ChatGPT for Lawyers 101
By Barbara Engstrom, Executive Director King County Law Library
I’ll be the first to admit that when I hear about the latest, greatest “advancement” in legal technology, I usually greet it with an eyeroll. Most of the time these legal tech wonders fizzle out and are forgotten in the time it takes to blow all that angel investor funding. ChatGPT however, shows no signs of abating, so I guess it’s time for all of us to get up to speed. I recently attended a fascinating CLE on ChatGPT presented by LeighAnne Thompson, Reference Librarian and Adjunct Professor at Seattle University School of Law and Jonathan Franklin, Digital Innovation Librarian at University of Washington School of Law. I asked LeighAnne and Jonathan to help us understand some of the basics of ChatGPT for law practice.
LeighAnne and Jonathan, you were both early adopters of ChatGPT. What sparked your interest in doing the deep dive into ChatGPT when it was still so new? Are you applying it to your work differently now than you initially did?
Jonathan: I was scheduled to teach a Legal Technology class in winter quarter, starting in early January 2023. ChatGPT arrived on November 30, 2022 and it was immediately obvious that the class needed to address this new tool, so I added it to the syllabus for the final two weeks. The quarter started with normal (pre-chatGPT) legal tech but by the time we got to mid-February, the whole second half of the class addressed generative AI. From there, I kept following it, with a special interest in using APIs[1] and generative AI to support public access to the law regardless of physical location.
LeighAnne: I teach a course called Law Practice Technology & Ethics where students explore both traditional law practice technology and emerging technologies, such as blockchain, predictive analytics, etc. When ChatGPT arrived, it was evident that generative AI might have a profound impact on the delivery of legal services and that we need to educate law students about its benefits and risks. Initially, I encouraged my students to experiment with it for low stakes tasks like generating lists or summarizing information, while supervising it carefully. As proprietary legal generative AI enters the market, I anticipate having students use the tools for more complex tasks, such as drafting legal documents.
LeighAnne, in the CLE you mention that lawyers are already using AI in their work without realizing it. What are some examples of AI that have been around for a while in the legal arena?
The legal profession has been using traditional AI tools for many years. For example, AI is used for legal research, analyzing contracts, analyzing and categorizing documents in litigation or due diligence, automating documents, and for e-discovery. Lawyers are also using AI for data analytics to predict things such as the likelihood of a particular judge to rule on a particular type of motion.
Jonathan, I tend to conflate the terms associated with AI like machine learning, natural language processing, and generative AI. What’s the difference between these concepts and where does ChatGPT fall versus something like Westlaw’s search algorithm?
The exciting aspect of ChatGPT is that it is generative. It creates new content. The term generative AI is used to distinguish it from extractive AI, where the AI is used to find preexisting content, such as searching for docket content or document automation. While extractive AI delivers the retrieved documents, generative AI is trained on those documents and then summarizes or synthesizes that content into something new. ChatGPT is one example of a group of Large Language Models (LLMs) that also includes Bard, Bing, Gemini, GPT4, and many others.
Natural language processing is a larger field that is used with either type of AI. For example, both Lexis and ChatGPT use natural language processing to understand the users’ queries. Machine learning is a method for improving insights from data, such as a movie recommendation system, but is not necessarily related to understanding a question that the user asks the system.
Jonathan, with generative AI in its infancy, most people see it as fun tool to knock out some of the more mundane tasks of life and work. You note that there are real costs – both environmental and economic – associated with programs like ChatGPT. Can you describe the costs and discuss any plans for mitigation by either tech companies or governments?
It is easy to look at generative AI as ‘free’ because you can go to a website and use it without paying. For now, the costs are being carried by the companies trying to build market share and figure out how to build subscription services that cover their costs. The costs of running the AI hardware are substantial, from the cost of the computer chips from NVIDIA, to the cost of water for cooling the servers, to the cost of the employees who are building the tools. If each legal memo generated costs $100 or more, users might not play around with it the way they are, so enjoy this introductory period while you can. At the same time, there are numerous competitors in the field, so there will be competition that will likely drive prices down for general tools. Specialized tool based on proprietary data, such as protected legal content could remain quite expensive if there is minimal competition because so few companies have the requisite data.
LeighAnne, you mentioned speaking with attorneys currently using generative AI in their work and teaching law students about generative AI. What are some real-world examples of attorneys using generative AI and what are the biggest takeaways for law students with their explorations of generative AI in the classroom?
I have talked to attorneys who are using generative AI to generate lists, such as deposition, interview, and voir dire questions. Other attorneys are using it to get a quick first draft of documents such as engagement letters or simple leases. Of course, we have all read the cautionary tale of the attorneys who used it for legal research and ended up being sanctioned by a judge for submitting a brief with fake case citations.[2] Some of the biggest takeaways for law students are learning how to interact with the tools to get the best results, and how to use the tools responsibly and ethically.
Jonathan, you mentioned the importance of prompts in getting effective responses from programs like ChatGPT. What are the dos and don’ts for creating effective prompts?
We hope that prompt writing is temporarily an important task and that future systems will take your question and optimize the prompt behind the scenes, sort of a pre-LLM prompt improvement process. For the sake of transparency, I expect you will be able to see the full prompt, but some companies might consider that final prompt a trade secret.
Use active verbs and be as precise as possible. It is better to say: “Draft a 100-word biography of Susan Smith” vs. “Tell me something about Susan Smith”.
Determine the audience for the content. ‘Draft a memo to a judge’ or ‘Explain a solar eclipse to a six-year-old in Spanish’
Finally, consider the format you want the result in. Do you want it to create a PowerPoint deck, a memo, or a formal motion? Be sure to say what type of output you would like. It is even better if you include a sample. The result will be closer to what you are expecting.
Until these systems are proven not to use your content for training purposes or otherwise retain it, do not include private information and ensure that you keep client confidentiality at the forefront. The ethics of using and not using AI are critical to integrating it into legal practice.
Jonathan, in the CLE you gave a URL to do a test run on ChatGPT. Could you give the URL and offer suggestions for specific prompts that might replicate legal work tasks so our readers can see ChatGPT in action for themselves?
I chose this site because it did not require registration, login, or other friction, so it is ideal for a hand’s-on session, https://chat.chatgptdemo.net/
I recommend splitting your tasks into two buckets, legal issues and administrative issues. Administrative issues are better for seeing what these systems can do because they are less specialized. For example, it is easier to write a letter to an airline complaining about the lack of ginger ale than it is to write a motion for summary judgment. Billing, marketing, website content, and other more general uses are ideally suited to this tool.
For legal tasks, start with those that are common enough that there are already samples out there, such as a non-disclosure agreement, a lease, or cease & desist letter. Again, don’t enter the client information or any details, but you can request it leave blanks for the party names and add specific clauses that you might want.
Regardless of what you use this tool for, please remember that you need to read the content and that ethically, you are responsible for it – so if there is a clause that is awful for your client and you don’t catch it, you cannot ‘blame’ the AI because it is your responsibility to evaluate everything you produce.
LeighAnne, a bad prompt could clearly implicate the duty of confidentiality of RPC 1.6. What client confidentiality issues need to be on our radar? What other ethics rules does an attorney’s use of ChatGPT impact?
It is important to read and understand the Terms of Use, Data Security, and Privacy Agreements for any legal technology, including ChatGPT. For example, a lawyer needs to understand who owns the prompts and outputs, how the prompts will be stored and reviewed, and whether the inputs and outputs will be used to further train the model. ChatGPT’s FAQ page states: “As part of our commitment to safe and responsible AI, we review conversations to improve our systems and to ensure the content complies with our policies and safety requirements.” If a lawyer includes confidential client information in a prompt (for example, copying and pasting part of a legal document without removing client information), not only has there been a violation of Rule 1.6, but this also raises questions about waiving the attorney-client privilege. Further, ChatGPT states that prompts will be used to further train the model, which is problematic if confidential information is revealed in the prompt.
The use of legal technology, including ChatGPT, raises many ethics rules. The competence rule requires lawyers to stay abreast of the benefits and risks of using technology. Some other significant rules include communication, supervision, candor, and reasonableness of fees.
This technology is developing at breathtaking speed. LeighAnne you mentioned some of the things that attorneys are currently using generative AI for. What’s down the road? What new legal products are in development?
Because the technology is moving at such a fast pace, it’s difficult to predict what is too far down the road. Currently, we are seeing legal generative AI products being rolled out that will use conversational prompts to perform research, drafting, summarizing, and analysis. Companies promoting these products promise features to reduce some of the problems we have seen with ChatGPT. Because the models will be based on underlying legal data sets, the generated content is more likely to be accurate and less likely to be invented content (“hallucinations”). Again, lawyers using these tools need to understand the benefits and risks, including the privacy and security risks. For example, Lexis+ AI states that it will encrypt prompt conversations, purge uploaded documents at the end of each session, and that users can delete their prompt history.
Jonathan and LeighAnne, what do you consider the top benefits and the top risks of attorneys using ChatGPT? What advice do you have for an attorney wanting to dip a toe in the generative AI waters?
Jonathan: The benefits for now are efficiency in completing rote non-mission-critical tasks. Spending 20 minutes drafting a letter vs. spending 3 minutes to get a draft and 5 minutes editing the draft. In addition, it is hard to start with a blank piece of paper, so these systems give you a starting point to react to, making them worthwhile even if you don’t end up using anything they generate. If it got you to complete tasks you dreaded doing by giving you terrible drafts, that is still a benefit.
In the near term, my concerns are related to humans thinking the AI is smarter than it is. Just as we have seen in self-driving cars, we might all want it to be a reality, but this technology is not sufficiently trustworthy when it is free. That is not to say some legal tools from reputable vendors won’t be reliable, but the casual use of a free site on the internet is truly ‘you get what you pay for’. Secondarily, I worry about deepfakes, or fabricated audio, images, and video that never happened, but looks realistic. It is far too easy to trust your eyes when humans using these tools can create false narratives.
LeighAnne: Like Jonathan, I think a top benefit is efficiency. Another benefit I see is in sparking creativity. For example, when I draft a list of deposition questions and then ask ChatGPT to draft a list, sometimes the ChatGPT list will inspire me to think about the issue differently.
In addition to the risks Jonathan identified, I think a real risk is that lawyers will use ChatGPT without understanding the flaws in the underlying model, including bias and misinformation, and without appreciating the risks of client harm if not used and supervised competently.
My advice to lawyers wanting to “dip a toe” in generative AI tools is to be aware of the benefits and risks of the tools, and to use them effectively and responsibly. As we see legal generative AI tools hit the market, I caution lawyers to carefully supervise the outputs.
What resources do you suggest for staying up to date with AI in the legal realm?
https://www.legaltechmonitor.com/
https://thebrainyacts.beehiiv.com/
Want to Learn More about ChatGPT?
Keep an eye on the KCLL events calendar and attend the ChatGPT for Lawyers 101 yourself. You’ll have the opportunity to ask questions and talk about how much ChatGPT has changed since this column came out. As always, for questions about any of your legal research issues, contact the law library at services@kcll.org or visit our website, www.kcll.org.
[1] “An application programming interface (API) is a way for two or more computer programs to communicate with each other. In contrast to a user interface, which connects a computer to a person, an application programming interface connects computers or pieces of software to each other. It is not intended to be used directly by a person (the end user) other than a computer programmer who is incorporating it into the software” See Wikipedia https://en.wikipedia.org/wiki/API
[2] See Benjamin Weiser and Nate Schweber, The ChatGPT Lawyer Explains Himself (New York Times June 8 2023) https://www.nytimes.com/2023/06/08/nyregion/lawyer-chatgpt-sanctions.html