You Can’t Outsource Understanding—How 2 Universities Are Teaching AI
By the second week of Dijo Alexander’s applied artificial intelligence course at the University of Wisconsin-Milwaukee, students are building an AI-first potato chip company. Students create the brand, write its story, design a logo, compose an advertising jingle, and even produce a video. The exercise gives them a low-risk way to explore what AI can help them do before they apply the same tools to forecasting, operations, and client work.
While students marvel at what they can produce, they also wonder whether it is really their work. They did not draw every image, write every line of code, or compose every note. Alexander, a professor of practice at UWM’s Lubar College of Business, assures them otherwise.
“This is your work,” he says. “You did it with the new set of tools that are available to you.”
Students Are Using AI Without Guidance
This exchange illustrates a problem higher education has not yet resolved. Students are already using AI, often in the shadows and with little guidance, while universities still debate whether the technology belongs in the classroom. The result is not less AI use, but less thoughtful use.
In Microsoft’s 2026 survey of 3,345 respondents across K-12 and higher education, 92% of students and education leaders, along with 88% of educators, said they had used AI for school-related purposes. Yet 77% of students and 53% of educators reported receiving no formal AI training.
In Teaching AI The Tool Is Not The Lesson
Alexander does not describe his course as an AI class.
“This is a business class where we will help you look through an AI lens,” he said.
His students are not being trained as machine learning engineers. They are learning how AI changes the work of marketing, finance, operations, and management.
He also avoids much of the jargon that has grown up around generative AI. Instead of dwelling on prompt engineering, he talks about verbalization. Students have to explain what they want, supply the relevant context, and then judge whether the result is useful. These are not new skills; they are extensions of writing, reasoning, and communication.
That is one reason MBA student Nick Stricharchuk found his undergraduate English major surprisingly useful. Working well with AI requires clear instructions and a clear understanding of the assignment.
He makes the point with a typewriter analogy.“No one was looking at the typewriter going, ‘Why haven’t you written me the great American novel yet?’” he said.
The tool does not choose the problem, supply judgment, or accept responsibility. The person using it still has to do those things.
As the semester progresses, Alexander’s students move from the potato chip exercise to real business problems. A project on food demand in UWM’s dining operations led students to consider forecasting, waste reduction, and applications for external businesses. Other students have taken prototypes back to their employers and found themselves presenting to senior executives earlier than they expected.
AI lowers the cost of trying an idea. A business student can build a prototype, show it to a client, and quickly learn whether it solves the problem. Alexander tells students to start with a skateboard before trying to build a Cadillac.
“You cannot outsource understanding,” Alexander said.
AI can collect information, evaluate options, and build a prototype. It cannot decide whether the result makes sense or serves what the client needs.
Medicine Makes The Stakes Clear For Teaching AI
At the University of Kentucky, the same argument is taking shape in medical education, where the risks are higher and the limits of the technology are easier to see.
Tama S. Thé, a pediatric emergency physician and assistant professor at the University of Kentucky College of Medicine, expects AI to change medicine through predictive analytics, remote monitoring, and clinical decision support. The challenge, according to Thé, is that physicians are adopting AI faster than medical schools are teaching them how to use it.
Thé’s student Hunter Colson encountered the problem from the learner’s side. Medical school still runs on dense lectures followed by hours of flashcard drilling. Colson found that the hardest part was not the complexity of any single fact. It was connecting thousands of facts and retrieving the right one at the right time.
Colson and two classmates turned to AI, building a Socratic tutor that could turn static course materials into an active learning experience. Instead of giving students an answer, the system asks questions, identifies gaps, and builds connections among concepts.
Making this work required as much medical knowledge as technical experimentation. A general AI model does not know which misconception a medical student is likely to have or which question will expose a weak point.
“The domain expertise just becomes a lever for the AI,” Colson said.
Practical AI use does not make theoretical knowledge optional. It makes it easier to see who actually understands the material.
A student who cannot recognize a weak explanation, a missing assumption, or a fabricated citation is not prepared to use AI safely. The answer is not to keep students away from the technology but to provide enough subject knowledge and supervised practice for them to judge the output.
Colson draws a clear line between different uses. Asking a general-purpose chatbot what medication to give a critically ill patient is reckless. Asking a system grounded in assigned course materials to turn a lecture into a guided study session is a different activity with a different risk.
AI literacy includes knowing the difference.
When Teaching AI Start With The Problem
Both universities organize their AI work around problems rather than products.
At the University of Kentucky, an undergraduate AI literacy course is built around the Bluegrass AI proposal. Students identify a problem in a community they care about and develop a proposal for using AI to address it. The course introduces practical skills such as providing context, detecting hallucinations, and working with agents.
The university’s AI incubator follows the same model. It brings together people from medicine, engineering, law, agriculture, physical therapy and other fields to build projects that no one discipline could develop as effectively on its own. One student built a tool that uses public tax records to identify companies that might support a shelter serving survivors of domestic abuse.
That work is closely tied to Kentucky’s public mission. In rural communities hours from medical specialists, or cut off by flooding, remote clinical support and drone delivery are not demonstrations built for prestige but responses to conditions people already face. For a public university whose graduates often remain in Kentucky, this is as much workforce development as experimentation.
The University of Wisconsin-Milwaukee serves many first-generation and working students in a region built around traditional businesses. Its students work on operational problems for the university and for Milwaukee-area businesses. Many are already employed, so they can test ideas in the workplace while still in the course.
Microsoft’s role at both institutions is infrastructure and scale: Microsoft 365 Copilot for more than 70,000 students and employees and a protected sandbox at UK; agent frameworks, SAP, and Snowflake at UWM.
Access, however, is not an educational strategy.
Microsoft’s own education report calls for ongoing training, practical guardrails, and the integration of AI into everyday teaching and school operations. This is a more demanding goal than simply distributing licenses. Not every assignment should allow AI. A task designed to test independent recall may require students to work without it; a task designed to analyze evidence, test alternatives, or build a prototype may be better with it.
The choice should be explicit. A blanket ban leaves students experimenting without guidance; unrestricted use leaves them without standards.
Teaching AI - A Skill, Not A Shortcut
Universities have two easy responses to AI, and neither is adequate. They can treat it mainly as an academic integrity threat and try to keep it outside the classroom, as the University of Chicago Law School is doing. Or they can provide broad access and assume students will figure it out.
UK and UWM offer a more practical approach: bring AI into the work of learning, tie it to disciplinary knowledge, and require students to remain responsible for what they produce.
Students will graduate into workplaces where the use of available tools is expected. Employers are unlikely to reward someone for spending five hours on work that can be completed responsibly in 20 minutes. They will value people who know what to ask, when to distrust an answer, and how to verify it.
AI used well can help students explore, experiment, and discover where their understanding is weak. Universities should not teach students to hand their thinking over to AI, and they should not pretend the tools do not exist. They should teach them to think well enough to choose the right tool for the job and how to use it well.