Hood College put AI skeptics on its task force, shaping policy

When Hood College built its AI task force, the volunteers came first: early adopters from STEM, education and computing who wanted to experiment. Then the co-chairs went looking for the people most likely to slow them down. One faculty member had raised real concerns about the ethics of AI. Another, from one of the college's own tech departments, had spoken up at faculty events about students leaning on the technology before they understand the basic concepts underneath it. Both got invitations, and both joined.

That choice shaped everything that followed. So how do you write an AI policy that your early adopters and your sharpest critics can both live with? Between February and June 2026, Hood's 18-person task force ran a campus-wide scan and turned the results into an AI hub and a guidance document for the college's president. According to co-chair April Boulton, associate provost and dean of the graduate school, it worked because the group argued, not in spite of it.

The skeptics got their seats on purpose

The task force came out of a request from Hood's president late in the fall of 2025. Boulton remembers the president's frustration that AI had shown up in every inbox and every higher ed news story, and her push for a response that covered the whole campus. Faculty in separate departments would not talk it through on their own while staff talked about efficiency somewhere else. The point was to build what Boulton called "an information highway" between academic and administrative units from day one.

The pairing at the top set the tone. Boulton came from the academic side. Her co-chair is Hood's risk manager and director of business services. Formally appointed in February 2026, the two had a short, specific charge: run an environmental scan of what the campus was experiencing and wanted, gather broad representation, and deliver a guidance document in time for the June trustee meeting.

The first shortlist ran about 10 people. The obvious picks were the director of the Center for Teaching and Learning and the coordinator of online instruction. A call to department chairs brought in a wave of enthusiasts. Looking at that list, Boulton saw a gap. "I think we need to have some folks who are gonna... for lack of a better phrase, keep us honest and keep us grounded in, you know, the concerns around this technology," she said.

The staff side filled out the rest: someone from records and library staff, the director of admission, a representative from faculty services and administrative support, and institutional research. The group topped out at 18. Boulton said it felt unwieldy and too big, and that it did not make the compressed timeline any easier. She still thinks it was the right call.

The friction kept arriving at one request

Boulton is direct about what that mix did to the meetings. The size of the group and the clearly opposed points of view made the work harder and "slowed us down considerably," she said, and some sessions got heated, especially the ones on syllabus designations. The co-chairs treated that as part of the job.

The environmental scan gave those arguments something to push against. The task force built it on NIST best practices for AI in higher education and drew engagement from more than 40% of faculty and staff, enough that Boulton felt confident in the feedback. One request kept coming back: clear policy, with some flexibility by discipline. Respondents did not want a free-for-all where every instructor sets their own rules. They also did not want a campus-wide ban.

That surprised the co-chairs. Boulton noted that some institutions have taken strong anti-AI positions, and she had not expected so many people at Hood to ask for guardrails rather than freedom. Writing guardrails that fit a lab course and a writing seminar at once is exactly where the friction came from.

The result is two layers. Hood now has a college-wide generative AI and emergent technology policy, and departments and administrative units are encouraged to build more specific rules on top of it. Boulton describes it as a framework that gives the campus the structure it asked for while staying open enough to be refined by each discipline or unit.

Data rules were the easy part to write and the hard part to stick

Privacy and intellectual property did not stay in the classroom. Faculty worried about protecting their own IP and about making sure submitted work actually reflects student learning. Offices that handle student records, including financial aid, institutional research and assessment, the registrar and admission, had a different worry. These tools make it easy to dump a large dataset in and ask for patterns, and Boulton said that convenience "comes with inherent risk."

The first rule was simple. Faculty and staff may only use platforms the college has vetted and licensed. Free tools Hood has not invested in are off limits for anything involving sensitive data or student work. Sensitive data also became a major theme of the year-one professional development that kicked off August 1, where a contracting partner covered best practices for sensitive data inputs and outputs.

Writing the rule did not end the work. Boulton still hears colleagues talk about platforms the college has not invested in, and she still has to remind them to use the licensed product. "I don't like wearing this hat," she said. For anyone drafting a similar policy, the takeaway is that the approved-tools list is where the work starts, and keeping people on it means teaching it again and again.

Assessment without a detector, and room for students who say no

On authentic assessment, Boulton does not claim Hood has solved it. What faculty get is "a buffet of solutions and workshops and tools," and the answers vary a lot by discipline. Bench-based and experiential work, along with some formula-driven STEM exercises, has held up better. Fall professional development, using both in-house experts and an outside contractor, helped clusters of disciplines move toward assignments that resist AI. Boulton will not call any assignment AI-proof, because she does not think that is possible.

Faculty asked for detection software early on. Boulton said she has not seen good evidence that it is not biased or that it is not wrong too often, and she sees no real appetite for it at Hood right now. What is growing instead are interviews. One graduate faculty member assigned a grant funding problem he might normally give a thesis student and told students to use AI to solve it. Each then sat for a 30-minute interview, graded on a rubric, where he picked apart the work and they had to explain their reasoning and choices.

The same logic has reached admissions. Hood has cut many of the required essays from graduate applications and replaced them with 10- to 15-minute interviews for top candidates.

The friction on the task force mirrors the campus. Boulton described student opinion as bimodal, just like the faculty, with first-year students arriving with strong views before any professor has shared theirs. Most people sit in a "messy middle" that is still working out how it feels. A group of enthusiasts wants to use AI for innovation and efficiency. Somewhere between a quarter and a third of students and faculty are firmly opposed, citing environmental impact and damage to creativity and critical thinking.

Last year that opposition turned into students boycotting assignments in a few courses. When an objection is grounded in a religious or spiritual tradition, Boulton said, the law requires an alternative. For sincere ethical objections that are not religious, Hood encourages faculty to offer alternatives where they can. The exception is a course that is about generative AI, where an alternative is unlikely and the student can take a different course.

A policy built to be rewritten

The task force wrapped up in what Boulton called "a little bit of a mad rush." The structure it left behind assumes the work will keep changing. The ad hoc group is becoming a permanent committee with planned reviews each year and on two- to three-year cycles. Boulton argues the guidance should really be updated every semester, in real time, unlike majors and general education requirements that are often reviewed on five- and ten-year cycles.

Even the wording is meant to last. Hood chose "emergent technology" over "generative AI" or "large language model" because something is coming that those terms will not cover. That phrase appears in a new graduate learning outcome that graduate faculty and program directors adopted at grad council the month before the interview. It commits every graduate program to teach what the technology is doing to its industry, how to use it well, and when to slow down.

Boulton has seen how fast a sure bet can expire. A year ago, colleges were urged to build certificates and badges in prompt engineering. Now, she said, the platforms have adapted so much that the best practices fit into half a day or a couple of lectures. She expects some of today's new AI degrees will need restructuring for the same reason.

The same caution now shapes how Hood picks vendors. In 2021 and 2022, choosing a partner meant a pitch, reference checks and a cost comparison. Now the "prove it" phase feels like "a five part series," with long interviews and real use cases the vendor has to work through. Boulton compared it to hiring, where even virtual interviews can be gamed and so take longer than they used to.

That is why she measures success the way she does. In three to four years, Boulton said, "success won't be that our graduates know the current technology. It will be that they can learn the next one." Hood already uses AI to track alumni on LinkedIn, and she wants to see graduates working on projects and in roles that did not exist when they earned their degrees. A policy that stays useful when the tools change has to be written the same way, and arguing with the skeptics is how Hood got there.

Boulton spoke with Darin Francis for the Hard Truths in AI Education series on Signals in Higher Ed, sponsored by Major League Hacking.