you questions. So maybe it's an incentive for them to do that, you know? Yeah.
Jeff Lewis 13:10
Hey, can I ask what do you think of the new format in, you know, the post COVID era, the justices when they are asking questions down the line in more orderly fashion, as opposed to the old way? Here? They're everywhere. What do you think of that new question asking format.
Adam Unikowsky 13:26
So it's now kind of like a hybrid. So during COVID, when they're on the phone, it would just be down the line. Now they're back in the courtroom, and they do the free for all. But then there's questions at the end. So like the way it works, and I was like, they wait for Justice Thomas to ask question, if he has one, maybe a couple. And then there's the usual free for all. And then at the end, they go through all the justices and ask their questions. I think it's great. I see. It used to be quite dissatisfying, like your 30 minutes expired. And like, That's it, there's more questions and like, you know, if it goes on a little longer, so justices have a chance to ask whatever questions they have. I think it's all the better. I mean, you know, a few extra minutes. You know, I hope it improves the quality of the decision making, maybe it does. And so I think that's great.
Tim Kowal 14:04
Now, Adam, as I read from your bio, up at the top, you have undergraduate and graduate degrees in electrical engineering and computer science. Could you tell us a little bit more about that at the time you were studying electrical engineering and computer science? Did you have designs on going into law with that under your belt? Or was the path to law a turn that would come later on?
Adam Unikowsky 14:25
It was late so you know, I started as a physics major, which I ended up actually being a double major also, I really enjoyed that. And then I thought, you know, maybe I should try something more practical. So I did electrical engineering, computer science, which is really just like computer engineering, computer science, which I enjoyed as well. Then I ended up getting a master's degree and I don't exactly know how I somehow got diverted to law school. I mean, I like to argue like everyone else, and I was on my school's to colleges debate team, believe it or not, and so I just kind of thought it would be an interesting career. And it was almost like I remember just like, one day on a lark. I was just like, Yeah, I'm gonna go to law school, but it wasn't you know, I didn't say Get out that way somewhere around senior year, I was just like, No. Let's give this a shot. And that's it.
Tim Kowal 15:05
Okay, so on a lark, I'll decide to go to law school and on a lark, I'll get Supreme Court clerkship and on a lark, all big nine, that oral argument appearances and the US Supreme Court.
Adam Unikowsky 15:15
Well, you know, but it's I mean, physics and also, you know, engineering is like, it's great. It's absolutely fascinating. I learned a lot, you know, I'll never regret.
Tim Kowal 15:23
Now that must play a role in your current quest to replace all human law clerks with artificial intelligence. Well, as your background play in there,
Adam Unikowsky 15:32
I mean, some of the people that computer people I met in college were amazingly brilliant, I have incredible admiration for many of the people that I encountered at the time. And you know, and now one can see the type of unbelievable technology that maybe not them specifically, but amazing computer engineers have produced I mean, gotta say, I don't know if you all use Chet GPT and things like that. And, you know, Ali's able to fusion and all these technologies, but like, it's mind blowing. I was absolutely flabbergasted when I saw their capabilities. Yeah.
Tim Kowal 16:01
Okay, so Adam Unocal skis article on his substack blog, which will cite in the so in the show notes is should AI replace law clerk. So here's the pitch from the blog post. And then this is after Adam offers some disclaimers about how AI probably should not replace judges. He goes on to say, here's the price of the of the piece. However, in the shorter term AI could and in my view, should serve the same role that law clerks play. Now the AI would review the briefs, summarize the arguments and make a recommendation as to how the case should come out. After the judge decides how the case should be resolved, the judge could summarize the rationale to the AI and the AI could prepare a draft of the judicial opinion based on that rationale, the judge would review and edit the AI would cite check and offer further comments. And the opinion would then be released and quote. So can you expand on that a little bit that makes it sound fairly practicable, and not terrifying? You know, if it's used as a tool to supplement what the judge is doing all along, which in a way is what the law clerks are doing now?
Adam Unikowsky 17:03
Yeah, I mean, the goal of the judicial process is to produce accurate, well written opinions. Swiftly, right. I mean, judges are public servants. And we want to have and their goal is to resolve disputes. I mean, it's great that law students have good career opportunities, I enjoy being a law clerk. But ultimately, as practical matter, judges are there to resolve cases accurately, fairly in an unbiased way and quickly. And the question is, Can AI help them do that? And I think it can, I mean, if we assume we'll soon get to a place where an AIS memo is going to be as good or as or better than a recent law students memo. And also the AI can complete this process in like 10 seconds, you know, I think it can do it. And if it can do it, it should do it.
Tim Kowal 17:46
Let's talk about today. Um, do you think the AI is good enough, right now, as we are having this conversation to completely replace human law clerks? Or do you think that at least one law clerk is still going to be there to check the work of the AI? Where do you where do you think we are, as of this moment in April 2023.
Adam Unikowsky 18:05
So like, I kind of feel like it keeps changing, like every week, like, you know, when I wrote this post, still GPT three now, or GPT, four, which does all kinds of interesting things. And, you know, there's competition. So like, I haven't even checked the most recent data is like a month ago that I worked with, are not ready quite ready for primetime yet. But I wouldn't be surprised. And I haven't even tried some of the newer stuff. And then I, you know, within a year or two years, I wouldn't be surprised if like someone not knowing who wrote it, compared a law clerk memo to an AI memo would say the members better. It's more thorough, didn't miss anything, whatever.
Tim Kowal 18:43
Yeah. Yeah, how do we have to share I mean, because we are fairly early on in this AI endeavor, while the share an anecdote because maybe that's about all we have at these early stages. I've run some chat GPT driven memos actually, through co counsel, the case text product, and they have been great at helping me get up to speed for a new client call, for example, they send me a brief or an opinion, a judicial decision that they'd like to review. And I can run it through and they can summarize it, or I want to know about a particular legal issue. So I have given me a quick and dirty summary of California case law on a particular issue. And it gets me up to speed, you know, in five minutes, or otherwise, it might have taken me half hour or an hour and with the assistance of a junior associate, but there are limitations that I spotted. I had a conversation this morning, where I made use of a couple of quick memos that I ran just five minutes before the phone call. But then there was one case that the potential client told me now here's the, you know, the definitive case in this area, and I go back to these memos that I ran, and it hadn't talked about that case. I don't know how I missed it. But now of course there are there going to be oversights and incorrect things in a human law clerk or junior associates memo as well. But yeah, I guess maybe the point is that AI is not a panacea, where where it's it's going to have advantages and that it's it's going to be faster, certainly faster, more accurate. In some ways, but maybe less accurate in other ways, and how do we ferret that out? So we know how to compensate for its deficiencies?
Adam Unikowsky 20:07
Well, that depends on on what his deficiencies are, and there's going to slowly decrease. So like, I mean, you know, if the AI will miss things, but if it misses fewer things than a human, then I think you want to use the better rather than the worse, right? I mean, nothing is perfect, but you want to, you know, human, certainly human law clerks, you know, reflecting on that period of my career, I was my 20s, I was pretty dumb. I don't really trust the former version of myself to provide. To be honest, not only that, but you know, the AI is have read every single case, I mean, at least once wants to train on the full database. I don't know if that's happened yet. But eventually, they'll train and help us law and they'll have read every single case and memorize it, that's certainly a useful feature. And you know, and it's cheap. And it's also unbiased, right? I mean, the AI and its humans may have unconscious or implicit biases, based on you know, the characteristic of a litigant or a lawyer, and then the A won't have that because computer and I think that's another advantage
Jeff Lewis 21:00
out of I've got a couple of questions. And part of my ignorance, I never clerked anywhere. So I have maybe a romanticized view of the process. Maybe it's a little less transactional than what you laid out in your article. Is there something to a judge in a law clerk interacting and a judge, maybe stepping down from the ivory tower and learning from a clerk? And vice versa? That would be lost if AI replaced a law clerk?
Adam Unikowsky 21:27
Well, to some extent, yeah. I mean, I had a great time clerking for both the judges, I got along with them. Well, we had a lot of very interesting conversations, and I don't know what they got out of it. But I got a lot out of it, that's for sure. You know, but I'm pretty practical person. And the judges job is to decide cases, according to law, especially in lower courts. It's just mastering the facts in the record and the relevant legal principles. The other thing is, you can still have the law clerks. I mean, they're not that expensive. law clerks don't make a huge money. You know, it's just the types of things they'll do might be different, maybe you'll have fewer of them. And you I mean, like, you can serve law clerks. If the judge wants to hang out with a younger person, that's fine. You know, I mean, a lot of being a law clerk, as you know, sitting in your office, and like figuring out some issue you've never seen before, right? And when you graduate from law school, or anything, and then make mistakes, that's a pretty typical awkward day. And so I'm not sure we should romanticize that either. I guess.
Tim Kowal 22:20
Are you familiar with the the technique, I guess you call it a prompting the AI prompting chat GPT. I'm not an expert in this. I've not tried it myself. But I guess the idea would be, you can give a search prompt to chat GPT say, you know, tell me if the International Shoe standard is met in this case, such that minimum contacts are established when the defendant was served at an airport, you know, when traveling in the state, something like that, but then you could give it further prompts that Oh, and this was just for a passing through a layover flight. And also, by the way, he has no other context, I guess you just keep feeding it more information and more prompts, scale it into the right jurisdiction, oh, this is a state court that's going to be deciding it rather than a federal court or what state it is keep getting more and more specificity in the prompt, and it will massage and give you different or better results. And I wonder if that is going to wind up playing a big role in how to give proper prompts to chat GPT, whatever the AI interface technology is, and do attorneys law students, law clerks need to be trained better, and how to use how to get the best results out of AI.
Adam Unikowsky 23:26
So you're right, the AI can remember your prior answer and modify its answers based on new information. And in fact, there's always websites explaining interesting ways to get chopped up to do what you want, like, you know, answer this, assuming I'm an educated person in the voice of a college professor or something, it will do it actually, you know, in the voice of an economist or something, if you like that mode of thinking. So yes, I think that hasn't technology improves, I, we will have to learn to use it in a way to harness its capabilities.
Jeff Lewis 23:57
Adam, I'm still stuck. I'm sorry, I'm still stuck on stepping up to the podium without notes. And I'm wondering, preparation must be super important for you, as AI changed the way in which you prepare for oral argument?
Adam Unikowsky 24:11
No, I haven't really used AI in my practice. It's more just been like, Whoa, this is neat. Like, I wouldn't say that I use it every day, I find that if you're trying to read something during an argument, then you're losing or you're winning, and they're not listening anyway, in which case, you just just stop because you're gonna you know, so I feel like, of course, you have the briefs there, because they'll just ask you something about the briefs. And you need to say, oh, at page 38. Let me look at the footnote. But you know, yeah, meaning, you know, so that's the answer your question. Yeah,
Jeff Lewis 24:40
I gotta tell you, you know, the courts we practice in mostly California state courts. 90% of the time, the opinions already written in full and much of oral argument, I think is the chance to play up policy or the why and the passion arguments. So I don't see much value in reading. But I gotta say I always have little cards with notes like little segment notes in case the hard question. comes up that I don't really want to advance, but guess I have to answer it. Interesting. Interesting. Have you played around with chat GBT to do a mock oral argument? I've done that. It's really fascinating.
Adam Unikowsky 25:11
I haven't, but it's good idea. Yeah.
Tim Kowal 25:14
Yeah. Along those lines, I've been meaning to check up to see if I remember years ago, Josh Blackman was involved with the project to catch him going off years old memory now. But I think his project was to try to feed into an AI system, all of the past opinions of various Supreme Court justices to kind of build a an ideological profile, so to speak, so that the AI could prognosticate on how a particular justice would come out in a particular case with particular facts and law at issue. And gosh, again, this was years ago that remember, he was explaining this project, and I haven't kept up with it. But I wonder I mean, that project must be supercharged by this point. I wonder if you're going to the United States Supreme Court, or for that matter, any Intermediate Court of Appeals seems like AI, probably not too far off, if we're not there already, where the AI could probably prognosticate how this particular constituent panel would come out. If there is a on a certain issue, even if it is a an ideologically charged issue.
Adam Unikowsky 26:09
I think I mean, that sounds like a particularly high degree of difficulty. But I mean, you can imagine some cases. So consider something where, you know, certain judges have certain views, a qualified immunity or something. All right, I can certainly see a world in which the AI is capable of like comparing the facts of one case, to the relevant legal standard, you know, figuring out the level of generality, the legal standard, then looking at the judges qualified immunity cases and saying, Okay, this is a qualified immunity Hawk, it's really got to be identical to the facts of the prior cases. And if it's more of a dove, well, then, you know, it's sort of in the same ballpark close enough. And so you can imagine that I mean, like, I think it's many years off to really, you know, read all the judges judicial opinions, and then read a brief like, that strikes me is hard. But look, my philosophy with this stuff is if a human can do it, then there's nothing special about the human brain that can't be replicated, if you have a bunch of circuits rather than neurons. And so, we mean, you are capable of reading a brief and saying, Hey, I read this judge's decisions, this is what I think the judge is going to do. And if we can do it, then I don't see any reason why computer can't do it, you know, with the as the technology improves,
Tim Kowal 27:15
yeah, that's gonna be the trick is that will be the space to watch to see if there is something that AI consistently for whatever reason doesn't seem to be able to do a certain thing that we humans can do. That will be the space to watch to see what is it that makes us humans special. If anything, I think that that may be a neat opportunity that AI gives us is that we're not just piecing together legal arguments, AI, now we've got, you know, their AI breath hot on our neck, that they're doing what we're doing on our mediocre days, we have to make every day you know, we have to push that envelope. So we could do that thing that AI can't do. Yeah,
Adam Unikowsky 27:50
I mean, we know a lot of lawyering not all but a lot of lawyering is a little bit mundane. I mean, it's applying the facts of the law, certainly a lot of judging. And unpublished opinion is applying the facts of the law. Like, I'll agree that an AI may not be able to come up with incredible scientific advancements prove amazing math theorems or whatever. But like how much of judging or what law clerks do, or how much of what we do every day is really quite at that level of of intellect. So I don't think lawyers will be replaced, but many tasks we do, I think, can probably be automated. That's what I think.
Tim Kowal 28:21
Yeah. You mentioned something about, you know, when you were saying that maybe anything that we human lawyers can do an AI can do. And I thought, Well, what about when you're trying? You're asking a specific research question, like I was asking you a research question about to co counsel the other day about terminating sanctions? And can you get relief on that. And the result I got out of it, it was interesting, because some of the results, they come back with ones that are directly on point here was an issue of terminating sanctions. And here's how the court came down on it. But there was another one involving here was, you know, attempt to disqualify an expert. And even though it's the results that even though it's not right on point to the research requests, it is analogous to, you know, because it effectively ended the case when this expert was excluded or something like that. Wow, that's pretty creative. You know, I would not have gotten that on a Boolean search result through traditional legal research tool.
Adam Unikowsky 29:07
I mean, the way these AIs work is they're not really creative. I mean, they're basically taking a huge database and figuring out using some probabilistic model what the most likely next word is based on a bunch of input words, based on huge training dataset. But you know, when we're lawyers, in many cases, not all, that's what we're doing. We're looking at a set of facts and looking at how prior courts have evaluated similar facts. And in fact, when you're writing a brief, you kind of don't want to have to be graded, right. It's much easier if you say there's a case that's directly on point and it forecloses plaintiffs or defendants claims. In fact, the most effective briefs are like I wrote in the blog entry that when a court calls your arguments, creative, it almost always means they've just rejected your argument. And so for many of the legal tasks, I don't know what percentage but many of them seems to me that these models of just you know, looking at prior decisions and doctrines Send materials and just figuring out what the next word should be based on prior cases is kind of what it should do if you speak.
Tim Kowal 30:06
Yeah, that's an interesting point about trying to be as uncreative as possible. You only want to be creative where you don't have any good arguments, you know, engineering Maxim, you want to fulfill your project with as few innovations as possible.
Adam Unikowsky 30:19
Exactly. You know, it's so much easier when you have a case that forecloses the other side, it's one paragraph, you move on to the next section, right? No policy arguments. No extrapolation. It's nothing. You went. And you know, so that's the best brief. Yeah,
Tim Kowal 30:35
well, yes. So was your point there that will AI be able to make those creative arguments, the policy arguments, this the social ramifications of a judicial holding going a certain way?
Adam Unikowsky 30:45
I mean, AI is actually pretty good already making policy arguments, you ask it, like, give me five reasons why, you know, this is better than that. And it'll it'll kind of do it actually fairly well, because of its mastery of lots of policy arguments out there. But I think AI may like your your nightmare, when you're a breed when you're a brief writer is like, screwing up, like not finding the right case. Right? Like, there's sometimes there's just an answer that you haven't figured out. And, you know, that's something the AI will help you with. And the other thing is, when you're writing a brief, I mean, you want it to be a little bit punchy. But I mean, you know, it's got to be a little boring to like, it can't be like, you know, because you're right, you don't have your panel is going to be it's written, you know, in this sort of calm, methodical style, you know, rather than tons of possess, like got to limit the possess, and that's the kind of writing style that the AI is good at. So like, you know, maybe AI won't be a spectacular novelist, but nor are most lawyers, nor should they be, at least in their briefs.
Tim Kowal 31:39
Right. Okay, Adam, let me go back to your article. Again, the article is should AI replace law clerks? In your article, you anticipate the objection? So you're telling me there aren't any harms to letting AI tell judges what to do? And Adam's answer is, yes, that is basically what I am telling you. Could you elaborate on that? Well, the word
Adam Unikowsky 31:58
basically is doing a lot of work there. You know, I think that one important thing is not saying that the AI is going to replace the law, the judge, it's an assistant and the judge still has to exercise his or her independent judgment. And so the benefit of the AI is that it's harnessing the the human abilities with the computer's ability, like AI does some things better than people, right? Has AI knows more stuff, right? than you do? And I do. And anyone does, because AI has read every single case, AI may like judgment, but number one, the judge is there to exercise judgment. That's why it's called the judge. And number two, I'm not 100% persuaded law clerks have a particularly good judgment either reflecting on my own experience.
Tim Kowal 32:37
Yeah, yeah, that's the things that comes through in the rest of the analysis. In your post some of the reasons that you give for embracing a I amount of something on the order of Well, there are we have some drawbacks to our current system, our current way of doing things. So for instance, you mentioned that judges already don't read all of the cases, they do rely to some extent on summaries provided by their clerks. And so, you know, if they're relying instead on summaries provided by AI, then, you know, your judge is really losing anything. Yeah, maybe they're gaining something.
Adam Unikowsky 33:10
It depends on the judge, like Justice Scalia would actually read every single case cited any opinion that he would write, he was like, obsessed with not citing a case incorrectly, he would really do that. But you know, it's easier to do that when you're in the Supreme Court hearing 60 cases a year than when you're on the Court of Appeals hearing that 1000s. And certainly trial courts are just too. And so you know, as you say, if the judge is going to rely on a summary, there are advantages to on an AI to complement the judges abilities. That's right. You also
Tim Kowal 33:38
mentioned or anticipate the objection, that well, AI is a black box. So we don't know what we're getting in your response to that as well. law clerks are already invisible, we don't have any influence on the decisions or the biases, the inputs that the that are provided by the current human law clerks.
Adam Unikowsky 33:57
Well, so there's that, like, you got an oral argument that you're doing. And then you see these, you know, 27 year olds and formal clothes and you're like, okay, then with blank stares, you know, not, it's not clear why I feel that much more connected to them. Plus, you don't really understand the human brain either. I mean, no one really, I don't really understand how the AI works. I've read these discussions, you know, Transformers and whatever, I still don't fully understand exactly what it's doing. But you know, the creative process is also quite mysterious. I think just the output that matters if the AI produces a consistently great set of memos by
Tim Kowal 34:29
Yeah, you know, and it's funny that you say that you don't understand how it works. It makes me feel a little bit better because I also don't understand how it works and I don't have any advanced degrees in electrical engineering or computer programming, but for that matter, you're also in good company because the co founder of case Tech's says he doesn't completely understand how it works either to some extent, you know, they program some of these synapses, for lack of a better term, these these algorithms and then they're kind of off to the races. It's almost like self learning in a way he puts he has put a lot of things in air quotes, you know, when it says that AI is reading or understanding language, yes to put air quotes around reading or understanding, it's not quite reading or understanding in human in the same way that humans do it. But whatever way it's doing it, it's producing useful results that do effectively mimic human work product.
Adam Unikowsky 35:17
Yeah. So I read these articles about the transformers and all that stuff, you know, you read it, and you're like, okay, but like, does it really work? And I guess, we don't really have an intuition about what happens when you train something like billions of times on like, billions of you know, gigabytes, or more terabytes and terabytes worth of data over a long time. At some point, it's like, you know, just use a silly example, like how at one language changes to another, like, how is it the case that like, you know, in 1000 years, you know, Latins split into French and Romanian, Spanish, Italian, and even more than that, like, whatever the original human language was, was split into many different languages that how could that have happened? And yet we know it did, factually. And part of it is that our brains can't process how like, you know, magnitude of time, and incremental changes can all add up together. It's not really related, but it just hard to process just sufficient amount of time. And the magnitude of computer operations that AI does eventually can I guess, produce these amazing results?
Tim Kowal 36:17
No, yeah, I like that analogy. Yeah, there are a lot of things, a lot of social phenomenon, we don't have either the right language or brains are just not made. Like for example, I'm thinking how for some people, the explanation to every social phenomenon they don't like is some form of conspiracy theory. Well, you know, some things may look like a conspiracy theory. And there may be something strange that we can't explain going on. But sometimes we lack any better way of explaining it other than conspiracy theory. And maybe it's true here where we lack any better way of explaining what AI is doing, other than it's reading and understanding and analyzing cases for us. But yeah, whatever is to transformers are code. Yeah, successfully mimicking? Well,
Adam Unikowsky 36:52
I took an AI course back when I was in college, we're talking like 2001, or something. And they taught us about neural nets. And like, I remember thinking it was lame, I was just like, okay, fitting something, or have and the whole neural thing seemed like a gimmick, and like, okay, so you train it on a bunch of things, and like, you fit a curve to something and maybe like, it'll get a little better at chess or something. But I guess you do, you know, I mean, I didn't expect that. So I have to say, I was surprised at what I'm seeing, in principle, look, the human brain can do it. So there's some kind of computation happening here. And there's no principle reason why, you know, silicon set of neurons can't replicate that although works very differently. But I wouldn't have believed it until I saw it.
Tim Kowal 37:29
You mentioned that you've read a lot of, you know, dystopian science fiction in your time, do you have any trepidations about the future of AI? is, are there any dark turns potentially, in our future?
Adam Unikowsky 37:39
You know, I read a lot of, you know, AI concern, and there's a whole community of people out there who think AI is very dangerous, and are really worried about alignment. And, you know, I read all that stuff, because it's very interesting. You know, I don't, I don't feel like I have a good intuition. Like, I had no intuition that AI could even accomplish this before it came out. So I'm not sure I have a very good intuition of like, what's going to happen? I think there's a benefit to having multiple API's, you know, made by different people to ensure that, you know, the biases of assembler developer, don't prop it, you know, that may be subtle don't propagate. You know, AI alignment is such an abstract topic. I don't even know what I don't know about this. So I'm, you know, just try to read it every week and and try to make sure I know what I don't know, and then try to know it.
Tim Kowal 38:23
Yes. Well, that does that fat give cause for some trepidation that we don't know what we don't know, you know, should we be careful in how much we rely on it? And maybe if we rely on it, we should do so only in small increments?
Adam Unikowsky 38:35
Yeah. I mean, that may be true. In some cases. I mean, with respect to you know, law clerks, you know, the judge discipline judge should rely only to a limited extent, only to the extent that it's persuasive. That's how a judge should treat a law clerks memo. You know, if you're persuaded by it, fine. But you know, you have to kick the tires, you can't assume that it's correct, you should look at the cases. And so if there's like that the level of review that we assume should happen, and I think probably does typically happen, I don't know about every single judge in the country, then I think it's fine to use the AI as an input to a human process, if it's being checked, which it should be, you know, I have another post about AI is being arbitrators, which I actually think may be a good idea in some ways, too. But of course, I wouldn't want an AI Supreme Court that would be that would be weird.
Tim Kowal 39:21
And could AI to the extent that it's processing and relying on decisions that are out there, some of which may have an ideological tinge? Could AI thereby produce ideological results? Is there a garbage in garbage out problem?
Adam Unikowsky 39:35
You know, so there's a lot of talk that Chechi Beatty is too left wing and things like that, you know, and then, you know, you see on the internet, like, Oh, it'll praise Biden, but one praise Trump and things like that, you know, so yes, you don't want an AI pushing you in a particular direction. Lord knows that lots of law clerks are ideological to write. AI is not alone in that regard. Number one, though, an ideological case. I'm sure the judge will be making his or her own decision. Like I think the fact that AI says, you know, some ideologically charged thing will necessarily persuade the judge the AI is most useful outside of that context of cases. And also the AI can be programmed to produce both sides of an ideological issue as well. So I don't anticipate, like dogmatic AI is in particular social issues, like moving judges any more than dogmatic law clerks.
Tim Kowal 40:24
Okay. All right. And last question, do you have any predictions for what we're going to see next? How do you see these things playing out? Do you? Do you predict that there are any judges who are going to follow your advice and start replacing their human law clerks with AI? Or do you think they need to start seeing some proof of comp more proof of concept before they do that? And what are the steps that you see playing out before your proposal might be adopted by some appellate judges?
Adam Unikowsky 40:49
Well, yeah, there's gonna be a lot of proof of concept. I mean, I think it'll start as like, the judge will have the former law clerks, and the judge may ask the law clerk to use AI, or the judge may use the AI himself or herself. Like, it'll be the judge and the law clerks and the AI just to see the stuff it'll do. And also, originally, it might be for like, little stuff, you know, find a case that says this. Or, you know, here's one sentence of that, you know, can I get fees for this? When do I have to apply? What's the deadline to apply for fees? What's the standard for fees? What happens if I applied late is your tolling, you know, whatever, things like that, and then say, like, write a little paragraph rejecting the argument for fees on the ground that it's untimely, if it is little stuff. And then if people are like, wow, that's doing it, and they check the case? And it seems right. Yeah. And then it'll inch forward.
Tim Kowal 41:38
Yeah, yeah. It's a brave new world. I think, you know, maybe we'll even start seeing AI, listening to our conversations and providing us live fact checks, and, you know, additional tidbits that we could put into our conversations. What do you think, Jeff, maybe we'll ask Pablo, if we can get that when it comes available and enhance our content? Right?
Adam Unikowsky 41:56
You have Alexa? Alexa was listening. I mean, it seems goofy. I was listening, but it was in your house. I mean, you say, hey, Alexa listens. Right? And so anything you say, can have Alexa, install this listening? Yeah, I have some problems with that, too. But it's if you're worried about that, then a lot of other things have to change.
Tim Kowal 42:11
Yeah. I was told by an Amazon engineer that it's, it's not listening. I don't know how it works. Whether it's listening only for the word Alexa, or hey, Alexa, whatever that code word is. But yeah, how do you trust it? I guess you'd have to ask it was so trusting. All right. Well, Adam Yuna kowski, thank you so much for joining us and talking about your article. Should AI replace law clerks. We'll have a link to that in the show notes. We'll be following your work on substack. And I hope you'll come back to the podcast again in the future.
Adam Unikowsky 42:40
I'd love to thanks so much for inviting me. Yeah, and that wraps up
Jeff Lewis 42:43
our episode. We want to thank case techs again for sponsoring our podcast and each week we include links to the cases we discuss these in casetext and listeners of the podcast can find a 25% discount available to them when they sign up a casetext.com/calp. That's casetext.com/calp.
Tim Kowal 42:58
And if you have suggestions for future episodes, or guests that we should include on the podcast please email us at info at cow podcast.com. And in our upcoming episodes like for more tips on how to lay the groundwork for an appeal when preparing for trial.
You have just listened to the California appellate podcast, a discussion of timely trial tips and the latest cases a news coming from the California Court of Appeal and the California Supreme Court. For more information about the cases discussed in today's episode, our hosts and other episodes, visit the California appellate law podcast website at ca l podcast.com. That's c a l podcast.com. Thanks to Jonathan Cara for our intro music. Thank you for listening and please join us again