Latest / AI Tools for Practicing Lawyers / Field Note: Even Biglaw Gets The AI Hallucination Blues

Transcript

Field Note: Even Biglaw Gets The AI Hallucination Blues · AI Tools for Practicing Lawyers

• 0:03
  SPEAKER_01: Welcome to AI Tools for Practicing Lawyers.
• 0:07
  SPEAKER_01: Practical, no-nonsense guidance to help attorneys put AI power to work in their practice right now.
• 0:14
  SPEAKER_00: Many lawyers still think that AI hallucinations are what happens to careless lawyers or small firms or solo practitioners in their tiny little offices or people who just don't know what they're doing.
• 0:29
  SPEAKER_00: But now we've had a really fresh reminder that even elite firms can get burned.
• 0:36
  SPEAKER_00: Even big law gets the AI hallucination blues.
• 0:39
  SPEAKER_00: Today I'm gonna connect something huge that just happened in the world of AI hallucinations to a field note that I did a few weeks ago of Confession of an AI Hallucinator, where I explained how smart lawyers under pressure can be lured into trusting fake authority.
• 1:02
  SPEAKER_00: So, you know, I'm constantly reading about AI news to help prepare for this podcast.
• 1:11
  SPEAKER_00: And sometimes I forget that the whole world isn't obsessed on AI in the legal world.
• 1:18
  SPEAKER_00: So everybody may not be fully drilled in on that AI can sound authoritative even when they're wrong.
• 1:27
  SPEAKER_00: Maybe even especially when they're wrong.
• 1:30
  SPEAKER_00: And the fake citations that they push out, they look real.
• 1:34
  SPEAKER_00: And it's not only fake citations, but maybe even more perniciously, it's fake holdings or fake rationale or fake language in actual citations that they cite that that may not get picked up.
• 1:50
  SPEAKER_00: Lawyers under deadline pressure absolutely become vulnerable.
• 1:54
  SPEAKER_00: So this kind of stuff is absolutely more likely to happen in emergency last-minute filings.
• 2:01
  SPEAKER_00: And that's exactly what happened to the 850-plus lawyer firm of Sullivan and Cromwell in a case that just got released last week, where they sent a letter to the judge after they were exposed at having added all sorts of hallucination mistakes to an emergency brief that was filed on April 9th in the United States Bankruptcy Court for the Southern District of New York.
• 2:34
  SPEAKER_00: They am they they first of all they were caught, and that's a bad look.
• 2:40
  SPEAKER_00: They didn't catch the bad citations, their opposing law firm caught the bad citations.
• 2:47
  SPEAKER_00: And there was some discussion with opposing counsel, and then Sullivan and Cromwell had to write a letter of apology that they filed by ECF, basically saying, please don't sanction us.
• 3:01
  SPEAKER_00: We're really good guys, but this one got away from us.
• 3:05
  SPEAKER_00: We have AI hallucination protocols in place, and we just didn't follow them.
• 3:12
  SPEAKER_00: I don't know what the upshot is going to be on this case, but it's all over the big law news.
• 3:18
  SPEAKER_00: And I I wanted to pass it along to you.
• 3:21
  SPEAKER_00: Now, what they did in the Sullivan Cromwell case is they added a Schedule A to this letter of apology.
• 3:30
  SPEAKER_00: And this letter of apology is going to be posted on our website under the deliverables section.
• 3:38
  SPEAKER_00: I'll have a link to that in the show notes, and you'll be able to see what happened.
• 3:43
  SPEAKER_00: They attached the schedule A that listed the errors one by one.
• 3:49
  SPEAKER_00: There were citations with wrong numbers, pin sites didn't match.
• 3:54
  SPEAKER_00: There were quotes that read like no judge ever wrote them.
• 3:57
  SPEAKER_00: There was awkward language, there were garbled words, duplicate words, there was mangled formatting.
• 4:05
  SPEAKER_00: This was not a single chatbot mistake.
• 4:08
  SPEAKER_00: This was a rushed AI-assisted drafting followed by rushed human review under deadline pressure.
• 4:17
  SPEAKER_00: And that's when you're going to see hallucinations, that's when you're not going to catch the hallucinations, and that's when they become dangerous.
• 4:24
  SPEAKER_00: So the real lesson isn't that one elite firm slipped, it's not that AI is evil, but that polished output can create false confidence precisely when time is shortest.
• 4:39
  SPEAKER_00: So I've been working on a different approach to legal research where you don't have to deal with the AI verifications.
• 4:48
  SPEAKER_00: And you know, my approach is gonna be contrary to the trend in the legal industry.
• 4:58
  SPEAKER_00: The trend in the legal industry is to add all of these dedicated legal databases like Westlaw and like SIS and like Bloomberg to AI tools.
• 5:12
  SPEAKER_00: So for$500 per seat per month,$1,000 per seat per month, lawyers will be able to do their legal research through an AI prompt window or some kind of mechanism like that.
• 5:29
  SPEAKER_00: And frankly, I think that's a mistake.
• 5:31
  SPEAKER_00: I think there are great uses for AI and legal research, but that's not it.
• 5:36
  SPEAKER_00: You know, I like there's a to me, there's an airport metaphor.
• 5:40
  SPEAKER_00: You have a you have an hour of flight.
• 5:42
  SPEAKER_00: Oh, it's only an hour-long flight to someplace that's 500 miles away or 350 miles away.
• 5:48
  SPEAKER_00: But by the time you drive to the airport, park the car, get the shuttle, or maybe you get an Uber to drive you, you go through ticketing, or you go through security, you wait at the gate for an hour, you take the flight, you taxi on the runway, you land, then you taxi some more, then maybe uh maybe you don't have to wait for the baggage claim.
• 6:15
  SPEAKER_00: But you got you got three hours in there, solid.
• 6:18
  SPEAKER_00: You know, it might have just been easier to drive.
• 6:21
  SPEAKER_00: That's where I'm coming from with AI legal research.
• 6:26
  SPEAKER_00: When you went into Westlaw or Lexis and they had natural language searches, which frankly I always thought were were subpar.
• 6:36
  SPEAKER_00: I I never thought they were that great.
• 6:38
  SPEAKER_00: But because what you're doing is you're gonna put a natural language search into the search box in your research tool, and then the research tool is gonna translate that into a Boolean search.
• 6:52
  SPEAKER_00: It's gonna try to suss out what you put in there as keywords and then do its own Boolean search and then produce results, which may or may not be great.
• 7:06
  SPEAKER_00: I never loved them.
• 7:07
  SPEAKER_00: And the reason is they're all they're gonna create a single Boolean search with keywords that I'm not controlling.
• 7:13
  SPEAKER_00: I think a much better approach is to go to your AI tool and ask, give me, give me five good Boolean searches for this natural language search.
• 7:27
  SPEAKER_00: And then instead of putting your natural language search into the research application box, put the natural language search into your prompt box and your AI tool and have your AI tool kick out four, five, six, however many good alternate keyword Boolean searches for you.
• 7:50
  SPEAKER_00: Then run those searches through your research tool and see what you get.
• 7:57
  SPEAKER_00: Then, you know, years ago I wanted to do a series on software tools for lawyers where we would talk about some of the features in Microsoft Word, Excel, Gmail, apps that lawyers are using all the time, but they're not fully educated on all the features.
• 8:18
  SPEAKER_00: And that goes multiple times for legal research tools.
• 8:24
  SPEAKER_00: Ask the AI that just gave you all these Boolean tools through screenshots or a description of what you're doing to help you dig deeper, make sure that the cases you found are good law, make sure there aren't duplicate cases and all the multiple searches you ran, and more or less educate you as to some of the deeper, less obvious features in that legal research tool.
• 8:52
  SPEAKER_00: So, what you've done now is you've used AI to come up with great, maybe creative Boolean searches you may not have discovered on your own, and you've used AI to dig deeper into the research tool that you're already paying for.
• 9:10
  SPEAKER_00: Then you've got what should be pretty effective, real case, real citation, legal research.
• 9:18
  SPEAKER_00: You want to take it a step further, upload those cases that you found into your AI tool, constrain a search for reasonable and usable blurbs that are within those cases, and double check those blurbs to make sure they're really there.
• 9:38
  SPEAKER_00: And then you're done with your research.
• 9:40
  SPEAKER_00: You've got cases that work, and then you could use those cases and that research output to have AI draft a great brief for you.
• 9:52
  SPEAKER_00: And you don't have to worry about going back and verifying those cases, although you should just to make sure at that point that the AI hasn't come up with new hallucinations in the research that you have created and discovered through this other method.
• 10:11
  SPEAKER_00: And you're not going to have to do verification protocols after the fact.
• 10:16
  SPEAKER_00: And I would really strongly urge lawyers who want to use AI to help with research, but don't want to find themselves in the same very embarrassing position that Sullivan and Cromwell found themselves in.
• 10:31
  SPEAKER_00: So even big law does get the AI hallucination blues.
• 10:36
  SPEAKER_00: Until next time, this is Ron Drescher for AI Tools for Practicing Lawyers.
• 10:42
  SPEAKER_01: That's it for today's episode of AI Tools for Practicing Lawyers.
• 10:46
  SPEAKER_01: Thanks for listening.
• 10:48
  SPEAKER_01: We'll see you next time.