AI in Practice

So welcome everybody to today's learning objectives. And with us today is Tom Martin, who is the CEO of LawDroid and a legal AI advocate, a lawyer, speaker. I've known Tom for a really long time. I think I met you first at an ARG conference. And Tom's a graduate of Yale and UCLA Law School. He lives in Vancouver, so it's a bit early in the morning for him right now. So we appreciate you getting up and spending time with us this morning, but I'm going to let you talk about what you guys are doing with AI and what you've seen and the kind of questions you're getting asked and all that good stuff.

I'm happy to, and thank you so much, Katherine, for inviting me. Brilliant. I've been following your work and writing for so many years. Really excellent stuff and sources of information about new innovations. And actually, that's, you know, that's part of the lifeblood of LawDroid and what's interested me in doing this kind of stuff is I'm a 25-year lawyer, so I do have a lot of practical experience as a lawyer, but I've always had a love of technology. And so combining the two is something that I, I really enjoy doing day in, day out. And so today we're going to talk about AI for the practice of law, and we're going to go into some concepts that you may have heard explained before, some you might not have. So forgive me if we're going over some things that you may have learned in another webinar, but it's a good foundation. So first off, not to be melodramatic, but for me, November 30th, 2022 was the day that the world changed in terms of artificial intelligence becoming something that was for real. You know, that was when ChatGPT became available and the fluency and intelligence with which you could communicate with the machine was mind-blowing. And I don't want us to lose sight of that because I think there's been a, a lot of, you know, goal move, you know, goalpost moving in terms of artificial intelligence that now we expect so much more from it, but it's able to accomplish so much with what we have right now. It's pretty amazing. And so we're going to talk a bit a bit about that. This was a quote that I've used from time to time. And yeah, the repercussions of what, what we could do now with this technology is still becoming, it seems like on a daily basis, a new innovation, a new application of generative AI and AI generally. So what do we mean when we talk about artificial intelligence? I know that As lawyers, we're very precise with our language and what does exactly that mean? It's actually a pretty large group of concepts, large language models, of which generative pre-trained transformers is one type of large language model that's used to represent semantic meaning, words, and allows for machines to understand what we're saying. Conversational AI is a way of using artificial intelligence to have natural language conversations, as you see now in ChatGPT and other chatbots. I started LawDroid 7 years ago. 7 years ago, we had chatbots, we didn't have as much intelligence. And so now we're actually starting to reach the expectations that people had back then. Generative AI as I think we all know now, is where you can, you know, ask a question, request some output, and then it actually comes up with what you're asking for. We'll see in a second how that works actually behind the scenes. Fine-tuning is when you take a model and then you expose it to a certain set of data. And what you're trying to do by exposing it to that data is to get it more familiar with, let's say, workers' compensation, how those rules work. And so then in the future, when you ask questions related to that, it's more targeted in its responses. ChatGPT, I think we're all familiar with at this point, but it's the consumer front, consumer-facing version of OpenAI's technology. And a chatbot is how, is the user interface that it uses conversational AI to communicate with us. There's all kinds of applications that are using AI now. This slide is a little older because that over the past couple of months, we've seen an explosion of every single company across the world, including legal tech companies issuing their own AI bolt-on or There's been additional AI startups, but we have Synthesia, which allows you to create videos using artificial intelligence. Notion has built in an AI component to their software. You have Jasper, OpenAI's ChatGPT, of course, Midjourney for creating images, and all sorts of others that are available. So what is ChatGPT? It is not an oracle. I think the biggest misconception that we've all been learning together over time— oh, sorry. I was just checking a potential question there, but it's not an oracle. It's not an expert in the facts of the world that's current as of today's date. And you shouldn't be trying to use it that way. It is an easy way to think of it as a language calculator. And I think you've probably heard that from other people that you input some question or input some task that you want it to perform. And then the output is what it calculates based upon the language that you've entered, what approximates best what should come next. And so in that way, it's calculating language as opposed to a number calculator. It does that, but it does a lot more than that. The RLHF that I have down here is reinforced learning from human feedback. And that's actually what made ChatGPT just a quantum leap better than anything that came before it because Before we were exposed to ChatGPT, OpenAI had an army of people that were getting the output from the system and then giving their feedback as to whether or not that was what they're expecting. And they incorporated that feedback about if it was a good experience to make the model better. And so that's why the human element made it something that felt so fluent and intelligent to us. So you have the language calculator aspect, but you also have the human feedback and you also have it being trained on the kind of formats that we're used to. So we take this for granted, but like asking a question and expecting an answer that is relevant to that question, that's something that had to be trained on. And as well as being trained on writing stories or like creative output. So all of these things, all of these things came together, and that's why it's so good. Let's take a look at how it works. When we input a prompt into ChatGPT, like, what is the meaning of life, it gets broken down like this in a request that gets sent over to OpenAI. So the first is, you know, it sends over an API key to open the door to get access to the technology. Then we select what kind of model we want to use. It could be 3.5 Turbo, it could be GPT-4, it could even be an older model like DaVinci here. The prompt of course gets communicated because that's what we're asking for. A calculation is made as to how many tokens that amounts to. And you may know that when you send a request over to OpenAI, there's a token count that's kept and you get charged for that if you're using the API, for example. And there's also a temperature that you could set, and the temperature could be from 0, which is no creativity, no randomness in the response, versus let's say 0.5, which is a more creative response. And the creativity is because instead of giving you the best next calculated output, it's giving you something that's close to it. So it introduces a little bit of randomness or creativity in the kind of answer it gives to you. So all of that is happening when you're sending this order over to OpenAI to read through and then give you a response. It sends over the response. The meaning of life is a deeply philosophical and existential question, as we've seen. And This is how things go back and forth. And it's not magic. It is just statistical calculations, but ones that are getting extremely accurate. How do we get the information into ChatGPT in the first place? How does it know these things? Well, we work in language like a cow goes, for example. But to get that phrase, "A cow goes," which we want to complete that, "A cow goes," we have to translate it into numbers so that the machine can understand it. That process of translating it into numbers is called embedding, and it gets embedded as a vector. So if we remember any of our physics, a vector could be, you know, a point in 3-dimensional space. But I didn't get into law to be a physicist. But the, the point is that we convert letters into numbers, and that gets put into a vector database where OpenAI can understand this and, and draw out answers. So when it's in its number state, something like a cow goes would be in this 3D space very close to a related concept. So think of it like a 3-dimensional space where different concepts are in the same close proximity to each other, just like a constellation. So a cow goes would be closely associated with moo, so that if you asked OpenAI what does a cow go, it would say moo because it can find the answer in a very close space. So that data that informs ChatGPT or ChatGPT 3.5 about the world so that it's aware of these different facts are held in, in what's called a parameter. And a parameter is closely equivalent to a fact, right? So GPT-3 was aware of 175 billion parameters and GPT-4 is orders of magnitude larger, 100 trillion, right? So it has a vastly greater experience of what the world is like and what the facts are that make it up. Can't imagine how large GPT-5 will be, but we'll see that probably pretty soon. You might, might ask me, what about hallucinations? It's a big problem. The Avianca case, you know, made a mess of things. And so one thing I want to distinguish, actually, before we get there, one thing I want to distinguish for you is ChatGPT on the one hand, versus applications that are built on top of that that technology. Okay. So the Avianca and Schwartz case, that was them, from them working directly with ChatGPT and asking questions, trying to use it as an oracle, which remember I said not to. And then on the other hand, using an application that's built on top of that GPT technology, where they could add in additional safeguards. So what does that look like? Here at the base here, you see ChatGPT. So that is just straight exposure to that model without anything else. And that's why there's risks of hallucinations when you deal with straight ChatGPT. Prompts or prompt engineering, as you know, is now an occupation, is where you add an additional layer of sophistication to your interaction with ChatGPT. Now you could ask it to play the role of an expert in workers' compensation, for example, and to limit your responses to, let's say, a specific jurisdiction, and to make sure that your responses are in the form of an answer to my question. And you could give it direction about the kind of output you're looking for. So that prompt engineering is very important. Because now you can make it much more effective and specific. But where we can take it up to a whole other level of sophistication is what we call RAG. So RAG stands for Retrieval Augmented Generation. Retrieval Augmented Generation. And what RAG is, is where you connect a large language model like ChatGPT, you connect it to a reliable source of truth. What is a reliable source of truth? It could be something like, for example, the Harvard Case Law Project. The Harvard Case Law Project is a repository of US case law, federal, in all 50 states. And so when you connect this GPT technology to reliable sources of truth like that, you get reliable answers. So it's about that garbage in, garbage out principle. If you tie it to something that is a reliable source of truth, which could be your case file, it could be you upload all the, all the documents in your case, and then you use the technology to answer questions drawn from that information. So when it does that, it's extremely reliable, accurate, and informative. What about privacy? Isn't that a problem? Well, first off, if we're using the API for OpenAI, so we're building on top of OpenAI's technology, they specifically have a policy of not using that information for training purposes. So the fear that you input something that is client-related and that's going to get revealed to some third party somewhere else in the world. And oh my God, my client's going to sue me for that. It's not going to happen. If you're using an application by a vendor like LawDroid, or like Casetext, or any other, that is not going to be used for training purposes. So that fear is unfounded. When it comes to using ChatGPT itself, they now have this little switch where you could turn off allowing them to use the chat history for training purposes. Okay, so you could turn that off. And then, as you probably have heard, they now have a business version which allows you to do the same thing. So many of these privacy concerns, you don't have to worry about it if you take the steps to ensure that they're not going to use it for training purposes. The other issue is, of course, with any vendor as well as ChatGPT or OpenAI, you want to make sure that they're encrypting the data and also further that they allow you to delete data that you would want deleted. What about ethics? Isn't this a nightmare for ethical compliance? Well, first off, I think using ChatGPT and other vendors' technology that's based on ChatGPT, it actually helps you to comply in some ways with ethics requirements. Rule 1.1 requires us to be competent in using technology. And so by actually interacting with the technology, you become competent in the use of it because you've become more familiar with it. God knows that You know, Schwartz and Company are now much more familiar with how ChatGPT works and they're not going to get into the same problems that they had. And they've educated everyone else because of the popularity of what happened as to the limitations of ChatGPT technology. And like I said, again, that's ChatGPT versus applications that are built on top of that technology. That add additional safeguards. Rule 1.6, confidentiality. Well, we just talked about that. That data is not being used for training purposes. It is being encrypted if you're picking the right vendor. And also, your data can be deleted from their system upon request. 2.1, it is, of course, a bedrock principle of the practice of law that you review your work product before you file it with the court, give it to your client, have them sign it. And so that remains a principle that stays in place regardless of what technology you use. So yes, even when using ChatGPT, Copilot, Co-Counsel, any of these products, you still want to review it. Does that mean that you're spending twice the amount of time? It's not. It's gonna save you a lot of time to have the work product initially created by the, by the AI, and then reviewing it would take just as much time as possibly reviewing it from a paralegal or an associate. So what can it do? So this technology helps with a, a number of different things. It is really good at brainstorming ideas. Blog article writing, for example, is a great way to use it. Or I think fundamentally we all are stymied when we are faced with a blank page and we don't know where to start. And it's really good for getting us started on ideas and even a first draft of an article that we could write. And I think with our, you know, legal brains, when we're faced with something that doesn't really fit what we want, we're gonna break it down and then rewrite it. So it helps to actually get us started and even to put it into our own voice. Beyond creation, it's really good at extraction. So summarization, it'll take an article and summarize it by extracting information from it. Or you could even upload documents like let's say, let's say a transcript of a deposition, and to summarize, to summarize it, but with an eye towards maybe specific facts that you're trying to establish. So summarization is another application. Excuse me. Grammar check, you can go through and check that, help you with writing, and also translate into many different languages, which is quite good at. Equivalent to Google Translate, if not better in some instances. Here's an example prompt. You might have seen some examples like this, but if you put it in ChatGPT, you know, write me this demand letter, and you give it a few different issues like damages and then breaking down the damages into medical bills, car repairs, and chiropractor fees, it does a fairly good job of writing at least the beginning of a demand letter. Of course, this begs the question for many practitioners, like, this is a great start, but what if, you know, I could connect it to the case file where I have, I actually have the medical bills in there, or I have the list of car repairs and chiropractic fees, as well as the doctor notes? And the good news is that actually with some plugins you could use for ChatGPT, or with third-party vendors that build on top of the technology, you can upload those documents and you can make them more useful to writing a draft like this that could be a much more comprehensive draft. Of course, keeping in mind that you need to review it to make sure it covers everything you want. Here, this is an example of getting you started with creative writing or somewhat creative. Because you want to write, let's take this example, an ebook about probate administration, California. It gives you an outline. I know that for myself, once I have an outline, I could start really jumping in into it and taking it step by step. It's, it's that, you know, overwhelming sense of I don't know what I'm going to write or where I'm going to start that paralyzes me. But if I have an outline, it makes it much easier to get started. And this does a really good job of that. And a marketing application is writing follow-up email sequences. I don't like doing that. You probably don't like doing that. We're not marketers. But this thing does a pretty good job of drafting up a sequence. And this is about data security compliance. And then I put the objective here from introduction, from introducing yourself to offering a free legal consultation. So it understands that the objective is to lead them towards getting a, a consult with you about these, these issues that you're bringing to their attention. And this language here in particular, which ChatGPT just came up with, is, you know, with the increasing threats to data security, it is important for businesses to take necessary steps. So it's kind of like, you know, it's, it's kind of pushing that button. Hey, there might be some risk here. You want to be careful. And is using that emotional, you know, using that psychology of, of risk to encourage people to, to get a consult with you. So it's quite good. It's probably been trained on millions of examples of marketing. This one is, I really liked is it gives you a way to respond to something negative, like a negative Google review. If any of you work for a period of time, like there's something that comes up, right? In the moment, you want to defend yourself, you want to jump on it and just be really adamant. And this, this in a way gives that nice professional distance. By allowing it to respond for you. And it hits all the right points that you should in answering a negative review, which is, you know, thanking them and then, you know, we take all feedback seriously and reach out to contact us. You know, does a textbook good example. But I thought this was interesting in that it helps to moderate, you know, the emotional response that we would have naturally. There's a lot of great prompts in Awesome ChatGPT Prompts, which is available on GitHub. If you go in there, there's like hundreds and hundreds of prompts that might be useful that you could use in your daily practice. Mitch Jackson has some that can help with writing styles, and Dazza Greenwood at MIT has his own. I think he even had a webinar, a course on legal prompt engineering, and there's a repository that he has as well. So these are some really great resources for playing around with prompts and seeing what you can get out of ChatGPT. This is a webinar that I gave with some really excellent people, including Daza, Jackie Schaefer, Natalie Ann Knowlton. And actually, ChatGPT itself was a panelist at the beginning of the year. So this is on YouTube. Check it out if you want to. ChatGPT and its implications for the legal industry. This is an article that I wrote about how you can use ChatGPT as a creative partner called Harness the Power of Generative AI with Creative Collaboration. That's on Artificial Lawyer. I would definitely recommend you check it out because what I point out there is this dichotomy between sometimes lawyers expecting generative AI to perform a push-button perfect output versus it working with you as a creative collaborator. So that distinction, like, it's the successful strategy is to work with it as a creative collaborator where you, where you ultimately are responsible for the output, not expecting it to be push-button perfect. And when I say push-button perfect, I mean, You give it a few directions, like let's say to write a successful Supreme Court brief on whatever the issue is, you push the button, then you expect to be able to file that directly with the Supreme Court and win the day. Like, that's not how you should use it. That's how Schwartz used it. Don't make that mistake. Think of it more as a creative collaborator. And so, These are the benefits. It could help to eliminate procrastination, definitely less effort in getting started, give you more insights. You could ask it to give it feedback, to give you feedback from different points of view. Like if you're trying to argue against me, what are some of those arguments? You could get more work done and hopefully have a better work-life balance. So the applications with which you could hopefully accomplish that include these. So my company's LawDroid. Casetext, as we know, its co-counsel is amazing. It gets really great results and has many different applications. Clearbrief is really great for motion analysis and writing. Harvey, although a little mysterious, has got some really large clients. And is doing some great things. Jurisage, which just recently merged with CiteWrite, I believe, is a great source of Canadian case law summaries and analysis. Spellbook, it's an add-on widget in Microsoft Word that helps with drafting contracts and analyzing them. And they're also from Canada. So some really great applications here. I would encourage you to check them all out and play around with them. I'd like to show you a little bit about LawDroid Copilot, which is our AI legal assistant, and what it could do. I'm going to show you a few demos, okay? And those videos are really fantastic in showcasing how you can use generative AI in a practical, in a practical way in a legal context. This is what it looks like. And I, so I wanted to give you an idea of what it looks like so you could see how it's different. So one of the things with dealing with ChatGPT is that you have to formulate a prompt. In fact, that has led to that whole occupation of being a prompt engineer. But most of us as lawyers, we don't wanna become or learn how to be a prompt engineer. We just want it to work, right? We just want the assistance. And so the bet that I placed was simplicity would be extremely useful. So we boil down prompts into buttons. So you just need to press a button and that will take you to the place you wanna go. And behind each, each of our buttons is a well-formulated prompt that can help you with what you're looking for. So if you want to summarize something, just click on summarize, correct grammar, it's all self-explanatory. Draft email or letter, for example, instead of having to feed in everything that you want to ChatGPT, you click draft email or letter, it asks you a series of follow-up questions to clarify, like who are you writing this letter to, What's the purpose of this letter? Is the tone that you want formal or informal? And you just answer a few of those questions and it drafts a letter for you that's perfectly appropriate. So in that way, it's just really straightforward and it's like having a conversation with a colleague or an assistant who can help you out. We even have an option for emotional support, which is you're working after hours, it's late at night and you're frustrated, you got it. Motion due tomorrow. It's actually pretty good at empathizing with people and walking them through, like, practical steps of how to break down the work, but also, you know, maybe even telling you a story of what it would be like to take a vacation in Tahiti. So this is what it looks like. I wanna show you an example of what it can do. First is lead generation. Here, creating a chatbot in Lodroid is as easy as creating a flowchart as you see here. And when you create that flowchart, it appears to a user like this where you can customize the welcome message, the questions that are being asked, and make it fit with what you're looking for. It doesn't have to be this. It could be customized to your own use. It captures contact information, That data we can share with integration partners such as Clio and other CRMs. And it's saved within our system. So the first name, last name, email address, all of that is saved within our system. And we also even enable answering questions on the fly based on data that you've uploaded. So here the user has a question about a Texas legal issue, like how do I get divorced? And we could instantly use ChatGPT technology to draw the answer from reliable sources that you've uploaded into our system. So it's not hallucinating. It's actually getting the answers from the information you've provided. And how do we get that into our system? You could create a knowledge base like this. You could paste the information into our system to create the knowledge base. Or if you have many different sources, you can upload those into a knowledge group. Which is an umbrella for holding all kinds of information. And then you can integrate it, with applications like Clio Grow and then port each of those data points, the name, email address, and all that so that it goes into the CRM. So we make it as easy as possible to get this information into your system and get new leads like this, so that you can grow your practice. Sorry, one second. Okay, so the next demo I'm going to show you has to do with doing basic legal research. It's an extremely popular application. One thing I want to point out is that what we can't do right now is to write a legal memo based off of a question you have. But what we, what we can do is we can make it lightning fast to get cases that are relevant to a search that you're performing and then analyze that case, summarize it, and you'll see here how that works. So you could, again, press the button for case law research. Run a search against U.S. case law. And here, let me pause it for a second just to show you that when you do that search, you can now do it just like you're speaking to a colleague. Hey, where are the cases involving child custody where parents stop paying child support in California? So you do have to give it a certain number of keywords, and also like the jurisdiction of where you're looking for it. And our system, like others, provides a ranked list of the cases that are relevant. You could display previews of that case just to get the gist of why that case has come up and why it might be relevant. You know, that kind of functionality is not setting the world on fire, but it does help, you know, to also, like, read the case if you want to. Now, that is not involving any generative AI. But again, if you want to review the case manually, you can do that. But we've— what we've built into it to take it up a notch is we've used generative AI to, let's say, brief a case. So I'm pausing it for a second because I want to tell you what's happening in the background. There's a lot going on. The first thing that we're doing is when you selected that case, let's say McCann versus McCann, that case is getting imported into our vector database. So remember the process I showed you from changing words into numbers? We're extracting all the text from that case and ingesting it into our vector database, turning it into numbers so that ChatGPT can understand it. Then we're sending all of that information over to ChatGPT to quickly read through and analyze the case. And once it's read the case, then I've fed into it a series of questions that it needs to answer. And that's when we get the output. So it analyzes the case, and then we get the factual summary. These are the facts of the case. We get the procedural history. And let me stop it there for a second. Again, you don't ask ChatGPT, what is the procedural history of a case? If you Just think of, think of a person off the street. If you asked a person off the street, what is the procedural history of McCann versus McCann? They would look at you like you're crazy, right? Because normal people don't talk about procedural history. They may talk about what did the judges in the, in the courts that decided this case before, before this, what did they decide? Right? That's how they would talk about it. And surprisingly, that's how you need to communicate with ChatGPT when you ask it to answer these questions. So when I asked it to provide the procedural history, I didn't ask it that. I asked it, you know, what the courts before this case, before this appeal had decided. And you could see here, The case has been heard by the Court of Appeal and they have decided this child support order does not automatically terminate. And it goes on. My point being is that it does a really good job once you ask it the right way. And then it goes on to, um, summarize the issues involved in the case, the rule of law that was applied, and, uh, it even goes into, you know, like the case holding, what did the judge ultimately decide? Not what is the case holding, what did the judge ultimately decide in this case? And then the reasoning about how the judge arrived at that decision. Um, and just to stop it there for a second, let's, let's take this in. Like, this is an entire case briefing that, remember back in law school, you're given all these cases and you have to run a case briefing. It takes time. And you have this in a minute or two. So I could hear you all thinking like, well, does this, is this reliable, Tom? What I could tell you is anecdotally. So when I used to practice in California, I did a lot of toxic tort personal injury cases, which involve issues of causation, which are complicated in toxic tort cases because there's usually long periods of exposure to chemicals.. And then how does that cause the injury, which could be multiple myeloma or leukemia or whatever it is, some kind of a lung injury. Proving causation is difficult. Point being, I did searches like this, found all of the top cases that I am familiar with, uh, that deal with causation in toxic tort cases. And I briefed them all and I read the briefs. And they were on point. It summarized the relevant facts of the case, the issues, and all of that. So from my experience, it has been highly reliable in creating these briefs, and it's an excellent way to become familiar with a case. And here you can ask follow-up questions. So like, once it briefs the case, you know, what's the name of the judge who wrote this opinion? You can ask follow-ups. Kind of like having a conversation with a colleague, you know, gives you the name of the, the judge. But you could ask it other things like, can you please provide me just a short summary of this case? So not a full-on brief, but just a summary. Uh, you, you might even ask it to apply the rule of law that it just defined up there to some other situation. Uh, so it's a great way of interacting with, with cases and becoming familiar with them. Now this one, this is the most recent thing. It's actually one I, I had telegraphed to Katherine earlier that I had hoped we would have ready at this point, and we do, and it's really amazing. Okay, so what it starts with is the ability for the users of our platform to upload their own documents. So it could be a motion, it could be deposition, deposition transcript, it could be whatever documents you have. You can now upload it to our system. And just like you were able— you, well, you could see that we could ask follow-up questions about cases. Now you can do that with your own documents and you could extract really useful, practical, and valuable information from them. But in this case, this is a workflow for analyzing emotion. So the frame for this is you're a litigator. You day in, day out are reading motions from opposing counsel and you have to write an opposing motion. I've been there myself. I don't like it. And, you know, if you're a junior, you might be handed the motion and then you have to, you have to figure out what the arguments are, you know, look for what are the citations this opposing counsel referenced and what are they claiming that those citations stand for? And then even further, can you develop some counterarguments? And beyond developing the counterarguments, do the cases that they cite really stand for the proposition that they're claiming they do? So let's take a look at how we can help with that. So first off, here we have these two new buttons at the bottom. Use a document is for a document that's already been uploaded into our system. Upload document clearly is to upload new documents. You could upload PDFs or Word documents. Just be sure that if you upload a PDF that it's been OCRed so that there's actually text that the AI can work with. So when you click Use a Document, there's a whole laundry list of documents I have in my system, but this one here, Motion for a Judgment Notwithstanding the Verdict, this actually is one that I found online. It's not one that I wrote. Don't have any ties to it or really know the people. I don't know the people who wrote it, but I thought it'd be a good example. So you select that motion, and as soon as you do that, it makes it available to our system. Like, it's ingested that information, and now you can perform a number of different tasks on it. You could summarize it. You can extract legal citations, extract the arguments. Let's try extracting the arguments. And, one second. So just as with the cases, when you click extract arguments, it is reading through the whole thing very quickly, finding the arguments, which from its, its training, its generalized training, it's understood it's, it's developed an understanding of what arguments are and how to get them out of a document. So, all right, so the first argument made in this motion for judgment notwithstanding the verdict is that there's been juror misconduct. So what is that about? They're saying that, uh, juror number 2 failed to disclose her employment history as an office manager at a law firm during the VAWD year. So that's a reason that you should throw out the case. Juror misconduct. Okay. And it even goes on here to say the defendant argues that this misconduct prejudiced the case and the case they cited for that proposition. Right. But let's go on. We have a second argument here is that the plaintiff failed to meet their burden of proof. That's why the case should be thrown out. Uh, defendant— or the judgment should be thrown out. The defendant contends that the plaintiff presented conflicting evidence regarding the accident circumstances. Okay. And then the third argument is just that the jury verdict was clearly erroneous, that it didn't follow the law. Um, Defendant cites the testimony of Cynthia, Cynthia Herrick, a witness, and physical evidence. So like, it's not just coming to the conclusion, it's tying it to specific facts that are presented in the motion and providing you with a summary of that argument. So this, again, you're not having to read through the motion, although you probably should, and to confirm everything and make sure that, you know, it is tracking. But very quickly it gives you like a lay of the land of what the arguments are. Oh, and see at the bottom there, would you like to create counterarguments? Uh, we'll see that in a second, but I wanna show you that this is coming from an actual document. This is the motion for judgment notwithstanding the verdict that I uploaded. The first argument is juror misconduct. It's not making it up. It's right there. All the facts that it extracted, it's right there. Second argument, failure to meet the burden of proof. And then all the facts that were the basis of extracting that argument right there. Third one, jury verdict clearly erroneous. So this thing is highly reliable when you give it a reliable source, source of information. And what better source of information is your own? And so when you upload a motion like this, it's like giving it to a paralegal or an associate to read and give you summaries. It does a really great job. But let's go even further. We want to create counterarguments. We want to get started on our own motion, and this thing can do it. It can get you started. So the first counterargument, juror misconduct. Our counterargument to that is that The failure of juror number 2 to disclose her employment history during voir dire, uh, is— that argument is flawed. And the reason it's flawed is because the claim that it would have made her biased in some way is speculative. They haven't shown or proven how that, that was actually the case, um, and they, they haven't provided any concrete evidence. The assertion that she used her experience as an office manager at a law firm is speculative and they haven't shown that it's influenced the jury's decision in any way. So it gives you a good start about how you would argue against the juror misconduct claim. Same is true for plaintiff's failure to meet the burden. And the same with, uh, clearly erroneous. The defendant's claim that the damage to the plaintiff's vehicle is consistent with the plaintiff entering the intersection is speculative, has not provided concrete evidence. Um, so look, one thing it's not doing, I'll grant you, is it's not giving you like the legal basis of those counterarguments. It's not saying, here are all the citations to support the counterargument. Um, we're not there yet with that, but what we are able to do is extract the legal citations from the motion that we were reading. And so here we get a summary of what those cases stand for as represented by opposing counsel. So we could see here that, uh, Bernand is cited for the general rule that, you know, the decision to grant or deny a motion for new trial is within the sound discretion of court. Walker versus Hall, that the trial court's discretion extends to matters concerning juror misconduct or irregularity. But let's say we want to check that out, so we enter that into our case law research that I showed you before. We find that case, Walker versus Hall, and in this situation we choose to summarize it. Okay, so keep in mind that when it's summarizing, it's drawing it from the actual case, so it's going to be reliable. We actually do have to read through this summary, so it doesn't do everything for us, but we read the summary and then we can put two and two together. Is it actually— does this case actually stand for the proposition that opposing counsel is claiming it does, or is it slightly not all four corners, is off point a bit, and how so? And then we could point it— point out how it is. And the summary really helps us to accelerate our analysis of the motion. So to be clear, this is not going to replace us. It's not going to replace your job or what you do. But as you can see here, it's going to highly accelerate and amplify your ability to do more with less. So it's a lot, excuse me, a lot we've gone over, but we do have a special offer right now, LogRocket Copilot, because it's in beta. As a lot of things are right now, it's only $5 per month to try it out. So if you just go to lodro.com, you can get early access to our offering and try it out. I'm very open to answering any questions anyone has about what I've shown you. But suffice it to say that we're at the beginning of this. There are so many different offerings that are becoming available from different vendors. That can really help you to do more with less resources and may even, I mean, is a boon in this economy that you can have additional help without having to hire more and throw bodies, you know, more bodies at the problem because that really is not, not a viable solution, I think, in this economy with inflation and all of that. So if you want to read more about what we're working on, and I do share some of my personal story in it too, my Substack is lawdroidmanifesto.com, and talk about different innovations in artificial intelligence and how it works in the legal industry. You know, just lots of food for thought. And, you know, I want to thank you very much for having me on to talk about AI for the practice of law. Please hit me with any questions you have, but you could follow me at LawDroid on Twitter or X, as we're calling it now. And you could check out more about LawDroid at lawdroid.com. And I'm very thankful for this opportunity to share these thoughts with you and hope you share yours with me.

Thanks so much, Tom. I really appreciate the explanation of the difference between using a tool like ChatGPT versus using the technology, applying it to your own data, because I think that's a huge differential. That is where, to me, all of the exciting stuff is going to be happening in terms of knowledge management for law firms and the ability to leverage. For instance, somebody asked me this morning, you know, how do you, how do you see this working in addition to your own gold standard documents, especially if you're a bigger firm and you've got a lot of documents? Thinking about things like if everybody had— if you could point it to everybody's LinkedIn, or you created a profile of everybody. If you use Microsoft, um, Directory appropriately in your 365 tenant, you could have people add things like what languages they speak, what cases they worked on before, what practice areas they had before. They— and so all of a sudden you're leveraging the people, which is something we talk about in legal ops leveraging people, processes, and technology. And so, you know, being able to find out who knows what around here. As a former librarian, I will tell you that I did as much searching as I did picking up the phone and contacting somebody who could answer a question for me just out of what they knew off the top of their head, you know. And so that's the kind of thing where I feel like We have a lot of opportunity, and that's what I'm enjoying showing people. So I'm very excited about this. And also, if you are using Microsoft Edge and you're a 365 subscriber, or even if you're not, you might want to try it out now because Bing now has, as of this week, has Bing Chat Enterprise built in to Microsoft Edge with your 365 subscription. So if you've got Standard, Business, E3, E5, whatever, the Bing Chat Enterprise will allow you to— you can't upload documents yet or anything, but you can actually have it summarize a Word document that you have pulled up from your 365 account. But it's also added a brand new terms of service. It says right at the top of it, protected, and and it is automatically defaulting not to sharing anything with Microsoft. Your searches are not cached. The communication is encrypted in transit and storage, but they're not actually storing anything. Once you turn off that chat, it's gone. They're not saving anything. So I think the market is meeting the needs of not just lawyers, but corporations and other people who are concerned about, I don't want to be training this stuff. I don't want bits and pieces of my, my information stuck in this system. I don't want to be— I, my boss came by the other day, goes, oh, are you training the AI? You know, because I was using it. And so anyway, I appreciate all the work that you've done. And for those on the call, just to know that Tom also has the LawDroid as a product, was a really cool front-facing, um, little bot that you could put on your website and basically have a chat interface that, that was, um, trained to answer questions. And it was the kind of thing where I, I guess if you've ever updated your, um, your tags and your, your car information at the North Carolina, um, Department of, uh, Transportation. You can do that with your clients using a bot on your website. But now we're talking about things like— there's a product I've been playing around with, Kaya, where I can upload my blog post and let people ask it questions. And so the search in WordPress for my blog is just straight up words. You put in a word and it gives you all the blog posts. It doesn't answer the question. And so this is where we're going. You know, we're, we're going from I have to read 16 articles to find the answer to my question to it's going to look through the articles and find it for me. Do we trust it? Probably not yet.

But, well, just to jump in there on, on, on that for a second is that, you know, we, we've been working with legal aid organizations over the past 7 years because it's a perfect marriage. You know, they have so many people that they want to help and they don't have enough people or a budget to help everyone, the thousands and thousands, millions of people that have unmet legal needs. But we've been working with, especially since we rolled out this new knowledge-based technology, we've been working with legal aid organizations across the country. Right now there's 5 that we're under contract with where what we're doing is they have years and years of this amazing, you know, brochures of legal information. Websites, PowerPoints, like all of the stuff that their volunteers have put together over the years. And we're unlocking that information for them. So just like you were talking about with your blog posts, one that comes to mind is the Texas Legal Center. They have hundreds and hundreds of Q&As and we've put all of that and digested it into our system.. And you could simply as a user, like as a citizen that has a legal issue, just ask like, how do I get divorced in Texas? How long is it gonna take? Right? And it calls through all of that information and immediately answers that question where before that, the burden would be on their shoulders to try to like go through the circuitous path of it's buried somewhere in a PDF, like 3 or 4 levels down. They immediately get answers to the question. And of course they're told, and if you have any further questions, we could, you know, possibly help you if you qualify. So it is just, you know, a game changer in helping people. But that's also what lawyers, all of you as practicing lawyers, can unlock for your own potential new clients. And clients, allowing you to be there even when you're busy.

Absolutely. I'm going to stop recording, so if anybody has questions they don't want to capture for posterity, I'll do that. But Tom, thank you so much for your time and this clear explanation of how all this works, and I will get this recorded and sent out to everybody.