ChatGPT on retAIner: goodcounsel packs AI in the legal briefcase
A year or so ago, amidst the initial frenzy of excitement about artificial intelligence (or more specifically, about the “large language models” or “LLMs,” which enabled non-programmers to interact with AI using natural language), I mused on our blog about whether our clients should fire us and hire ChatGPT instead. (Short answer – not yet.)
As I wrote at that time, “I expect that AI tools will become increasingly helpful as assistants,” and over the past year, this has certainly come to pass here at goodcounsel. We have been finding ways to incorporate AI into our practice every day, both the general-purpose technologies used by the public, such as ChatGPT, and specialized legal tools built on top of these AI platforms, such as those offered by our partner ClauseBase.
Below, I’d like to describe a few of the ways that goodcounsel uses AI to enhance its law practice.
Research
LLMs like ChatGPT4, which I often use, excel at assembling information about specific legal topics, which gives us a running start. However, the LLMs still offer some straight-up wrong answers to questions. For example, we have been fielding a lot of questions about the Corporate Transparency Act, which recently went into effect. Large operating companies that have an “operating presence at a physical office within the United States” are exempt from CTA compliance. I asked ChatGPT how the statute defines this concept. ChatGPT replied:
The CTA doesn’t explicitly define the term “operating presence at a physical office within the United States,” leading to questions about how this requirement applies, especially in the era of remote work.
As has often been observed, LLMs answer confidently even when they are wrong, and that was the case here. The Code of Federal Regulations defines the phrase “operating presence at a physical office within the United States” at 31 CFR 1010.380(f)(6). Fortunately, we did the extra research necessary to discover this.
I think of ChatGPT as a highly knowledgeable but slightly careless research assistant. (I struck “slightly” in light of this recent study [full text] from Stanford’s RegLab, finding that “hallucination rates range from 69% to 88% in response to specific legal queries for state-of-the-art language models.”) The AI’s processing power and vast store of knowledge means that it will have information to offer on just about any legal topic. However, the information it offers will not always be complete – or correct.
Drafting
Lawyers are supposed to be skilled writers and a great many are, but the profession suffers (in my opinion) from an excess of long-winded, repetitive, and ponderous drafting. One of goodcounsel’s values is to draft in concise, plain English. ChatGPT provides able assistance in this effort. It will take a verbose provision and quickly transform it into something much simpler. (More on this in an earlier post.) It may sometimes go too far, eliminating useful details or rendering language that is a bit too casual (even for us), but it definitely provides a helpful first draft.
Legal-specific tools
Our partner, ClauseBase, is typical of legal technology vendors who are developing AI tools for legal applications. Though its core mission remains document assembly (i.e., helping lawyers create template clauses to reuse across documents), ClauseBase now also offers reviewing tools that integrate directly into drafting applications such as Microsoft Word.
For instance, ClauseBase can swiftly summarize a 50-page agreement or generate a comprehensive “cheat sheet” of defined terms with corresponding cross-references. (Should lawyers be drafting 50-page documents in the first place? No, but since they do, it’s nice to have AI tools to help us digest them.) ClauseBase also improves proofreading, identifying inconsistencies and omissions with greater accuracy than humans – and relieving us of one of the more boring aspects of our job.
Lawyers who are not finding ways to use AI tools are doing their clients a disservice. They spend more time getting work done, which, in the stubbornly persistent hourly billing model, may be financially rewarding for the lawyer, but does not help the client. goodcounsel invests in AI tools and capabilities, most of all because as professionals, we believe we should always be practicing with the best tools for the job. And, because we frequently set up client engagements on a fixed-fee rather than hourly basis, our incentives are aligned with completing work efficiently.
P.S. I asked ChatGPT if it was “offended” by my characterization of it as “a highly knowledgeable but slightly careless research assistant.” It replied: “Not at all! It’s a fair characterization in the context of explaining the capabilities and current limitations of AI like me. It’s important for users to understand that while AI can be a powerful tool, it’s not infallible and should be used with a critical eye. Your description helps set realistic expectations.”
To give credit where credit is due, ChatGPT offered suggestions for the above text, a few of which I accepted.
Categorised as: Artificial intelligence and the practice of law, Law Practice Innovation, Lawyering