A familiar bit of bureaucratic reasoning, described as “politician’s logic” in the British comedy Yes, Prime Minister, goes something like this: Something must be done. This is something. Therefore, we must do it.
I suspect some of the rush to create AI courses at colleges and universities reflects this sort of thinking. Universities recognize that artificial intelligence is changing the workplace. They want graduates to be prepared for those changes. Creating a course, or perhaps a sequence of courses, seems like an obvious response.
But it isn’t clear that creating an AI course will actually address the problem.
AI is already changing accounting, marketing, finance, nursing, engineering, economics, and dozens of other fields. Universities are understandably concerned about preparing students to use these tools. But what does it mean to promise graduates a working knowledge of AI? And who is qualified to teach it?
The problem is that understanding how AI is used in a particular profession requires considerable knowledge of that profession. Even people with the same job title will vary in their specific expertise, and much of their knowledge comes from doing particular things in particular settings. No one professor can know all of that.
This is the knowledge problem Friedrich Hayek described. Knowledge is dispersed among many people in different places and circumstances. Much of that knowledge is local, contextual, and difficult to communicate to others. Leonard Read illustrated a related point in his essay, “I, Pencil.” A pencil looks like a simple thing. But if you go through everything needed to make one—the materials, the tools, the machinery, and all the different processes—you find that no one person knows how to make a pencil. Different people in different places know how to do different parts. No worker possesses all that knowledge, and no manager or mastermind does either. And it is just a pencil.
Marketing is a pencil. Nursing is a pencil. Accounting is a pencil. And each of these disciplines contains specialties that are pencils of their own. This doesn’t mean we can’t teach the basic principles of accounting; we can and we should. Those principles allow accountants to understand and make use of the tools available to them. But the tools change, and the details of their use vary across different specialties. AI is a rapidly changing tool. Knowing the principles of accounting is not the same thing as knowing how AI is being used in auditing.
Consider my own limitations. I’m an economist, and I’ve never taken an accounting course. I know very little about basic bookkeeping, much less auditing. I can ask an LLM how AI is used in auditing, and it will give an impressive answer. I can learn about document review, anomaly detection, fraud detection, and internal controls.
But I still wouldn’t know enough about auditing to judge how useful those applications are, whether the descriptions were accurate, or what important details might be missing. I certainly wouldn’t be qualified to teach students how to use AI in an auditing practice. AI compounds this problem because we tend to confuse information with knowledge.
Having information is not the same as possessing knowledge. We know some things because we’ve studied them, some because we’ve practiced them, and some because experience has taught us lessons we would rather not have learned. The expression “ask me how I know” usually refers to that last category.
This distinction matters for higher education. AI lets people appear to have knowledge they don’t. Indeed, that is at the heart of much of the concern about students using AI for academic work.
Students can use AI to give others the wrong impression. They might present ideas that aren’t their own, use language they didn’t write, or make it appear that they understand something they haven’t really thought through. But the bigger problem is presenting work as evidence of knowledge or understanding one doesn’t actually have.
Universities should be careful not to do something very similar. When a university promises to prepare students to use AI in accounting, marketing, finance, nursing, and other professions, it is making claims about what its faculty are qualified to teach. But how much do they really know about how AI is used in these different fields? Even within their own disciplines, faculty members are still learning how AI is being used. And so are the people working in those professions.
A better approach would be to identify faculty members who can develop expertise in how AI is used in particular areas of their disciplines, then incorporate that knowledge into upper-level courses. But faculty can’t be expected to develop all of that expertise on their own. Much of the relevant knowledge is held by people doing the work. Universities need to find ways to draw on that knowledge, whether through adjunct faculty, guest speakers, or professional advisory boards. AI in auditing belongs in an auditing course, but practicing auditors may know more about using AI than accounting professors do. The same is true of marketing research, econometrics, and countless other specialties.
The point isn’t to expect faculty to know everything. It’s to find ways to make use of knowledge that is dispersed among many different people. Humility is required. Hayek’s lesson is that knowledge simply cannot be gathered and packaged into tidy course modules.
This is certainly harder than creating a course on AI and asking a handful of faculty members to teach it. It means that many faculty members will have to develop new expertise and continue learning as the technology changes. But that’s precisely the problem we’re trying to address. The relevant knowledge is spread across many different people and disciplines. Proposing to gather all of this into a single course doesn’t somehow bring that knowledge together.
A university-wide course on AI may have a place if the focus is on the ethics of using AI in academic work. Students and faculty alike need to think carefully about how to use these tools without misrepresenting what they know, understand, or have produced themselves. That problem is genuinely shared across disciplines, but the specialized uses of AI in hundreds of different occupations are not.
Universities should be particularly cautious about overpromising and underdelivering. The opposite is much better for an institution’s reputation, particularly over the long run. There is something to be said for a little epistemic humility.
AI has made information much easier to obtain and computation vastly more powerful. It has not solved the knowledge problem.