Artificial Intelligence Is Used by 87 Percent, but Only 26 Percent Have a Formal Strategy: The Governance Gap in Higher Education
Original Chinese title: 87% 已經在用人工智慧，只有 26% 有正式策略：UNESCO 調查 200 所大學後提醒，真正落後的可能不是工具，而是治理
A UNESCO IESALC survey of 200 higher-education institutions in 19 countries found that 87 percent use artificial intelligence in at least one area, while only 26 percent have a formal strategy.
鍾靜蓉｜台科大數位教育博士
A researcher in digital teaching strategy and metacognitive data reasoning who focuses on artificial intelligence in teaching, assessment, faculty development and higher-education governance.
Artificial intelligence in universities is no longer a question of whether it will be used. Students, teachers, researchers and administrators already use it. The difficult questions are which uses are acceptable, which must be disclosed, what data cannot be uploaded, which assessments need redesign and who is accountable when a system fails.
A 2026 regional study by UNESCO IESALC and the United Nations University Institute for the Advanced Study of Sustainability surveyed 200 higher-education institutions in 19 Latin American and Caribbean countries. Eighty-seven percent used artificial intelligence in at least one area, but only 26 percent had a formal strategy. Together, these figures show that daily practice has spread much faster than institutional governance.
The gap first appears in classrooms. Without shared rules, each instructor decides whether students may use generative tools. One bans them, another requires disclosure, another treats them as writing partners, and another says nothing. Students face inconsistent expectations across one institution and learn to guess each teacher's boundary instead of developing artificial-intelligence literacy.
Assessment is the second challenge. When artificial intelligence can summarize, translate, code, rewrite and build presentations, a take-home report alone reveals less about what a student can do. Assignments need not disappear, but evaluation can move from the final product toward the process. Students may preserve inquiry records, compare outputs, identify errors, explain sources, defend work orally and state where their own judgment entered. Reasoning, criteria, verification and responsibility matter more than polished prose alone.
Third comes data governance. Uploading unpublished manuscripts, interview transcripts, personal data or participant information to an external model can breach privacy and research ethics. Sending an entire class's work to an unknown grading service creates similar risks. A university strategy must therefore cover data classification, approved services, sensitive information, third-party contracts, retention, cybersecurity and accountability, not merely publish a recommended-tool list.
Fourth is equity. Paid systems may offer larger context windows, stronger models, better file analysis or more usage. If a course assumes access to an expensive service, household income becomes an educational advantage. Institutions encouraging artificial intelligence need to consider campus licenses, shared computing, open models and workable alternatives so that inability to pay does not become inability to learn.
Fifth is faculty expertise. Policies often concentrate on student cheating while teachers also need new knowledge. They must understand hallucinations, the difference between retrieval and generation, changing data, varying outputs and when human judgment is necessary. A single two-hour workshop cannot supply that capacity; it requires continuing professional development.
The survey also found adoption moving upward from faculty, researchers and students rather than following policy. Technology entered practice before rules arrived. Governance based only on restriction can alienate users, while no governance leaves responsibility unclear. Better rules draw on the experience of people already using the systems.
This is where Two-Eyed Seeing can contribute. Central governance understands law, security, procurement, academic ethics and quality assurance; teachers and students understand daily situations. A university can set a shared rule against uploading sensitive data while allowing disciplines to define appropriate use according to learning goals. It can require a common disclosure practice without forcing every course into the same assignment format.
Courses involving Indigenous and cultural knowledge add another layer. Not everything that can be digitized should be placed in a public model, and ordinary copyright does not capture every right attached to cultural data. A campus retrieval system, knowledge assistant or teaching model should build source, interpretive authority, local context and data governance into its design. Data is not a set of fragments without history or relationships.
Artificial intelligence can also improve teaching governance. Institutions may analyze anonymized course-design needs, teachers may produce exercises at varied levels and review them, and students may ask systems for counterexamples, sources or perspectives. The point is not to give judgment to a model, but to use the model to make human criteria more explicit.
The gap between 87 and 26 percent is not a story about a lagging region. It is a mirror for universities worldwide. Once use is widespread, digital maturity is no longer measured by whether a university adopted artificial intelligence. It is measured by whether the institution knows why it is used, who is responsible, where data goes, how learning is assessed and how access remains fair.
The next stage of competition may depend less on buying the newest model than on building better governance. Good rules neither prohibit every innovation nor leave every risk with an individual instructor. They give students clear boundaries, protect teaching expertise, allow administrators to follow risks and preserve a place for different cultural and knowledge systems. When artificial intelligence becomes infrastructure, shared decision-making is the capacity universities must upgrade.
A formal strategy must answer everyday questions
A formal strategy must be more than a declaration. Teachers, students, researchers and administrators need to know which uses are allowed, which assignments require disclosure, what data cannot be uploaded, who reviews third-party services, how disputes are appealed and when rules are reconsidered. Disciplines need not answer identically, but shared boundaries must be transparent so students do not guess and teachers do not carry platform risk alone.
Assessment design needs the most collective discussion. A blanket ban may drive use out of sight, while judging only the output hides what a student learned. Practical options include declaring the scope of use, preserving verification and revision records, explaining work orally, comparing evidence and grading interpretation, citation, data protection and reflection. Academic integrity becomes an honest account of work, judgment and responsibility rather than a yes-or-no question about touching a tool.
Indigenous knowledge and cultural data require explicit governance. Possessing scans, transcripts or research files does not automatically grant a university the right to upload, retrain on or publicly retrieve them. Teaching and research systems should record provenance, permission, retention, secondary use and withdrawal. When rights are uncertain, knowledge holders and responsible governance bodies should decide. Artificial intelligence can support bounded exercises and verification, but not replace human judgment about relationships, consent and interpretive authority.
Yuan Media AI | Continue by role
Choose your role to view follow-up questions designed around its responsibilities.
Sources and further reading
AI use and content-safety disclosure
This English edition is an AI-assisted translation based on official and research sources, with established facts, open questions and analysis kept distinct.