Findings

Throughout these discussions a set of recurring themes emerged regarding major issues in policy development. We have grouped these into three major areas: policy logistics, student agency, and marginalized students. These topics were raised and refined across multiple discussions and generally reflected real challenges experienced by members of the DOERS community in their work. Unlike the high-level principles synthesized above, these are often administrative or granular, describing specific practices to adopt or problems to avoid.

Policy Logistics

• Clear guidance on what we mean by AI use. As a threshold matter, policies must clearly define their scope. As the term “artificial intelligence” is increasingly used to mean everything and nothing, policies must make it clear whether they encompass, for example, word processing programs that offer AI-supported suggestion features, image tools that rely on genAI for touch-up, and so forth. Without further clarity, a generic “policy on AI” is likely to be unhelpful and counterproductive.
• Policies should not be built tool by tool, but rather around types of activities. In our conversations it became clear that this scope is best designed around common use cases rather than particular tools. Policies should be written to address practices that are in bounds or out of bounds because there are many tools that can be used in a variety of ways and because the rapid churn of new tools means that any tool-specific policy will be quickly outdated.
• Policies for students need to be paired with transparency and policies for instructors. While many policy decisions are currently being driven by concerns about inappropriate or dishonest student use, participants noted that these policies must first make the rules clear for all students. Students are just as uncertain about what appropriate use of these tools looks like and require clear guidance about what is in and out of bounds. Likewise, students face the same set of anxiety about instructors’ inappropriate and dehumanizing use of genAI for grading and evaluation. Policies should be transparent and make it clear how both students and instructors should and should not use these tools.
• Rather than enforcing a single policy, institutions should offer a menu of policy options. Because there is no one-size-fits-all policy for genAI use, some institutions have adopted a menu of potential options for classroom policy. This allows everyone to have a clear, polished, and vetted set of language that will be familiar for students and can be enforced in a consistent fashion across courses, while also supporting the flexibility needed for a diverse curriculum. Instructors who intend for active use of AI can adopt the appropriate broad policy while those who want to limit use can adopt a more closed policy in their classroom. This mix of consistency and flexibility is critical both because everyone needs clear rules but also because of ambiguities created by the inclusion of genAI in technology tools such as Grammarly and increasingly in course shells and publisher-provided learning platforms.
• Policies should recognize principles of academic freedom and student agency and require clear communication about pedagogical values. The open education community includes many different perspectives on the value of genAI as well as diverse concerns about environmental, social, and dignitary issues with these tools. In all cases, policy should be grounded in an accurate and fact-based understanding of the tools and the open education community has a critical role in providing this perspective. Guided by the best information, institutions may choose to offer alternative options that address concerns or to ask that there is clear communication about why it is pedagogically necessary to use or prohibit use of particular tools.
• Policies should rely on normal safeguard for vetting and contextualizing genAI tools. While genAI is a relatively new and fast-moving area of educational technology, it does not exist in isolation from other institutional policies. Policies should rely on well-established safeguards for edtech tools offered by trusted units on campus. Whitelists of vetted tools from IT, guidance on best practice from librarians and instructional designers, and support for values-led practice from open education experts are critical resources for making the best choices about these tools.
• Build policy for values, not copyright concerns. While copyright issues in training genAI models are making headlines, this is an issue that is less likely to impact educators on the ground. Building copyright confidence across educational work continues to be important to understand what educators and librarians can do, but broadly the use of these tools doesn’t change those guardrails. Instead, policy decisions should be built to support good pedagogy and discourage academic dishonesty. Any legal concerns are likely addressed by existing policies and specific terms of use, which often offer indemnification for authorized users. As in most educational contexts, the best policies will lead with pedagogy and values, rather than being sidetracked by broad copyright policy concerns.
• Clear and ethical process for updating. Because changes in technology and best practice are inevitable, policies should have clear and ethical processes for regular updates. These processes should be evidence-based and include appropriate stakeholder input. In particular, it is critical to include student and faculty perspectives in the process in addition to the technical, legal, and administrative expertise necessary to contextualize policy decisions.

Open Pedagogy, Student Agency, and Open Values

• Center pedagogical goals of open education. Open education centers a set of explicit pedagogies and values, from renewable assignments to ethical support for agency. AI policy should be designed to support and center these values. Where permitting or forbidding use of genAI is not aligned with these values, policy may need another look.
• GenAI as a signal – and an engine for open education. Like any tool, genAI may be a better or worse fit for individual people, assignments, and courses. But policy makers and embedded instructors should be aware of genAI use as a signal for structural mismatch, not just laziness. GenAI can be a good tool for some uses, particularly time consuming instrumental work that may be used as an imperfect proxy for understanding and pragmatic creation. Where unauthorized use recurs, it may be a sign that an assignment is not authentic and an opportunity to replace a disposable assignment with more open pedagogy.
• Beware the assumption of cheating. As with any new technology, genAI creates new ambiguity as to what appropriate educational behavior looks like. Regardless of how genAI is adopted, institutions face a danger of doubling down on suspicion and surveillance. Policies that begin with an assumption of dishonesty and focus on catching bad actors rather than supporting authentic learning are likely to do harm to all students and can poison the entire educational enterprise.
• Good tool use is a skill.  Focusing on permitting for forbidding genAI risks missing the more important fact that any tool can be used well or poorly, and quality generally improves with practice. Policy and pedagogy should focus on when and how to use tools well, rather than making categorical assumptions about whether the tool is inherently good or bad. Likewise, policy developed based on an unsophisticated understanding of the tool risks endorsing poor uses or forbidding uses that might be valuable.
• Use genAI for tasks it is suited to. Generally speaking, these tools are effective as a way to buttress technological skill and poorly-suited as a substitute for substantive knowledge. There is significant value in explicitly helping students understand this distinction which is well-aligned with open education’s commitment to critical use of tools. Policies should encourage intentional use in appropriate context and discourage misaligned use, rather than collapsing this distinction into a binary yes/no rule about use.

Open Pedagogy and the Marginalized Students

• Balance career readiness with the need for practice. For many institutions, use of AI has an explicit purpose in addition to addressing subject-specific content: preparing students to succeed in the job market. This need for career readiness may be especially keen for under-resourced, marginalized, and first generation students. In many cases, there is also pedagogical value in learning how systems work by doing things ”the hard way” especially at first. Learning arithmetic can have value even in a world with calculators, but a well-prepared engineer also knows how to make the most out of digital, labor saving tools. Web designers may benefit from hand-coding to learn the fundamentals and from learning to use code editors, developer tools, and the like. Similarly, there are cases where using genAI offers a shortcut that undermines fundamental understanding and cases where students would be poorly-served if they are not prepared to use genAI in their professional roles. As such, policies should balance pragmatic career preparation – including informed skepticism about the tools – with the genuine need to practice fundamentals.
• Provide clear signals for when “good” is good enough. Gen AI can be a powerful tool for combating scarcity but institutions need clear process and bright lines about what is good enough, especially when serving marginalized students. In many cases the outputs of genAI may be adequate but imperfect as with AI-driven accessibility resources, translations, and similar. Clear acknowledgement that genAI is neither perfect nor without value is a good first step and the best policies will be clear about who makes the decision about when “good” is good enough for a particular purpose.
• Center the needs of neurodivergent students. Student agency is a core principle of open education and the need to provide latitude for students to adopt the tools that best-meet their needs is especially strong for neurodivergent students. Those with documented disabilities often have explicit needs that can be addressed by use of genAI. Students who struggle with executive function may also benefit from tools that help with engagement, offer regular reminders, and move them beyond the tyranny of the blank page. As with all areas of disability, the best policies may not require students to disclose and support for neurodivergent students may also benefit those who are not diagnosed. As with curb cuts that help everyone, there may be cases where genAI explicitly supports neurodiverse students first and ultimately all students.
• Make space for students’ own evaluation of their needs. As with issues like accessibility and privacy, individual students may be best-positioned to evaluate their needs around tools like genAI. While there will be contexts where use of these tools is incompatible with pedagogical goals, it is critical to make space for student-led decisions about the use of different specific tools, including for neurodiversity and other invisible needs.
• Evaluate accessibility in the specifics. Accessibility is both good policy and a legal requirement for all institutions. GenAI is already a tool that is being used to meet accessibility needs and new uses are likely to emerge in the coming years. Generalized policies should not paper over specific needs and opportunities.
• Make space for non-native speakers. GenAI also presents important support for language learners and non-native speakers, as a communication tool in the classroom and especially for creating and improving materials for public use. Graduate students preparing scholarly articles, students engaged in public, renewable assignments, and others being asked to do the vulnerable work of sharing openly should have latitude to use the tools that support them sharing the best version of their work.