ChatGPT: Benign Title, Revolutionary Development (State Bar of Texas Annual Meeting 2023)

Episode Notes

Transcript

[Music]

Intro: Welcome to the State Bar of Texas Podcast, your monthly source for conversations and curated content to improve your law practice, with your host Rocky Dhir.

Laurence Colletti: Hello. Welcome to another episode of State Bar of Texas Podcast. And today, today we’re recording live from the Bars Annual Meeting here in Austin, Texas. And we’re at the JW Marriott which is just off Congress Avenue. If you haven’t been there, Congress Avenue is really interesting street, I ran it early this morning, but incredibly fascinating, lots of great stories of vendors. So anyway, check it out, but that’s okay. We are going to get started. Now, I got a wonderful guest joining us, it’s Korin Munsterman, and she is fresh off her presentation, ChatGPT, Benign Title Revolutionary Development, welcome to show. Thank you for joining us.

Korin Munsterman: Thank you very much.

Laurence Colletti: So you’re also a law professor. So let’s look a little bit more about your background. So I think it’s going to tie nicely into our topic today. So where do you work? What do you do?

Korin Munsterman: I work at UNT Dallas, College of Law, I teach law practice technology, I teach social media and the law. I’m developing a course for AI on the law, which I mentioned in my talk that I’m kind of regretting I said I would do it in fall because too much is coming at me at one time and I run the legal education technology department.

Laurence Colletti: Excellent. So we have the right expert for our topic today.

Korin Munsterman: Well expert, yeah, sure.

Laurence Colletti: Fair enough. So ChatGPT, obviously it’s been on the news, people aware of it. There’s a lot of products and services being designed on its platform and so, just in broad strokes pertaining to the legal profession. What are some of the more revolutionary products you’ve seen developed on that platform?

Korin Munsterman: Well, so far I’ve seen to that are really, really solid products and I’ve watched demos and I recommend some people all the time, go get vendor demos, you can learn a lot because they’re going to anticipate questions that they got from other people who have gone through the demos. So watch the demos. One is Co-Counsel from Best Case. So what it does is leverage AI and ChatGPT or generative AI to do things to streamline the work of attorneys. So it will summarize a deposition. You can ask it to cite to particular incidents within the deposition. So let’s say it’s a car accident. You want to know the road conditions. It can give you the page in the line number every single place that that appears. It can then — if you’re going to depose somebody else, it can then generate deposition questions off the deposition that it just summarized or it can do a cold. So say, you’re going to depose an expert on metadata associated with images on an iPhone, right? And you’re not really sure where to start, you can enter that prompt and it will give you a bunch of questions that you should ask and if you have questions that occur to you because of the questions it gave you, you can tweak the original prompt and have a give you more questions. It can extract data from contracts, it can write motions or draft motions, rather, because it’s always the first draft, never just file that, always the first draft.

Laurence Colletti: Hopefully not.

Korin Munsterman: Right. So it can do a lot of things for you. The other product is spell book, which is works with contract management and contract review and it can summarize case law for you and it can summarize parts of the contract for you and help you extract data. There’s another product that I use in my class where my students can get a certification on contract review, it’s a company called LegalSifter and what it does is it combines artificial intelligence and human intelligence and will find all of the missing terms, find terms that have been found but may need to be tweaked, recommend language that you might change it to, you can enter your own playbook into the system so there are a lot of really interesting products out there now.

Laurence Colletti: Alright, next question is about ethics. So you know, I had a great — I wish I’ve been recording our pregame conversation but just talking about ethics and the use of artificial intelligence models for legal work. And so, recently I’ve been implementing ChatGPT into kind of drafting, at least pre-drafts, so a blog that I do for my own website, it just testing it. I’m just trying to learn, it’s a learning tool for me but doing a junction with one of our shows that we have California Innocence Project, just running over the different types of topics that we talk about and so, one of the things that I noticed is that, it’ll throw in correct citation, I don’t mean just a little off. I mean, way off like this never existed citation.

Korin Munsterman: Right.

Laurence Colletti: And so, I understand and I think we all read about the attorney that got trouble court for submitting wrongful or wrong information to the court and of course, got lit up in court for it. And so, let’s talk about the ethics because it’s complicated. I think garbage in garbage out and these artificial intelligence responses only as good as the information that goes in, that’s one. And then beyond that there’s processes and procedures that kick information that does its research and summarizes its information but this doesn’t always work and so, let’s begin there. So what did attorneys that may not be so familiar with these models need to be — what do you be aware of when they get into this world of artificial intelligence, helping them with that legal work?

(00:05:11)

Korin Munsterman: First of all, they need to be aware that it’s very versatile, but it only is predicting the next logical word, that is all it is doing, it is not thinking, it doesn’t know it’s speaking to a human. It’s an “it”, even though it uses the word “I” or “my” when it’s responding to you. Based on statistical models of language that are applied there’s a bunch of parameters that are applied to its corpus of data, which includes all of the public information that’s available online which includes things like fiction, right? It’s got 65 billion books in it, some of those fictions. Some of it is fake data, some of it is wrong data, some of these outdated data, it’s not analyzing for accuracy and it’s not lying. It’s hallucinating or I think there’s another word confabulation that’s brought some psychiatry that where people have a disorder where they come up with stories with no intent to deceit, right? It’s the same kind of thing. It’s just a predictive model of language, that is it, so be aware of that. Second, it’s a drafting tool only, right? That’s it. If you want to learn something about a new topic, it’s a great place to start but always verify. Verify, verify, verify the data. It is true that it comes up with fake citations, right? Because all it’s doing — it’s not checking to see whether there is a citation, it’s just doing the statistical analysis of language against words, right, and predicting the next word. So that case that you mentioned, not only were the citations wrong but when the court ordered them to provide the citations, it provided quotation from the case and had all of its internal citations were wrong too, where it did not exist, right? So they had other technological problems within the law firm itself that got them into this kind of trouble. So my best advice is to get some CLE training on legal tech. If this person understood the databases that he should have had access to and how to check the data, check the cases and citations and all of that, he wouldn’t have gotten to the trouble that he is. And by the way, he’s not the only one. There’s an attorney in Colorado who didn’t get nearly depress, but he’s also done the same thing. And as a result of these cases, federal judges mostly around the country, but there will be more have issued standing orders saying that if you have used any of these platforms to generate your filings, you must disclose it and you must verify and certify that you have double-checked it, a human being has double checked that data to verify its accuracy.

Laurence Colletti: Well let’s also talk about in terms of the ethics, client data. And so, you know, we’re talking about platforms that, you know, help with the legal work, maybe they’re helping with the research but that takes prompts and you have to communicate with it. You have to share certain information about your client with this platform and then it kicks out a result. And so, you were expressing some concerns about where does that information go, like how are these prompts stored, who has access to it, how do you protect your client, so let’s talk about that.

Korin Munsterman: Okay. So with any vendor, with any time that you’re using tech, if you’re storing data in the cloud, if you are using a case management system that has data on the cloud, if you’re using e-discovery platform, whatever you are using, you have to make sure that you understand whether the data is encrypted, where it is stored, who has access to it, what kind of security rights are on that, has the vendor ever had a breach, if you want to export all of your data, can you get a verification that all of your data has been deleted from their servers, what are they using the data for, so social media is replete with all these different cases where they use people’s data in order to “improve their services”, so AI is doing the same thing, generative AI is bringing in the data that you enter your prompts when you give it a thumbs-up or a thumbs-down, you’re training the software how to perform better when you regenerate a response indicating that you weren’t pleased with the first one, it’s also improving upon and storing all that data in order to get better. So, it’s not just learning from the corpus of data that it was fed with and trained on, you’re also adding data to it. And while GPT-4 says, it makes every effort to protect user privacy, that is true, it does, I would like to know more about that, really, dig more into that. But user privacy is different from data privacy, right? So what kind of the data that you enter into their, my best advice is do not enter anything confidential, I mentioned earlier that Bard, Google has — Bard is Google’s generative AI platform and Google prohibits its employees from entering code into its own system for fear of a leak of their IP. So if Google’s not going to trust their own system, lawyers shouldn’t trust them either.

(00:10:13)

Laurence Colletti: Wow, that’s a little scary. But we fought from there. So I think it was interesting too is that, you know, in terms of the information, you know, it’s not just about the personally identifiable like information, it’s not just about that. But it’s also the what you can draw from the information, you know, we’re talking about like, we’re talking about a case and like there’s a high-profile case, without naming any names, so nothing out there is a problem, high-profile case in this county again, not necessarily any problem there, high-profile case, this issue now you’re starting to be able to triangulate confidential information.

Korin Munsterman: Right.

Laurence Colletti: And so I think that’s one of the things people need to be aware of is that this tool is learning from you as much as you’re learning from it. And so, you had to be careful about where that information goes, it’s just something to be aware of like this information could you triangulate what that case was the matter about and could you draw any conclusions based on that which would hurt your client.

Korin Munsterman: Exactly. It’s just I teach my students this all the time. People will — and it’s very common for lawyers to raise their hand at a CLE session and ask for advice from other attorneys. Like, you know, I’m practicing in this family law case and it’s a new judge and does anybody have any information, that’s fine, right? But once you put your email domain, make that available so they know what firm it is and once you isolate it to a county particularly as a small area and maybe the nature of the case will help disclose whether like say you’re representing an airline headquartered in Texas, well that kind of narrowed the field quite a bit, right? So you just any kind of reporter who’s looking at things could discern who your client is, and discern some negative information about that client. So, you have to be very careful about how you phrase things, especially in any of these platforms, social media, ChatGPT, even at a CLE session.

Laurence Colletti: Alright, last question for you. So I feel like I’m behind already on these technologies and it just seems like every innovation cycle, just gets faster and faster, gets more and more difficult to keep up. So let’s say there’s a law firm up that was just working hard, they’re really good at what they do but they know they need to start paying attention to this and they haven’t had time and now they’re like, all right guys, seriously, we got to like pay attention, where should they go, just get those first initial steps?

Korin Munsterman: Well, they would have been fortunate had they attended my session on ChatGPT, they would have learned a lot.

Laurence Colletti: There you go.

Korin Munsterman: But if you’re really just starting from scratch and you need some background and you don’t have a colleague or a kid, you know, your children are usually pretty good teachers when it comes to technology or anybody to ask, you know, the best resources really is YouTube, it’s always YouTube.

Laurence Colletti: Alright, well thank you so much for that. And we’ve reached the end of our program but I want to say thank you Korin Munsterman for joining us and if our listeners have questions or you want to follow up or attend one of your sessions upcoming, where can they find you?

Korin Munsterman: They can find me on the UNT Dallas College of Law website. Feel free to email me. I will be happy to help.

Laurence Colletti: Anything on social media?

Korin Munsterman: I stay off social media. I teach social media and the law. Did I mention that? I stay off social media.

Laurence Colletti: Alright, and that’s where we’ll leave it. But first, I want to say thank you to our listeners for tuning in and listeners out there, if you like what you heard, do our show a solid here and please leave a positive review for us on Apple Podcast, Google Podcast, Spotify, or wherever you listen to your podcast. I’m Laurence Colletti until next time, thank you for listening.

Outro: If you would like more information about today’s show, please visit legaltalknetwork.com, go to texasbar.com/podcast, subscribe via Apple Podcast and RSS. Find both the State Bar of Texas and Legal Talk Network on Twitter, Facebook and Linkedin, or download the free app from Legal Talk Network in Google Play and iTunes. The views expressed by the participants of this program are their own and do not represent the views of nor are they endorsed by the State Bar of Texas, Legal Talk Network or their respective officers directors, employees, agents, representatives, shareholders, or subsidiaries. None of the content should be considered legal advice. As always consult a lawyer.