Daniel B. Rodriguez (Northwestern), Law School Deans as Cheerleaders, and the Delicate Marketing Dance, Part I: The Case of Agentic AI:
Times in the AI world are changing, and fast. The latest iterations of AI are described as “agentic.” Tech companies such as OpenAI and Anthropic are developing tools that are notably autonomous, able to engage in what looks much what we regard as reasoning. They aren’t limited to using LLMs to respond directly to human prompts. Therein lies the essential difference. By contrast to generative AI tools developed and refined over the past few years, agentic AI is capable of carrying out a complex set of tasks through an iterative process that doesn’t necessarily depend on significant human action and interference. One thought leader described to a group of us working on ethical protocols for the use of AI by practicing lawyers that the human-machine interaction should distinguish between the human being in the loop, as traditional versions of LLMs presuppose in order for these tools to be effective, to humans being on the loop. Indeed, the essential utility of the agentic AI tools is that they enable humans to develop a goal and then task the tool with doing the research and also the reasoning to yield outputs that will realize this goal. One technologist has described the agentic AI advantage as being that these bots can engage in “recursive self-improvement.” Whereas generative AI has as the ultimate end product newly discovered content, agentic AI is designed to realize defined goals and to undertake the sequence of tasks necessary (not only research, but also reasoning and analysis). In the legal practice context, it is the difference between a tool that is highly successful at doing legal research and a tool that can undertake a full-throated legal task such as, for example, constructing a non-disclosure agreement that meets all the relevant legal requirements of a particular jurisdiction and accomplishes the objectives set out by the lawyer at the time that her “agent” is tasked with doing this project.
A search from ChatGPT describes this comparison in the form a very simple chart.
Generative AI typically performs one-step transformations:
Prompt → Output
Agentic AI performs multi-step reasoning chains:
Goal → Plan → Actions → Feedback → Revised Plan → Result
So now we come to the dilemma for the modern law dean.
Whereas generative AI could be described by its cheerleaders as a mechanism that would improve legal practice and the welfare of law school graduates, by giving them access to tools that would facilitate legal research and drafting, developers and champions of the newest agentic AI models insist that these products now or in the near future will fundamentally replace the need for many lawyers devoted to solving their client’s discrete problems. To be clear, this is not the same as predictions of how so-called Artificial General Intelligence (AGI) will unfold, so to reflect a world in which robots are completely autonomous and which there is no real daylight between the cognitive functioning of humans and of machines. Rather, it is a world in which humans remain necessarily on the loop — certainly to define the objectives of the client and also to frame for the bot the universe (which jurisdiction? Which sources of law? etc.) that is to be used for their project. Agentic AI tools function as mechanical agents to flesh-and-blood human principals. Nonetheless, what we colloquially refer to as manpower will be considerably affected by growing use of these tools. After all, why sic five associates on a project, working with, say, one or two partners, when the agentic AI can do basically anything that the directing partner needs or wants from their associates? Do the arithmetic, and we can see the worry that lawyer employment will be meaningfully affected and not in a positive way.
What is an enterprising dean supposed to do with all this? Is she supposed to push the collective opinion needle in the direction of more fear about these developments, either by introducing more skepticism about the utility and value of these tools, or insisting that there are sound reasons for consumers (including here both law firms and clients) to slow their roll, and maintain traditional patterns of hiring and mentoring? The dean may be caught between their beliefs as prognosticators and their perceived fiduciary responsibilities to their stakeholders. …
One of the “what’s next” questions the leaders of agentic AI and law are grappling with is how to adapt their still mostly off-the-rack tools to particular settings. Without suggesting that the best pathway is one that leads to a hundred or so bespoke tools that are designed around the identified needs of particular law schools, there is surely good cause for agentic AI developers to work collaboratively with law school faculty and leadership to improve the utility of such tools for what this newest generation of lawyers, and also their clients, will want and need to do legal practice in the “right” way. We have some experience in the law-tech ecosystem about the value of well-fashioned collaborations; and agentic AI seems like a good area for fueling and sustaining exciting and effective collaborations.
My optimistic ex-dean self tells me that there is potential for deans to work their way through the dilemma of championing new developments in technology that present serious risks to the employment model, and perhaps also the training model, of traditional legal education. One just needs to be exceptionally intentional and creative about exploring ways in which agentic AI presents opportunities, while also candidly acknowledging some of the risks. Likewise, deans can communicate to their important audiences that their law schools are not obstinately resisting, but are truly embracing, developments and initiatives in both generative and agentic AI that can possibly advance the welfare of individuals who lawyers service, while also enhancing rather than hobbling the education of new lawyers. To take just one quick slice at this, think about how progress in technology-enabled research and reasoning might help bring down the cost of legal services and help close the access to justice gap. (I will focus more squarely on A2J in my next post on this subject).
Bloomberg, AI and the Coming Mismatch Between Law Schools and Law Firms
Editor’s Note: If you would like to receive a daily email with links to legal education posts on TaxProf Blog, email me here.
