WEBVTT
00:00:03.759 --> 00:00:06.799
Welcome to AI Tools for Practicing Lawyers.
00:00:07.040 --> 00:00:13.839
Practical, no-nonsense guidance to help attorneys put AI power to work in their practice right now.
00:00:14.880 --> 00:00:22.559
I saw three AI and litigation stories this week that at first I thought were about three different things.
00:00:22.800 --> 00:00:35.840
But the more I thought about them, the more I realized they may be telling us something very important about a problem practicing litigators are going to encounter more and more often.
00:00:36.079 --> 00:00:37.359
And that's this.
00:00:37.840 --> 00:00:43.039
How do you litigate against a pro se party who is using AI?
00:00:44.240 --> 00:00:52.240
Because generative AI may have changed the economics of pro se litigation in a pretty fundamental way.
00:00:52.719 --> 00:01:01.359
Person who doesn't have a lawyer can now have something that looks remarkably like an inexhaustible junior associate.
00:01:01.520 --> 00:01:16.799
It doesn't get tired, doesn't charge by the hour, and perhaps most importantly, and most dangerously, it will happily draft another motion and another and another.
00:01:17.040 --> 00:01:26.640
There's a remarkable lawsuit pending in federal court in Chicago involving Nippon Life Insurance Company of America and OpenAI.
00:01:26.879 --> 00:01:36.799
According to Nippon's allegations, a woman named Graciella Della Torre had a dispute over long-term disability benefits, and she had a lawyer.
00:01:36.959 --> 00:01:41.280
Her case was settled and then dismissed with prejudice.
00:01:41.680 --> 00:01:50.079
Later, she became dissatisfied with what her lawyer had told her and really the entire engagement itself.
00:01:50.319 --> 00:01:55.519
So she uploaded an email from her lawyer to ChatGPT and asked about it.
00:01:55.760 --> 00:02:02.400
According to Nippon's complaint, ChatGPT reinforced her concerns about her lawyer's advice.
00:02:02.560 --> 00:02:12.800
And look, anybody who's using ChatGPT knows that it tends to echo back to you in much more sophisticated syntax what you told it.
00:02:13.120 --> 00:02:20.960
So she fired the lawyer and began using ChatGPT to generate filings attempting to reopen the case.
00:02:21.199 --> 00:02:26.319
Now, the federal judge rejected that effort, but that didn't end the litigation.
00:02:26.479 --> 00:02:30.719
She filed another lawsuit and continued filing motions and notices.
00:02:30.960 --> 00:02:39.680
And Nippon claims that dealing with all of this ultimately cost it$300,000.
00:02:40.960 --> 00:02:51.919
Nippon has now sued OpenAI, arguing, among some other things, that ChatGPT was effectively engaged in the unauthorized practice of law.
00:02:52.319 --> 00:02:57.039
Of course, OpenAI defends by saying the lawsuit has no merit.
00:02:57.439 --> 00:02:59.520
I don't know who's going to win this lawsuit.
00:02:59.599 --> 00:03:02.560
And that's not really the part of the story that interests me.
00:03:02.719 --> 00:03:11.280
What interests me is the$300,000 that Nippon spends battling against this pro se litigant.
00:03:11.439 --> 00:03:13.520
And I'm going to tell you something right now.
00:03:13.680 --> 00:03:20.639
Everybody in the law, everybody in the bar is going to be clutching their pearls, saying that this is a horrible thing.
00:03:20.960 --> 00:03:29.759
And looking at it objectively, I don't think it's such a horrible thing as a human involved in the world.
00:03:30.080 --> 00:03:39.680
Because huge corporations have really denigrated their interaction with consumers.
00:03:40.000 --> 00:03:47.280
And if consumers get fed up and start fighting back, I don't know that that's such a terrible thing.
00:03:47.439 --> 00:03:51.680
But whether or not it's a terrible thing, it's happening in the world.
00:03:52.000 --> 00:03:56.800
And we in the bar need to figure out how to address it.
00:03:57.039 --> 00:04:00.319
Because imagine you're the lawyer representing Nippon.
00:04:00.479 --> 00:04:02.000
Your client's case was settled.
00:04:02.159 --> 00:04:04.000
It was dismissed with prejudice.
00:04:04.240 --> 00:04:07.520
Normally you think, that's it, we're done.
00:04:07.759 --> 00:04:21.439
Except now the former plaintiff has access to a machine that can generate legal arguments, motions, briefs, and new theories, essentially without any real cost.
00:04:21.839 --> 00:04:26.879
And every time you respond, she can ask the machine for another reply.
00:04:27.040 --> 00:04:33.040
And every time you file something, she can upload it to the machine and ask, how do I fight this?
00:04:34.160 --> 00:04:38.480
That creates an asymmetry lawyers haven't really had to deal with before.
00:04:38.720 --> 00:04:44.720
The pro se litigant used to have a natural limiting factor: time, knowledge.
00:04:44.959 --> 00:04:50.160
You know, pro se litigants used to have to beg lawyer friends to help them with their cases.
00:04:50.480 --> 00:04:51.680
Not anymore.
00:04:51.920 --> 00:04:57.120
And that means lawyers representing the other side may need a different strategy.
00:04:57.600 --> 00:05:06.240
So I've gathered what I think might be some useful tips for lawyers dealing with pro sey litigants.
00:05:06.480 --> 00:05:09.839
And this is going to accelerate.
00:05:10.480 --> 00:05:18.879
More and more people are going to take on bigger corporations without counsel.
00:05:19.360 --> 00:05:23.839
So I think the first lesson from the Neppon case is pretty straightforward.
00:05:24.079 --> 00:05:32.959
Don't assume you have to litigate every AI-generated argument as though an experienced lawyer deliberately selected it.
00:05:33.279 --> 00:05:38.560
AI is extraordinarily good at generating more arguments.
00:05:38.800 --> 00:05:49.439
If you respond to every argument with 10 pages, well, that litigant is going to paste those 10 pages into an AI tool and generate 20 pages in response.
00:05:49.680 --> 00:05:55.600
Then you respond to the 20, and pretty soon somebody has spent$300,000.
00:05:56.879 --> 00:06:06.000
I think the better strategy may be to force the litigation back onto procedure, the actual issues before the court.
00:06:06.319 --> 00:06:08.560
What requires a response?
00:06:08.879 --> 00:06:09.839
What doesn't?
00:06:10.079 --> 00:06:11.680
Can the issues be narrowed?
00:06:11.839 --> 00:06:13.600
Are page limits appropriate?
00:06:13.839 --> 00:06:16.079
Would a pre-motion conference help?
00:06:16.240 --> 00:06:21.279
Has the litigant begun repeating arguments the court has already rejected?
00:06:21.600 --> 00:06:26.319
Is it time to ask the court for an order governing future filings?
00:06:26.879 --> 00:06:40.639
Sort of like a pretrial discovery type order, but just with a little bit tighter guidelines and a little bit tighter allowances for litigants to file documents with the court.
00:06:40.879 --> 00:06:46.480
In other words, don't try to beat the AI at its own game by producing more words.
00:06:46.959 --> 00:06:49.360
Make the battlefield smaller.
00:06:49.680 --> 00:06:50.319
All right.
00:06:50.800 --> 00:06:54.800
Number two, verify everything.
00:06:55.360 --> 00:06:57.279
Here's another practical problem.
00:06:57.519 --> 00:07:02.160
Don't expect handwritten screeds being filed with the clerk.
00:07:02.319 --> 00:07:07.439
An AI-assisted pro se filing is gonna look pretty good.
00:07:07.600 --> 00:07:09.600
That doesn't mean any of it's right.
00:07:09.839 --> 00:07:14.079
We've already seen plenty of cases involving hallucinated authorities.
00:07:14.240 --> 00:07:29.199
So when you're on the receiving end of one of these filings, don't assume that just because the document looks good, that the case that's cited on page 17 actually says what that pro se litigant says it says.
00:07:29.439 --> 00:07:30.160
Check it.
00:07:30.399 --> 00:07:32.560
Don't just check whether the case exists.
00:07:32.720 --> 00:07:34.720
Check whether the quotation exists.
00:07:34.959 --> 00:07:40.000
Check whether the case actually stands for the proposition being asserted.
00:07:40.240 --> 00:07:46.399
And check whether the court, jurisdiction, and procedural posture make any sense.
00:07:46.639 --> 00:07:53.600
And remember, there are at least 21 ways that AI can hallucinate in a legal brief.
00:07:53.759 --> 00:08:00.480
I talked about those in an earlier field note that I'm going to refer to down in the show notes.
00:08:00.879 --> 00:08:03.199
But here's the other side of that rule.
00:08:03.439 --> 00:08:08.399
Don't assume the argument is wrong merely because AI helped to produce it.
00:08:08.639 --> 00:08:12.639
AI can help a pro se litigant find a legitimate argument.
00:08:12.879 --> 00:08:16.720
Your job isn't to prove that the person used ChatGPT.
00:08:16.959 --> 00:08:26.240
Your job is to separate the potentially meritorious argument from the enormous amount of material AI may have generated around it.
00:08:27.120 --> 00:08:31.519
Rule number three: create a record of the pattern.
00:08:31.759 --> 00:08:39.360
If the filings become repetitive, frivolous, or abusive, don't just keep responding forever.
00:08:39.600 --> 00:08:43.919
At some point, the pattern itself may become more important.
00:08:44.240 --> 00:08:48.000
And that's where another case from this week gets really interesting.
00:08:48.240 --> 00:08:56.480
A self-represented plaintiff named Matthew Elliott is suing the New York Bariatric Group in Connecticut State Court.
00:08:56.720 --> 00:09:02.320
Judge Walter Spader discovered something unusual in Elliot's filings.
00:09:02.559 --> 00:09:06.080
There were strange blank spaces on the page.
00:09:06.320 --> 00:09:14.320
When the court looked more closely, it found text printed in tiny white type on a white background.
00:09:14.480 --> 00:09:16.559
This was reported in Ars Technica.
00:09:16.879 --> 00:09:20.240
A human reader couldn't see it, but software could.
00:09:20.799 --> 00:09:27.919
The hidden text contained instructions directed to any AI system that might analyze the filings.
00:09:28.080 --> 00:09:31.919
I mean, frankly, I looked at this text, I couldn't understand it.
00:09:32.080 --> 00:09:40.639
I didn't think it actually accomplished what that pro se litigant really intended for it to accomplish, but it was in there.
00:09:40.799 --> 00:09:56.159
And Judge Spader concluded that Elliot was attempting something called prompt injection, essentially trying to give hidden instructions to an AI system so that the system would produce an answer favorable to Elliot.
00:09:56.480 --> 00:10:16.960
There was just one problem, which is that unlike the UAE, which has taken on nationwide artificial intelligence procedures in all of their courtrooms, the Connecticut judicial branch doesn't use AI to review or decide filings.
00:10:17.200 --> 00:10:23.120
So Elliot was trying to manipulate an AI system that wasn't even there.
00:10:23.919 --> 00:10:28.320
He told Reuters that he was trying to audit the court's processes.
00:10:28.960 --> 00:10:30.960
The judge was not into this.
00:10:31.120 --> 00:10:44.639
And after additional hidden material appeared in subsequent filings, some of which was really ludicrous, the court sanctioned Elliot by taking away his ability to file electronically.
00:10:44.960 --> 00:10:52.480
Future filings have to be submitted on paper, which might be a pretty effective strategy for dealing with AI.
00:10:52.720 --> 00:10:54.720
But here's my favorite part of the story.
00:10:54.960 --> 00:10:57.279
Judge Spader's not anti-AI.
00:10:57.519 --> 00:10:59.679
Most judges these days really aren't.
00:10:59.919 --> 00:11:07.039
In fact, he disclosed that he used AI himself to prepare the decision sanctioning Elliott.
00:11:07.200 --> 00:11:09.279
Wow, how meta can you get?
00:11:09.519 --> 00:11:16.720
His objection wasn't that somebody used AI, his objection was to misconduct involving AI.
00:11:16.879 --> 00:11:19.360
And that's an important distinction.
00:11:20.080 --> 00:11:22.080
And that brings me to tip number four.
00:11:22.240 --> 00:11:27.200
Don't assume you get discovery into the Chat GPT conversations.
00:11:27.440 --> 00:11:32.879
Now we get to something my former podcast guest, Matt Lafferman, has been writing about.
00:11:33.120 --> 00:11:42.799
Matt is a partner at Denton's, co-head of the firm's Legal AI Lab, and someone I've had on this podcast to discuss AI, privilege, and work products.
00:11:43.120 --> 00:11:50.879
He and Rick Shearer recently wrote about the developing case law concerning AI interactions and discovery.
00:11:51.200 --> 00:12:01.759
And there's a temptation, if you're litigating against an AI-assisted pro se party, to think, great, I want those ChatGPT prompts.
00:12:02.000 --> 00:12:05.840
Well, maybe, but don't assume you're entitled to them.
00:12:06.159 --> 00:12:07.279
In Warner v.
00:12:07.360 --> 00:12:16.639
Gilbarco, a pro se plaintiff, used a paid ChatGPT account to draft filings and analyze materials related to her case.
00:12:16.799 --> 00:12:21.519
The defendant sought broad discovery concerning her use of ChatGPT.
00:12:22.320 --> 00:12:24.559
The court said no.
00:12:25.200 --> 00:12:36.080
It treated her inputs as reflecting her internal analysis and mental impressions and held that the material was protected by the work product doctrine.
00:12:36.320 --> 00:12:41.919
The court made an observation that I think is going to become increasingly important.
00:12:42.639 --> 00:12:47.039
Generative AI programs are tools, not persons.
00:12:47.360 --> 00:12:56.320
So the fact that a litigant used an AI tool doesn't automatically turn the litigant's thought process into discoverable material.
00:12:56.639 --> 00:13:19.519
And by the way, if you're going to find that an AI chatbot, like OpenAI's ChatGPT, or Claude or Gemini, are tools, not persons, then that should give you a lot more comfort in being able to record and have AI transcribe and summarize your conversations with clients.
00:13:19.840 --> 00:13:25.120
You're just submitting that information into a tool, not a third party.
00:13:25.440 --> 00:13:27.120
Now, in Tremblay v.
00:13:27.279 --> 00:13:33.919
OpenAI, lawyers used ChatGPT in investigating their case, and the opposing party sought the prompts.
00:13:34.000 --> 00:13:35.919
They're always going to seek the prompts.
00:13:36.240 --> 00:13:48.960
The court concluded that the prompts were opinion work products because the questions crafted by counsel themselves revealed counsel's mental impressions about how to investigate the case.
00:13:49.279 --> 00:13:51.360
So we need to be careful here.
00:13:51.600 --> 00:13:53.519
AI can create evidence.
00:13:53.840 --> 00:13:56.399
AI can create discoverable records.
00:13:56.720 --> 00:13:59.759
AI can be used to create protected work products.
00:14:00.000 --> 00:14:05.120
So there's no rule that says, well, they use Chat GPT, therefore I get the chance.
00:14:06.000 --> 00:14:10.240
Rule number five, AI use isn't the misconduct.
00:14:10.399 --> 00:14:13.519
And that brings me to the third thing I saw this week.
00:14:13.759 --> 00:14:23.440
Judge Sarah Smith of the Third Judicial Circuit in Illinois issued a remarkably sensible statement explaining her own use of AI.
00:14:23.919 --> 00:14:35.039
Judge Smith says she may use AI for things like outlining, editing, formatting, grammar, organization, and summarizing identified materials.
00:14:35.200 --> 00:14:40.879
My guess is with that laundry list, she probably used AI to draft the order.
00:14:41.200 --> 00:14:43.759
But she draws a clear line.
00:14:44.080 --> 00:14:47.200
AI does not exercise judicial judgment.
00:14:47.360 --> 00:14:57.360
It doesn't weigh evidence, it doesn't assess credibility, it doesn't make factual findings, doesn't interpret the law, and it doesn't decide cases.
00:14:57.759 --> 00:15:00.879
It's the judge's job to do those things.
00:15:01.200 --> 00:15:12.000
She also says that AI-generated legal authorities, quotations, and citations must be independently reviewed and verified.
00:15:13.039 --> 00:15:21.039
I think that's important because it demonstrates something we should keep in mind when we're dealing with AI-assisted pro-sale litigants.
00:15:21.360 --> 00:15:23.679
Using AI isn't misconduct.
00:15:23.919 --> 00:15:36.799
Lawyers are using AI, judges are using AI, I obviously use AI, and increasingly sophisticated pro se litigants are going to use AI too.
00:15:37.039 --> 00:15:39.200
So the problem isn't the tool.
00:15:39.440 --> 00:15:43.840
The question is what the person does with it.
00:15:44.240 --> 00:15:46.080
And that may be the bigger lesson.
00:15:46.240 --> 00:15:53.279
For the last few years, lawyers have spent an enormous amount of time asking: is AI going to replace lawyers?
00:15:53.519 --> 00:15:57.039
Maybe that's not the most immediate litigation problem.
00:15:57.679 --> 00:16:00.639
AI doesn't have to replace the lawyer on the other side.
00:16:00.799 --> 00:16:02.960
It can do something much simpler.
00:16:03.200 --> 00:16:10.879
It can give the person who doesn't have a lawyer the ability to litigate as though they have an inexhaustible one.
00:16:11.279 --> 00:16:26.559
Not necessarily a good lawyer, and not necessarily a lawyer who exercises professional judgment, but a lawyer who never gets tired, never worries about the bill, and is always willing to draft one more motion.
00:16:27.600 --> 00:16:37.039
If you're representing the party on the other side, the answer probably isn't to produce more paper than the machine can produce because you can't.
00:16:37.360 --> 00:16:40.559
The answer is better litigation management.
00:16:40.799 --> 00:16:58.320
Narrow the issues, use procedure, verify the authorities, identify patterns of abusive filings early, get the court involved when necessary, take legitimate arguments seriously, and don't confuse AI use with AI misconduct.
00:16:59.279 --> 00:17:04.559
Because the next generation of pro se litigants may not look anything like the last one.
00:17:04.799 --> 00:17:20.480
And if an insurance company really can spend$300,000 litigating over a case that had already been settled and dismissed with prejudice, that's probably something practicing lawyers should start thinking about right now.
00:17:21.839 --> 00:17:29.599
My name is Ron Drescher, and this has been a field note in AI Tools for Practicing Lawyers.
00:17:29.839 --> 00:17:31.279
See you next time.
00:17:32.400 --> 00:17:36.880
That's it for today's episode of AI Tools for Practicing Lawyers.
00:17:37.039 --> 00:17:38.079
Thanks for listening.
00:17:38.240 --> 00:17:39.920
We'll see you next time.
00:00:03.759 --> 00:00:06.799
Welcome to AI Tools for Practicing Lawyers.
00:00:07.040 --> 00:00:13.839
Practical, no-nonsense guidance to help attorneys put AI power to work in their practice right now.
00:00:14.880 --> 00:00:22.559
I saw three AI and litigation stories this week that at first I thought were about three different things.
00:00:22.800 --> 00:00:35.840
But the more I thought about them, the more I realized they may be telling us something very important about a problem practicing litigators are going to encounter more and more often.
00:00:36.079 --> 00:00:37.359
And that's this.
00:00:37.840 --> 00:00:43.039
How do you litigate against a pro se party who is using AI?
00:00:44.240 --> 00:00:52.240
Because generative AI may have changed the economics of pro se litigation in a pretty fundamental way.
00:00:52.719 --> 00:01:01.359
Person who doesn't have a lawyer can now have something that looks remarkably like an inexhaustible junior associate.
00:01:01.520 --> 00:01:16.799
It doesn't get tired, doesn't charge by the hour, and perhaps most importantly, and most dangerously, it will happily draft another motion and another and another.
00:01:17.040 --> 00:01:26.640
There's a remarkable lawsuit pending in federal court in Chicago involving Nippon Life Insurance Company of America and OpenAI.
00:01:26.879 --> 00:01:36.799
According to Nippon's allegations, a woman named Graciella Della Torre had a dispute over long-term disability benefits, and she had a lawyer.
00:01:36.959 --> 00:01:41.280
Her case was settled and then dismissed with prejudice.
00:01:41.680 --> 00:01:50.079
Later, she became dissatisfied with what her lawyer had told her and really the entire engagement itself.
00:01:50.319 --> 00:01:55.519
So she uploaded an email from her lawyer to ChatGPT and asked about it.
00:01:55.760 --> 00:02:02.400
According to Nippon's complaint, ChatGPT reinforced her concerns about her lawyer's advice.
00:02:02.560 --> 00:02:12.800
And look, anybody who's using ChatGPT knows that it tends to echo back to you in much more sophisticated syntax what you told it.
00:02:13.120 --> 00:02:20.960
So she fired the lawyer and began using ChatGPT to generate filings attempting to reopen the case.
00:02:21.199 --> 00:02:26.319
Now, the federal judge rejected that effort, but that didn't end the litigation.
00:02:26.479 --> 00:02:30.719
She filed another lawsuit and continued filing motions and notices.
00:02:30.960 --> 00:02:39.680
And Nippon claims that dealing with all of this ultimately cost it$300,000.
00:02:40.960 --> 00:02:51.919
Nippon has now sued OpenAI, arguing, among some other things, that ChatGPT was effectively engaged in the unauthorized practice of law.
00:02:52.319 --> 00:02:57.039
Of course, OpenAI defends by saying the lawsuit has no merit.
00:02:57.439 --> 00:02:59.520
I don't know who's going to win this lawsuit.
00:02:59.599 --> 00:03:02.560
And that's not really the part of the story that interests me.
00:03:02.719 --> 00:03:11.280
What interests me is the$300,000 that Nippon spends battling against this pro se litigant.
00:03:11.439 --> 00:03:13.520
And I'm going to tell you something right now.
00:03:13.680 --> 00:03:20.639
Everybody in the law, everybody in the bar is going to be clutching their pearls, saying that this is a horrible thing.
00:03:20.960 --> 00:03:29.759
And looking at it objectively, I don't think it's such a horrible thing as a human involved in the world.
00:03:30.080 --> 00:03:39.680
Because huge corporations have really denigrated their interaction with consumers.
00:03:40.000 --> 00:03:47.280
And if consumers get fed up and start fighting back, I don't know that that's such a terrible thing.
00:03:47.439 --> 00:03:51.680
But whether or not it's a terrible thing, it's happening in the world.
00:03:52.000 --> 00:03:56.800
And we in the bar need to figure out how to address it.
00:03:57.039 --> 00:04:00.319
Because imagine you're the lawyer representing Nippon.
00:04:00.479 --> 00:04:02.000
Your client's case was settled.
00:04:02.159 --> 00:04:04.000
It was dismissed with prejudice.
00:04:04.240 --> 00:04:07.520
Normally you think, that's it, we're done.
00:04:07.759 --> 00:04:21.439
Except now the former plaintiff has access to a machine that can generate legal arguments, motions, briefs, and new theories, essentially without any real cost.
00:04:21.839 --> 00:04:26.879
And every time you respond, she can ask the machine for another reply.
00:04:27.040 --> 00:04:33.040
And every time you file something, she can upload it to the machine and ask, how do I fight this?
00:04:34.160 --> 00:04:38.480
That creates an asymmetry lawyers haven't really had to deal with before.
00:04:38.720 --> 00:04:44.720
The pro se litigant used to have a natural limiting factor: time, knowledge.
00:04:44.959 --> 00:04:50.160
You know, pro se litigants used to have to beg lawyer friends to help them with their cases.
00:04:50.480 --> 00:04:51.680
Not anymore.
00:04:51.920 --> 00:04:57.120
And that means lawyers representing the other side may need a different strategy.
00:04:57.600 --> 00:05:06.240
So I've gathered what I think might be some useful tips for lawyers dealing with pro sey litigants.
00:05:06.480 --> 00:05:09.839
And this is going to accelerate.
00:05:10.480 --> 00:05:18.879
More and more people are going to take on bigger corporations without counsel.
00:05:19.360 --> 00:05:23.839
So I think the first lesson from the Neppon case is pretty straightforward.
00:05:24.079 --> 00:05:32.959
Don't assume you have to litigate every AI-generated argument as though an experienced lawyer deliberately selected it.
00:05:33.279 --> 00:05:38.560
AI is extraordinarily good at generating more arguments.
00:05:38.800 --> 00:05:49.439
If you respond to every argument with 10 pages, well, that litigant is going to paste those 10 pages into an AI tool and generate 20 pages in response.
00:05:49.680 --> 00:05:55.600
Then you respond to the 20, and pretty soon somebody has spent$300,000.
00:05:56.879 --> 00:06:06.000
I think the better strategy may be to force the litigation back onto procedure, the actual issues before the court.
00:06:06.319 --> 00:06:08.560
What requires a response?
00:06:08.879 --> 00:06:09.839
What doesn't?
00:06:10.079 --> 00:06:11.680
Can the issues be narrowed?
00:06:11.839 --> 00:06:13.600
Are page limits appropriate?
00:06:13.839 --> 00:06:16.079
Would a pre-motion conference help?
00:06:16.240 --> 00:06:21.279
Has the litigant begun repeating arguments the court has already rejected?
00:06:21.600 --> 00:06:26.319
Is it time to ask the court for an order governing future filings?
00:06:26.879 --> 00:06:40.639
Sort of like a pretrial discovery type order, but just with a little bit tighter guidelines and a little bit tighter allowances for litigants to file documents with the court.
00:06:40.879 --> 00:06:46.480
In other words, don't try to beat the AI at its own game by producing more words.
00:06:46.959 --> 00:06:49.360
Make the battlefield smaller.
00:06:49.680 --> 00:06:50.319
All right.
00:06:50.800 --> 00:06:54.800
Number two, verify everything.
00:06:55.360 --> 00:06:57.279
Here's another practical problem.
00:06:57.519 --> 00:07:02.160
Don't expect handwritten screeds being filed with the clerk.
00:07:02.319 --> 00:07:07.439
An AI-assisted pro se filing is gonna look pretty good.
00:07:07.600 --> 00:07:09.600
That doesn't mean any of it's right.
00:07:09.839 --> 00:07:14.079
We've already seen plenty of cases involving hallucinated authorities.
00:07:14.240 --> 00:07:29.199
So when you're on the receiving end of one of these filings, don't assume that just because the document looks good, that the case that's cited on page 17 actually says what that pro se litigant says it says.
00:07:29.439 --> 00:07:30.160
Check it.
00:07:30.399 --> 00:07:32.560
Don't just check whether the case exists.
00:07:32.720 --> 00:07:34.720
Check whether the quotation exists.
00:07:34.959 --> 00:07:40.000
Check whether the case actually stands for the proposition being asserted.
00:07:40.240 --> 00:07:46.399
And check whether the court, jurisdiction, and procedural posture make any sense.
00:07:46.639 --> 00:07:53.600
And remember, there are at least 21 ways that AI can hallucinate in a legal brief.
00:07:53.759 --> 00:08:00.480
I talked about those in an earlier field note that I'm going to refer to down in the show notes.
00:08:00.879 --> 00:08:03.199
But here's the other side of that rule.
00:08:03.439 --> 00:08:08.399
Don't assume the argument is wrong merely because AI helped to produce it.
00:08:08.639 --> 00:08:12.639
AI can help a pro se litigant find a legitimate argument.
00:08:12.879 --> 00:08:16.720
Your job isn't to prove that the person used ChatGPT.
00:08:16.959 --> 00:08:26.240
Your job is to separate the potentially meritorious argument from the enormous amount of material AI may have generated around it.
00:08:27.120 --> 00:08:31.519
Rule number three: create a record of the pattern.
00:08:31.759 --> 00:08:39.360
If the filings become repetitive, frivolous, or abusive, don't just keep responding forever.
00:08:39.600 --> 00:08:43.919
At some point, the pattern itself may become more important.
00:08:44.240 --> 00:08:48.000
And that's where another case from this week gets really interesting.
00:08:48.240 --> 00:08:56.480
A self-represented plaintiff named Matthew Elliott is suing the New York Bariatric Group in Connecticut State Court.
00:08:56.720 --> 00:09:02.320
Judge Walter Spader discovered something unusual in Elliot's filings.
00:09:02.559 --> 00:09:06.080
There were strange blank spaces on the page.
00:09:06.320 --> 00:09:14.320
When the court looked more closely, it found text printed in tiny white type on a white background.
00:09:14.480 --> 00:09:16.559
This was reported in Ars Technica.
00:09:16.879 --> 00:09:20.240
A human reader couldn't see it, but software could.
00:09:20.799 --> 00:09:27.919
The hidden text contained instructions directed to any AI system that might analyze the filings.
00:09:28.080 --> 00:09:31.919
I mean, frankly, I looked at this text, I couldn't understand it.
00:09:32.080 --> 00:09:40.639
I didn't think it actually accomplished what that pro se litigant really intended for it to accomplish, but it was in there.
00:09:40.799 --> 00:09:56.159
And Judge Spader concluded that Elliot was attempting something called prompt injection, essentially trying to give hidden instructions to an AI system so that the system would produce an answer favorable to Elliot.
00:09:56.480 --> 00:10:16.960
There was just one problem, which is that unlike the UAE, which has taken on nationwide artificial intelligence procedures in all of their courtrooms, the Connecticut judicial branch doesn't use AI to review or decide filings.
00:10:17.200 --> 00:10:23.120
So Elliot was trying to manipulate an AI system that wasn't even there.
00:10:23.919 --> 00:10:28.320
He told Reuters that he was trying to audit the court's processes.
00:10:28.960 --> 00:10:30.960
The judge was not into this.
00:10:31.120 --> 00:10:44.639
And after additional hidden material appeared in subsequent filings, some of which was really ludicrous, the court sanctioned Elliot by taking away his ability to file electronically.
00:10:44.960 --> 00:10:52.480
Future filings have to be submitted on paper, which might be a pretty effective strategy for dealing with AI.
00:10:52.720 --> 00:10:54.720
But here's my favorite part of the story.
00:10:54.960 --> 00:10:57.279
Judge Spader's not anti-AI.
00:10:57.519 --> 00:10:59.679
Most judges these days really aren't.
00:10:59.919 --> 00:11:07.039
In fact, he disclosed that he used AI himself to prepare the decision sanctioning Elliott.
00:11:07.200 --> 00:11:09.279
Wow, how meta can you get?
00:11:09.519 --> 00:11:16.720
His objection wasn't that somebody used AI, his objection was to misconduct involving AI.
00:11:16.879 --> 00:11:19.360
And that's an important distinction.
00:11:20.080 --> 00:11:22.080
And that brings me to tip number four.
00:11:22.240 --> 00:11:27.200
Don't assume you get discovery into the Chat GPT conversations.
00:11:27.440 --> 00:11:32.879
Now we get to something my former podcast guest, Matt Lafferman, has been writing about.
00:11:33.120 --> 00:11:42.799
Matt is a partner at Denton's, co-head of the firm's Legal AI Lab, and someone I've had on this podcast to discuss AI, privilege, and work products.
00:11:43.120 --> 00:11:50.879
He and Rick Shearer recently wrote about the developing case law concerning AI interactions and discovery.
00:11:51.200 --> 00:12:01.759
And there's a temptation, if you're litigating against an AI-assisted pro se party, to think, great, I want those ChatGPT prompts.
00:12:02.000 --> 00:12:05.840
Well, maybe, but don't assume you're entitled to them.
00:12:06.159 --> 00:12:07.279
In Warner v.
00:12:07.360 --> 00:12:16.639
Gilbarco, a pro se plaintiff, used a paid ChatGPT account to draft filings and analyze materials related to her case.
00:12:16.799 --> 00:12:21.519
The defendant sought broad discovery concerning her use of ChatGPT.
00:12:22.320 --> 00:12:24.559
The court said no.
00:12:25.200 --> 00:12:36.080
It treated her inputs as reflecting her internal analysis and mental impressions and held that the material was protected by the work product doctrine.
00:12:36.320 --> 00:12:41.919
The court made an observation that I think is going to become increasingly important.
00:12:42.639 --> 00:12:47.039
Generative AI programs are tools, not persons.
00:12:47.360 --> 00:12:56.320
So the fact that a litigant used an AI tool doesn't automatically turn the litigant's thought process into discoverable material.
00:12:56.639 --> 00:13:19.519
And by the way, if you're going to find that an AI chatbot, like OpenAI's ChatGPT, or Claude or Gemini, are tools, not persons, then that should give you a lot more comfort in being able to record and have AI transcribe and summarize your conversations with clients.
00:13:19.840 --> 00:13:25.120
You're just submitting that information into a tool, not a third party.
00:13:25.440 --> 00:13:27.120
Now, in Tremblay v.
00:13:27.279 --> 00:13:33.919
OpenAI, lawyers used ChatGPT in investigating their case, and the opposing party sought the prompts.
00:13:34.000 --> 00:13:35.919
They're always going to seek the prompts.
00:13:36.240 --> 00:13:48.960
The court concluded that the prompts were opinion work products because the questions crafted by counsel themselves revealed counsel's mental impressions about how to investigate the case.
00:13:49.279 --> 00:13:51.360
So we need to be careful here.
00:13:51.600 --> 00:13:53.519
AI can create evidence.
00:13:53.840 --> 00:13:56.399
AI can create discoverable records.
00:13:56.720 --> 00:13:59.759
AI can be used to create protected work products.
00:14:00.000 --> 00:14:05.120
So there's no rule that says, well, they use Chat GPT, therefore I get the chance.
00:14:06.000 --> 00:14:10.240
Rule number five, AI use isn't the misconduct.
00:14:10.399 --> 00:14:13.519
And that brings me to the third thing I saw this week.
00:14:13.759 --> 00:14:23.440
Judge Sarah Smith of the Third Judicial Circuit in Illinois issued a remarkably sensible statement explaining her own use of AI.
00:14:23.919 --> 00:14:35.039
Judge Smith says she may use AI for things like outlining, editing, formatting, grammar, organization, and summarizing identified materials.
00:14:35.200 --> 00:14:40.879
My guess is with that laundry list, she probably used AI to draft the order.
00:14:41.200 --> 00:14:43.759
But she draws a clear line.
00:14:44.080 --> 00:14:47.200
AI does not exercise judicial judgment.
00:14:47.360 --> 00:14:57.360
It doesn't weigh evidence, it doesn't assess credibility, it doesn't make factual findings, doesn't interpret the law, and it doesn't decide cases.
00:14:57.759 --> 00:15:00.879
It's the judge's job to do those things.
00:15:01.200 --> 00:15:12.000
She also says that AI-generated legal authorities, quotations, and citations must be independently reviewed and verified.
00:15:13.039 --> 00:15:21.039
I think that's important because it demonstrates something we should keep in mind when we're dealing with AI-assisted pro-sale litigants.
00:15:21.360 --> 00:15:23.679
Using AI isn't misconduct.
00:15:23.919 --> 00:15:36.799
Lawyers are using AI, judges are using AI, I obviously use AI, and increasingly sophisticated pro se litigants are going to use AI too.
00:15:37.039 --> 00:15:39.200
So the problem isn't the tool.
00:15:39.440 --> 00:15:43.840
The question is what the person does with it.
00:15:44.240 --> 00:15:46.080
And that may be the bigger lesson.
00:15:46.240 --> 00:15:53.279
For the last few years, lawyers have spent an enormous amount of time asking: is AI going to replace lawyers?
00:15:53.519 --> 00:15:57.039
Maybe that's not the most immediate litigation problem.
00:15:57.679 --> 00:16:00.639
AI doesn't have to replace the lawyer on the other side.
00:16:00.799 --> 00:16:02.960
It can do something much simpler.
00:16:03.200 --> 00:16:10.879
It can give the person who doesn't have a lawyer the ability to litigate as though they have an inexhaustible one.
00:16:11.279 --> 00:16:26.559
Not necessarily a good lawyer, and not necessarily a lawyer who exercises professional judgment, but a lawyer who never gets tired, never worries about the bill, and is always willing to draft one more motion.
00:16:27.600 --> 00:16:37.039
If you're representing the party on the other side, the answer probably isn't to produce more paper than the machine can produce because you can't.
00:16:37.360 --> 00:16:40.559
The answer is better litigation management.
00:16:40.799 --> 00:16:58.320
Narrow the issues, use procedure, verify the authorities, identify patterns of abusive filings early, get the court involved when necessary, take legitimate arguments seriously, and don't confuse AI use with AI misconduct.
00:16:59.279 --> 00:17:04.559
Because the next generation of pro se litigants may not look anything like the last one.
00:17:04.799 --> 00:17:20.480
And if an insurance company really can spend$300,000 litigating over a case that had already been settled and dismissed with prejudice, that's probably something practicing lawyers should start thinking about right now.
00:17:21.839 --> 00:17:29.599
My name is Ron Drescher, and this has been a field note in AI Tools for Practicing Lawyers.
00:17:29.839 --> 00:17:31.279
See you next time.
00:17:32.400 --> 00:17:36.880
That's it for today's episode of AI Tools for Practicing Lawyers.
00:17:37.039 --> 00:17:38.079
Thanks for listening.
00:17:38.240 --> 00:17:39.920
We'll see you next time.
