AI Strategy in Education: Start With the Problem

Sathish:

“You have been at Oakland University for years. Can you share your background and how things have changed in higher ed before we get into AI?”

Dan:

Thanks for having me. I have been with Oakland University in Metro Detroit for nearly 20 years, in roles ranging from financial aid to student recruitment to online learning and educational technology. My full-time role is director of support and innovation for our online learning team, and this academic year I was also appointed provost fellow for artificial intelligence. We are looking at how to strategically introduce AI into the organization, upskill our staff, incorporate it into our research portfolio, and guide faculty and students on that journey. Coming out of the pandemic and then the public release of ChatGPT, saying the last five years have been interesting would be a grand understatement.

Sathish:

“How prepared is your organization for this AI change? Are people anxious or excited?”

Dan:

I am a big believer in institutional readiness. Before we implement a change, we have to move people toward it, and the bigger the organization, the more complex that is. Higher education is different from most industries because we also carry teaching, learning, and critical thinking responsibilities, and agentic browsers can do a lot of damage in an online course. We are moving closer to introducing AI pilots and scaling them to see where we can bring AI in ethically and responsibly. We have held faculty discussions and tried to include student voices. It is not just me: our provost, CFO, CIO, and many others are driving this.

Strategically, I think the most value comes from academic and administrative units unpacking where AI helps them, rather than someone telling them. I ask: what are the pain points in your job, how could AI help relieve them, and what skills do you need to use these tools ethically and effectively alongside your human expertise? I do not present it as AI coming to take over your job.

Sathish:

“When designing a campus-wide AI strategy, what are the foundational pillars? Governance, academics, something else?”

Dan:

Certainly governance. Flagship institutions have a different bucket of resources. For an institution our size, we have to focus on ethics and on what tech stack gives faculty, researchers, and students a secure environment to experiment while protecting institutional data and personally identifiable information. How we govern IT, AI, and data is foundational. I see a lot of aspirations: I saw another organization do this, let’s do it too. I want those wins for our provost, our CIO, and the whole university, but they are hard to achieve without a solid foundation. It is like building a car without wheels or building a house starting with the attic.

Another pillar is knowing what we want to do. We are developing an intake process where you state your business case and problem statement, and we build use cases. We want to identify the problem and then match the tool to it, rather than taking a large language model and going looking for problems. That is backwards. Maybe you already have a system with embedded AI that can help, or maybe we need to look outside the enterprise. The intake process helps us prioritize projects, understand limitations, and estimate short- and long-term return on investment. We also have short-term goals around AI literacy. On February 13th, the Department of Labor released an AI literacy framework that could serve as a model for almost any organization getting serious about AI.

Sathish:

“So the strategy covers students as well as staff across the organization?”

Dan:

Correct. We will have an overarching strategy for tech infrastructure and cultural barriers, since every unit has a different appetite for change. But each stakeholder group needs a different approach. Students worry about jobs: am I in the right degree program? Deans of business schools and economists say nobody really knows. They suspect some job loss and some job creation, and I would be hesitant about anyone who claims to have the answer. Faculty researchers have different concerns. Their grant funders may or may not allow AI in proposals, and their data standards may conflict with our rules. But everyone needs a baseline understanding of the tools and how to use them effectively and efficiently, which drives me back to use cases: match the problem to the tool.

Sathish:

“If a student asks whether the course they are taking will make their future safe, what would you tell them?”

Dan:

I would say it depends, and I do not know, so let’s talk about it more. More students are reaching out to me for these exploratory conversations. Think about their day. A graphic design professor finds creative ways to use and critique AI design. Then the student goes to a creative writing class where AI is banned entirely. Then they go to their internship and are expected to push AI to its limits. Those things are in direct conflict, and it is confusing. I do not want to limit how faculty run their classrooms, but I empathize with students. They are eager, and honestly they have some of the best ideas for using these technologies. We want to teach them to use these tools to complement their critical thinking, and across higher ed we are still working out what teaching critical thinking means in an AI-enhanced classroom.

Sathish:

“Many students without clear answers turn to AI engines to ask which course to take. Is that the right approach when AI can hallucinate?”

Dan:

It is tricky. Agents that do the work for you are a security risk, because you have handed your credentials to a system without knowing what it does with your data. At the same time, I can click the Gemini tab in my Chrome browser and start interacting with any web page. We need to help people be smart consumers. When I shop for flights or hotels, I compare everywhere. But with large language models, people tend to lock into one or two, so they get skewed by how that model was trained. There are tools that run the same prompt across multiple models, which is smart, but you still have to check your sources and your work. Math teachers have always said, show me your work. We are still figuring out what that means when a student works with an AI tool or agent, and whether a final paper is really the best measure of learning.

I give our faculty a lot of credit for taking this on. When you consider a student’s workload, jobs, and family obligations, I understand why handing work to an agentic tool is appealing. But I would ask: do you know what is happening to your data, and do you understand how that could expose your organization to malicious code injection? AI providers are largely unregulated and will not wait for us. I hope industry partners with education to set guardrails, like keeping agentic browsers out of systems that may hold FERPA- or HIPAA-protected data. Legally, we are just not there yet.

Sathish:

“In the next four to five years, will universities redesign their curriculum for an AI-first world?”

Dan:

It could happen, possibly through accreditation, since setting learning outcomes is its own process. Today, the simplest way to bring AI into a course is a learning outcome to explore the ethical pros and cons, considerations, and limitations of AI as it relates to that course. That does not require using any AI tools. It is simply talking and learning. I am 20-some years removed from my undergraduate experience, and a lot has changed, so I love hearing students’ perspectives. If faculty are comfortable not being the expert or super user in the room, there is room for very productive discussions in any discipline.

Sathish:

“Are faculty confident, excited, or resisting these changes?”

Dan:

All of those. We have no control over the rate of change in AI; we are subject to what the tech industry does. What each of us controls is our rate of adoption. Like any classroom, you have early champions you can lean into, people dipping their toes into the shallow end, and people who dig their heels in, sometimes out of fear, anxiety, or uncertainty. Everyone falls somewhere on that learning continuum, just as we do not start students at a master’s degree. We are a little over three years into this AI era, and I think more disruption has to happen to move people along. As the gap between early adopters and holdouts widens, I think peer pressure, in a positive sense, will move more people. I will not mandate anything, and I believe faculty should control their classrooms. But students will demand more, so it is a delicate dance, and we are working to give faculty more training and opportunities.

Sathish:

“Beyond the classroom, where can AI make the biggest difference in university operations, marketing, or enrollment?”

Dan:

I think the earliest wins will be administrative and operational, because we have more control there. Every department has pain points but may not have had the chance to build a business case. A friend at another institution built a RAG model from markdown files describing the problems and common questions each business unit sees, then an agent that triages tech tickets to the right unit. It cut response times and frivolous tickets. We are exploring identity management and password resets synced across our systems. With 20,000-some users, those requests add up, and automating them frees people for higher-level tech issues. Admissions, enrollment, and marketing have opportunities too, but those units understand their own pain points best. A solid process for framing problem statements helps them identify the problem and then match tools to it.

Sathish:

“How is AI becoming a cybersecurity threat, with fake logins and spam, and how do you handle it?”

Dan:

Compromised accounts. One concern I hear at conferences is whether we can log threats when someone downloads a large language model to their machine and brings it onto our network. But the more obvious issue is agentic browsers and systems that have been given login credentials, because it is so easy to hand that information off to get schoolwork done. I feel for our CISOs and security teams. Bad actors are finding creative new ways to infiltrate using this technology. There are always good users and bad users, and our security teams are in for an interesting battle. I am thankful we have good people in place.

Sathish:

“What guardrails or policies should universities put in place to prevent these threats?”

Dan:

I do not believe in outright bans. They do not promote critical thinking. It is problematic that agentic systems can do so much on a student’s behalf, but organizations already use them, so students will need these skills after they graduate. From an education perspective, it would help if those tools could not so easily get into our learning management system. Even Gemini can use computer vision to see anything on my page. I wish providers were more mindful of data protections, so we can use these technologies productively without compromising critical thinking or student learning.

Sathish:

“When the university evaluates an AI tool to buy, how do you decide it is the right one?”

Dan:

It depends on the problem. For enterprise rollouts, some companies offer what amounts to an operating system over your organization that pulls data from all your tools and databases and generates reports through natural language. I see tremendous opportunity there. But we also have to consider consumption and environmental impact. Giving everyone access to 140-some models and the ability to query six to 10 at a time adds up quickly. Students in our computer science and engineering AI programs may need advanced reasoning models, while most people, administrators included, would do fine with fast, smaller models. Releasing everything to everyone without oversight or faculty insight would be short-sighted. We always look at accessibility, security, and compliance. A colleague in our university technology services group has been very helpful in establishing frameworks. We do not have to change our process entirely, but we need to understand how embedded AI or a full AI tool may look different from what we have evaluated before.

Sathish:

“Some organizations set aside a dedicated AI budget. We believe you should not do AI for AI’s sake. Is a dedicated budget the right approach?”

Dan:

It depends on the institution and its direction. Here in Michigan, the University of Michigan has been at the forefront, and it has access to different resources, companies, and alumni than Oakland does. I believe they were working on consortium pricing for Big Ten schools. A colleague of mine has helped establish a Michigan higher ed AI network where tech leaders discuss these implications. We do not want to overextend. The use-case approach shows what problems we are solving. Retention, for example, directly affects revenue. New tools can track student satisfaction, extracurricular involvement, or tutoring attendance so advising can be more proactive. If a spend retains 100 or 200 more students, that is how we weigh it over the short and long term. For schools that can set aside budget with a clear goal, fantastic. For schools with pinched budgets and lean workforces, you have to be more strategic about the problems you solve and the potential gains.

Sathish:

“Enrollment is declining at many universities. Do students prefer online education, or will traditional classes remain?”

Dan:

There is room for both. Online is getting hit hard right now because of what AI can do on a student’s behalf, but we have to rediscover what it looks like. In Michigan, enrollment has shrunk at most institutions. We are still feeling the effects of the auto industry’s struggles in 2009. I was in financial aid then and saw families struggle in ways they never expected. Many left the state for work and never came back, and their kids now attend universities elsewhere. Even so, credit to our enrollment management team: we are still seeing growth, though not as much as in the early and mid-2000s. We also have a new AVP for advising working with enrollment management and academic affairs to raise retention by five or six percent over the next few years. It is an ambitious goal, I fully support it, and he is looking at AI-enhanced tools to get there.

Students do gravitate toward the flexibility of online, and that was happening before recent AI advances. We offer hybrid courses, fully online programs, and HyFlex programs where you can attend in person, live online, or watch the recording later. But labs in nursing and engineering benefit from hands-on work, and connecting with peers, instructors, and advisors face to face matters. Some people really crave sitting across the table from someone on a problem. We just need to be creative in this AI era and teach skills in ways AI cannot easily do for students.

Sathish:

“If every university becomes AI-enabled, could any university compete with the elite schools within five years, since knowledge becomes accessible to everyone?”

Dan:

I do not know about three to five years. Oakland was established in 1957, competing with land-grant and flagship institutions founded long before. That may be too aspirational. A more productive question is who we want to be as a university. We welcome all, but what is our identity and purpose? I do not want to compete with the University of Michigan. I want to collaborate with Michigan, Michigan State, Western Michigan, and Central Michigan. Enrollment management competes for students differently, and I understand that. But if we worry too much about how green the grass is in someone else’s yard, we will not tend our own.

Sathish:

“What will AI-enabled campuses look like in the future? Students attending from home, talking to robots?”

Dan:

A year ago I would not have been talking about agentic browsers doing all your homework. I am curious how student needs will drive what the university becomes. Since I was in school, I have seen recreation centers go up, and they matter because involvement builds a sense of belonging. My hope is for more personalized, adaptive learning that meets students where they are and leverages their skills to get them where we want them to be. Maybe through AI enhancements, maybe robots in some areas, but certainly not in how faculty interact with students. I want that human component to remain at the center, because it is us who give AI its power, not the other way around.

Sathish:

“What advice would you give other institutions thinking about these strategies?”

Dan:

Keep talking. Talk to your students, your peers, and your industry partners outside your institution. I have learned so much just by asking and being present. Listen to your students. I originally thought they were worried about being accused of cheating. The first time I met with a group of students, they never mentioned cheating once. They talked about jobs. Hear that directly from them, early and often, because our strategy directly affects them and we are here to serve them. Identify the right stakeholders, and give people time to get over feeling behind. Support them. We are learning institutions. Let’s model the way and set the example, as we have for years.

Sathish:

“Could these tools help people who are lagging behind catch up?”

Dan:

I think they could, but you still need a skilled hand to keep people on the right path. I have already built the critical thinking skills and foundational knowledge I need for my job. A freshman or sophomore has not had that lived experience. A colleague told a group of students recently that AI can help them reach mid-management level faster than we have ever seen. But we have to be smart consumers: understand what quality input looks like, and do not just accept the output as correct. Review, edit, analyze, and accept or reject it. That is where critical thinking needs to be directed, and it is something education already does.

Sathish:

“Any final comments?”

Dan:

Nobody asked for AI, but it happened. Give yourself credit for at least trying to explore. It went from “we are going to go play” to “we have to go to work.” As long as you understand what information you should not put into AI and you protect your personal data, go play and learn. Experiment and fail often. Failure is learning. Share those experiences and see how others are failing and succeeding. We are all in this together, still figuring it out at every level of the organization. That is okay right now, but we need everybody’s help to push the ship along.