Nettles, Cow Dung, and Legal AI

I’m currently taking a class in Old English, just for fun, and we’ve been translating some medical texts on herbal remedies. A nettle poultice, for example, is good for chilblains and sore joints, and carrying periwinkle will keep one safe from demonic possession, snakes, and envy. It’s easy to laugh or cringe at some of the remedies. From a modern perspective, they might be obviously inefficacious (reciting “wenne, wenne, wenchichenne” over a cyst), obviously superfluous (to stop menstrual flow, use this herb then wait a few days), or just gross and probably harmful (applying cow dung to an open wound). But I’ve been trying to resist the presentist urge to laugh at the remedies and feel superior to their creators. That attitude gets in the way of my actual learning goal, to gain more insight into the lives of the people who wrote and used these remedies, plus it’s just inaccurate! The remedies were written down by intelligent people based on first-hand experiences, second-hand information, and frameworks for understanding the world that are radically different from mine. And many of these remedies would have worked: for example, modern studies have shown nettle to be effective against arthritis symptoms, cow dung actually does have some antibiotic properties although you should not apply it to your wounds, and here’s a great video about a salve shown to be effective against antibiotic-resistant MRSA bacteria.

Simultaneously, I’ve been a guest lecturer in first-year Legal Research and Writing classes this semester, and I’ve found myself trying to inoculate my students against a similar impulse. Although I’ve taught upper-level research classes before, this is my first time teaching 1Ls, so I’m more conscious of how I’m shaping their first impressions of legal research. One example: I assume that few new law grads will encounter print volumes of Shepard’s Citations in the wild, let alone be expected to use them. But the print Shepard’s is still a sophisticated technology, well-adapted to an especially knotty problem, and the best remedy for over a hundred years. I’m glad I don’t have to use it, and I don’t try to make my 1L students use it—I just show them a page before we start working with online citators, and one could teach citators effectively without ever mentioning print Shepard’s—but I don’t want them to take away that the old tools are necessarily worse, or reflect archaic practices that we Internet users have surpassed, and so I avoid talking about Shepard’s in those ways. Instead, I want one takeaway to be that we try to pick the best tool for the job, and that might change as the available technology, the job itself, our understanding of those things, and our criteria for what’s “best” change.

(Side note: Actually, we really just need tools that are good enough, but instead of rewriting that last sentence I’ll just put this caveat here. The drive for constant optimization and finding what’s best instead of what suffices seems like one root of the problem.)

I’m also trying to combat this attitude when I teach AI research tools. Read enough marketing material, and you might wonder if research tools that aren’t based in generative AI are now the equivalent of putting cow dung on your wounds. The product page for Lexis+AI describes it as “a comprehensive legal AI solution for drafting, research, and insights” (emphasis mine), combining Protégé and “authoritative LexisNexis content,” as though the raw content is the only thing worth using from pre-generative-AI Lexis. The new Westlaw homepage that faculty and librarians can preview sends a similar message; the page serves as an ornate frame for an “AI Deep Research” text entry box, with content and other tools relegated to a sidebar and “Keyword & Boolean” search available as a non-default option. The Deep Research press release does mention “Westlaw’s exclusive research toolset,” but it contemplates their use by an AI agent, not a human researcher. These platforms include disclaimers that AI results may contain errors and should be verified by a human, and in a recent interesting interview, LexisNexis’s CEO emphasized the need to keep a “human in the loop.” But if those humans become dependent on AI research tools—and disdainful of traditional tools—their abilities to independently verify the results may atrophy.

Just like with Shepard’s, I don’t want my students to take away that new generative-AI-based tools are inherently superior to other tools by virtue of their new-ness and AI-ness. I do think using Shepard’s online is preferable to the print for nearly every person and use case, but because of its actual usefulness for the task, not just because it’s online or newer. I’d like to set up my students to evaluate the flood of AI-based tools in the same way: AI or not, are they sufficient or meaningfully better for your purposes? What are those purposes anyway? (As 1Ls, they’re still learning what they even need research tools for.) What does “sufficient” or “better” mean for you? Does it include anything beyond the quality of the results the chatbot gives you, like time spent verifying the AI results, ease of use, cost, sustainability, impact on your own learning process? To be clear, I think individual students should and do answer those questions in different ways. But it’s important for them to consider those questions regardless of the novelty and technological basis of the tools.

Going back to the Old English remedies, I don’t mean to draw too direct of a comparison. The main thing that changed between print Shepard’s and online Shepard’s is that new technologies developed, not that the foundations of our legal system radically changed like our ideas about medicine did, and I think everyone agrees that print Shepard’s was effective. So I’d like us to see the older tools as more like using a nettle poultice for arthritic joints than like putting cow dung on a wound: effective and well-tailored to their purposes, even if we’ve found ways to improve on them or do things differently today, and with some core features that we can find useful no matter the technological setting. And before lumping in our current non-generative-AI-based tools with the poultices, talismans, and dung, I’d like us to consider which remain the best (or the good-enough) tools for our various purposes, and which AI tools and applications are more like the ineffective or even harmful remedies of the past.