by Eileen Korey
Nearly 30 years ago, two Cornellians published research to explain a phenomenon captured by this common phrase: “A little knowledge is a dangerous thing.” David Dunning and Justin Kruger could never have imagined that what became known as the Dunning-Kruger effect would have relevance to today’s use of artificial intelligence in legal studies and research. Their paper, “Unskilled and Unaware of It: How Difficulties in Recognizing One’s Own Incompetence Lead to Inflated Self-Assessments” was published in 1999.
“The way an AI model expresses itself, the outputs it generates, sound very persuasive because they are trained to please the user and present things in a decisive, confident manner,” says Kim Nayyer, Edward Cornell Law Librarian and associate dean for library services. “But these AI tools operate very differently from the way we think as humans. Even so-called reasoning tools are not truly reasoning in the way that human lawyers do.” The AI’s confidence in the way it presents findings can be deceptive for law students, lawyers, and anyone who relies on it for learning and research. In fact, a recently published study by Daniela Fernandes and others in Computers in Human Behavior suggests that AI users tend to overinflate their own competence, most likely because the AI tools provide a sense of confidence.
“They need to know what an AI model is capable of doing and what it’s not doing,” says Nayyer. “The wisdom that a lawyer acquires through experience and learning in law school and professional continuing education cannot be replaced by a generative AI tool, no matter how impressive the outputs are. Critical thinking and critical evaluation remain a lawyer’s most vital tools.”
Still, Nayyer appreciates that AI—in all its rapidly evolving forms—can be a powerful tool in learning and research. That’s why she continues to test emerging AI platforms designed for the legal community; share findings with students and faculty; offer e-courses, workshops, and continuing education for alumni; and teach classes to law students in which they can practice with AI tools like Lexis, Westlaw, and Clio vLex Vincent AI, and AI platforms Harvey and Legora.
AI literacy is a fundamental goal of Cornell’s new Center for Law and AI which brings together faculty and researchers in law and ethics, technology, and information science to develop innovative courses and research that advance how AI is shaping the practice of law.
“We were strong in the area of law and AI before the Center, but it was not easy to see, ” says Jed Stiglitz, center director, associate dean for academic affairs, and Richard and Lois Cole Professor of Law. “We needed to create a home, a beacon for people who wanted to be part of the conversation and study emerging issues in a more systematic way. Cornell has a lot of people working in this space. The Center helps make their work more visible and impactful.”
Stiglitz, who teaches administrative law, says he has “long been interested in natural language processing, how to use computer technology to understand language and apply it to legal research. The models that have emerged in the last couple of years have become more powerful—and quite seductive—in reading and processing millions of documents in the time it takes humans to read one.”
In reviewing documents, an AI language model is essentially continuously learning “legalese,” one word and fragment of a word at a time (also known as “tokens”). Stiglitz explains that the AI tool generates outputs based on the “tokens” that come before in a text, pre-dicting what the next “tokens” should be. “The AI models can simulate aspects of rea-soning or consideration of truth or morality or ethics,” says Stiglitz. “But simulation is not the same as human comprehension or judgment. It’s a pattern completion based on statistical relation-ships learned from training data.”
Stiglitz says the growing sophistication of AI tools demands that law students become “AI enabled and technologically fluent so they understand what AI can do, what it doesn’t do, and how humans can guide its development and usage.”
“I first encountered AI platforms in our lawyering program,” says Jessica Rosberger ’26, executive editor of the Cornell Journal of Lawand Public Policy. She recently wrote about her experiences in the journal: “Chatter about artificial intelligence (AI) echoes through the halls of law schools—whether it’s a professor disparaging AI and its capabilities in the classroom or a classmate whispering to you about a new way AI saved them time on some laborious task or assignment.”
When she first began her legal studies at Cornell, Rosberger says many students didn’t want to disclose they were using AI (“it felt like academic dishonesty”) but now it is “explicitly discussed” and used as a tool in studies and research. “I feel confident now going into practice to understand how the technology is evolving. I don’t know where it’s going, but I want to keep up with it,” says Rosberger, who joins the Manhattan District Attorney’s office after graduation. “In my view, Cornell is maintaining its integrity as a top law school by integrating AI into coursework and clinics. We have a professional responsibility to keep up with the technology and the platforms available to us.”
This sense of responsibility is especially evident at Cornell Tech, which offers courses and programs—including a Master of Laws (LL.M.) in Law, Technology, and Entrepreneurship degree—bringing together experts in engineering, computer science, design, business, and law “to build the foundations for new digital technologies—especially AI.”
“Our faculty are thought leaders in artificial intelligence and machine learning,” says David Reiss, clinical professor of law and research director of the Blassberg-Rice Center for Entrepreneurship Law. “The Law School has always been committed to ensuring that our graduates are practice ready. The integration of AI into the curriculum provides them with better tools. It will make them better lawyers and give them a leg up in practice.”
Reiss co-authored an article in Bloomberg Law explaining why law schools should teach how to integrate AI into practice. “Different AI products lead to wildly different results. Just demonstrating this to law students is very valuable, as it dispels the notion that AI responses can replace their independent judgment,” writes Reiss. Simulations were designed in which students were asked to complete the same transactional tasks, like drafting a contract or creating a client email, using different AI tools. The results were different because each platform drew from different sources or used different algorithms to interpret language—some provided a helpful first draft, while others did not, or even hallucinated.
“Our goal is student learning. It was for this reason that we like to deploy the AI tools at the end of our exercises: You do the work and then interrogate it with the AI tools of your choice,” explains Reiss. Just recognizing the capacity for different platforms to produce different results is critical. “AI is not a replacement for lawyers. We want our students to understand how it can be an enhancement for lawyers. It can increase efficiency and help meet tighter deadlines. The lawyers who adapt to AI will succeed and those who put their heads in the sand will fall behind.”
In an article published last year in arXiv, Stiglitz and col-leagues from Cornell’s Depart-ment of Information Science advanced the reasoning behind a center dedicated to AI and the law: “The challenges of legal AI are both specific to the legal domain, and confounded with the expectation of AI’s high performance in high-stakes settings. We identify three areas of special relevance to practitioners: data curation, data annotation, and output verification. First, it is difficult to obtain usable legal texts. Legal collections are inconsistent, analog, and scattered for reasons technical, economic, and jurisdictional . . . Second, legal data annotation typically requires significant expertise to identify complex phenomena such as modes of judicial reasoning or control-ling precedents . . . Finally, AI-supported work in the law is valuable only if results are verifiable and trustworthy . . . We call on both legal and AI practitioners to collaborate across disciplines and to release open-access materials to support the development of novel, high-performing, and reliable AI tools for legal applications.”
Some law students are already trying to improve AI for practical usage. When Biying Cheng ’25 was at Cornell Tech’s Entrepreneurship Clinic, she assisted a client who designed an AI chatbot to help tenants deal with landlord issues and housing court. “We devised questions to test what level of detail the chatbot could and should pro-vide, considering legal liability and jurisdictional issues,” says Cheng, who was already comfortable with AI. “I’m not a native speaker, and ChatGPT helped me draft emails, outline memos, and find resources. I called it Professor G and asked it to explains words I was unfamiliar with.”
Cheng’s pursuit of a J.D. came after receiving a master’s in international finance from Columbia University and a B.A. from The Chinese University of Hong Kong. Her work with the AI chatbot was a full-circle moment for someone who had witnessed Chinese students in Hong Kong participating in protests because of a serious housing problem. She saw how the law and AI could help New York City tenants facing housing issues. Cheng also co-authored an article with Professor Reiss on the real world impact of crypto and block-chain on tenants and real estate investors.
“The legal industry tends to be pretty conservative,” says Cheng, who now clerks for the U.S. District Court, Eastern District. “But lawyers should want to become fluent in AI as a way to understand how it can impact lives in better ways.”
“We’re going to look to our younger colleagues to move us forward,” says Reiss. He believes that AI can help “refine legal judgment” with the right kind of prompting and critical review. AI can help lawyers “stress test” their own reasoning , identify blind spots, and learn about novel issues.
Still, cautionary tales abound regarding AI “hallucinations” with judges sanctioning lawyers for citing cases that aren’t real. “I prefer the term fictitious output or fictitious citations,” says Nayyer. “An output that is false is a combination of words the tool has generated based on the language model it has access to. But just because they make errors doesn’t mean we should stop using them or improving them. A tool in the hands of an expert crafts-person still needs to be sharpened.”
“The question shouldn’t be whether this tool is perfect, but whether it makes you do the job better,” says Stiglitz. “We believe it’s our responsibility at the Center to help shape how AI is used responsibly to improve the practice of law.” ■
