Shortly after OpenAI released its first large language model, GPT, to the public in late 2022, I began working on ways to use LLMs to make labor law more accessible. At that point, the only way to get information about the National Labor Relations Act or the National Labor Relations Board was either from a legal database service like Westlaw, which costs thousands of dollars per year, or from the NLRB website, which is basically unusable.

Across the next couple of years, I used LLMs to help create NLRBResearch.com, which hosts two databases, one that contains all NLRB decisions and guidance as well as relevant Circuit Court and Supreme Court decisions and another that contains every document posted on every NLRB docket from 2010 to present. The databases are self-updating, full-text searchable, and contain LLM summaries of many of the documents.

During that same time, I created the NLRB Edge newsletter, which is essentially a legal tracking publication for the NLRB. Whenever new documents get added to the NLRB Research database, I use LLMs to produce summaries of those documents and then post them to the newsletter. There are other legal tracking publications that cover the NLRB, including Law360 and Bloomberg Law, but, like Westlaw, these subscriptions cost thousands of dollars per year.

This week, I have taken another step down this path by releasing The National Labor Relations Book. The book provides a novel introduction to labor law that should be useful to individuals who are relatively new to the topic, including law students, union staff, and rank-and-file workers, while also being useful to experienced lawyers who would benefit from having a handy outline of key areas of NLRA law that cites to recent decisions.

As with my other projects, I relied heavily on LLMs to help create the The National Labor Relations Book. In order to generate the text of the book, I programmatically identified the 100 most-cited cases involving the NLRA, found the 100 most recent cases citing to each of those cases, and then used LLMs to generate legal summaries using those 10,000 cases. I think this method is uniquely suited to the purpose of introducing an area of law because it focuses on the points of law that come up the most often. However, prior to the invention of LLMs, this approach to legal summary would have been too laborious to be feasible. 

Of course, using LLMs to reliably create a book like this is not as easy as opening up an LLM chatbot in the browser and telling it to write a book. A lot of human labor went into the construction of the book as well. I wrote the first couple of chapters, created the system of programs that facilitated the 10,100 API calls to Anthropic and Google, and then reviewed the result to identify and wring out errors.

Regardless of how it was created, I do think the book is a good resource for people in the orbit of labor law and hope that anyone curious will take a look at it (a free sample of the book can be found here). In addition to providing over 100 summaries of key NLRA legal precedent and how that precedent has been applied, the book includes a summary of the basic structure of how NLRB law is established. The book is also full of hyperlinked citations that, when clicked, take the reader directly to the cited document in the NLRB Law database. This feature allows readers to use the book as an easy entry point into further database research.

In the world of labor, the invention of LLMs has been met with some understandable ambivalence. Labor-saving technologies can result in job loss that the American system — with its lack of universal health insurance and generous unemployment benefits — is poorly suited at dealing with. But what I have found in my work is that LLMs can be used for good as well as bad, and in pro-labor and anti-labor ways. Law is an area that seems especially vulnerable to LLM disruption and so advocates for workers have no choice but to keep up. The National Labor Relations Book is an effort to do precisely this.