87% of Universities Use AI, but Only 26% Have a Formal Strategy: Is the Real AI Gap Who Governs It Rather Than Who Can Use It?
Original Chinese title: 87%的大學已經在用AI，只有26%有正式策略：真正的AI落差可能不是「會不會用」，而是誰負責治理？
A UNESCO IESALC and UNU-IAS study of 200 higher-education institutions across 19 Latin American and Caribbean countries found 87% using artificial intelligence somewhere, but only 26% with a formal strategy; it is regional evidence, not a Taiwan statistic.
Sulangal｜卑南族｜資深文化與影像工作者
A Puyuma cultural and visual-media worker focused on generated content, educational communication, and Indigenous representation.
# 87% of Universities Use AI, but Only 26% Have a Formal Strategy: Is the Real AI Gap Who Governs It Rather Than Who Can Use It?
What 87% and 26% Each Mean
The UNESCO IESALC and UNU-IAS study covers 200 higher-education institutions across 19 countries in Latin America and the Caribbean. Its official summary says 87% reported using artificial intelligence in at least one activity, while only 26% had a formal AI strategy. These percentages measure different things: adoption and institutionalization. They are not a statistic about Taiwan.
Put the Study Scope Beside the Headline
When 87% becomes a headline, readers can forget its denominator and geography. Nineteen countries and 200 institutions provide a regional institutional view, not a global census or a Taiwan sample. Careful translation states the population, timing, activity areas, and limits so the statistic is not turned into context-free anxiety or optimism.
Adoption Appears Across Several Activities
The news release says artificial intelligence appears in teaching and learning, research, administration, community engagement, and governance. Teaching and learning are most common, while 57% reported research use for data analysis and quality support. These figures do not prove every tool works; they show that governance must cover research data, administration, and public services as well as student work.
The Gap Is Institutional Pace
Faculty, researchers, and students may experiment before formal strategies exist. Rules then vary by course: disclosure may be required in one class and banned in another, while research and administrative errors may lack an accountable owner. Governance turns dispersed experimentation into an institution that can be questioned and held responsible.
Student Confusion Is a Governance Signal
The UNESCO IESALC news release says student training lags behind staff training, with about one in two students reporting confusion about appropriate use. This should not be reduced to student irresponsibility. Ask whether rules are clear, courses conflict, students without paid tools are excluded, and students can challenge a mistaken judgment.
Teachers Need Time for Co-Design
AI literacy is not a tool list emailed to teachers. Teachers need time to examine learning goals, data risk, assignment evidence, and assessment, deciding when a tool adds understanding and when it replaces practice. If institutions leave teachers to track fast-changing products alone, responsibility and inequality shift to individuals and departments.
Assessment Must Measure Learning, Not Polish
Governance is neither a total ban nor handing every assignment to a generative model. Define what the task measures, then use process records, oral explanation, source checking, version comparison, or practical work where appropriate. Students need clear boundaries. Academic integrity must be redesigned with learning, not outsourced to a detector.
Do Not Hide Public-Institution Constraints
The news release notes that private institutions tend to lead in strategy, staff training, and governance, while public institutions face resource and capacity constraints. Policy should therefore provide infrastructure, expertise, data protection, and training funds rather than impose identical duties without means. Otherwise governance becomes an unequal compliance burden.
Indigenous Education Adds Data Responsibility
In Indigenous education, artificial intelligence may process languages, images, families, and cultural materials. Beyond operation, governance must ask who authorized use, what cannot be public, whether models misrepresent communities, whether students are compelled to provide training data, and how benefits return. This is where educational quality meets data sovereignty, not an ornamental cultural add-on.
Policies Must Name Roles
A formal strategy should name who approves tools, protects data, answers appeals, monitors errors, and can pause use. If it only says use responsibly, no one knows whom to call after an incident. Governance should include student representatives and affected communities, not only IT departments and senior management.
Procurement Is Not Only an IT Matter
When a university uses a third-party AI service, procurement terms determine retention, training reuse, cross-border transfer, and subcontracting. Contracts should address data classification, deletion, audit rights, accessibility, language quality, and exit conditions, with teachers and students informed. Buying accounts is not buying governance.
Monitoring Must See Outcomes and Inequality
The study recommends monitoring and evaluation, which should not count only logins or courses. Institutions should examine learning outcomes, error rates, appeals, data incidents, performance across languages and access needs, teacher workload, and alternatives for people without tools. High average adoption is not governance success if one group is misjudged more often.
Begin with a Reversible Pilot
A university can pilot one course or administrative process, first defining purpose, data, risk, acceptable failure, and stop conditions. A pilot should not silently become a campus standard. Report who participated, who did not benefit, and where data went, then decide whether to revise or stop. Reversibility is part of governance capacity.
The Regional Study Is a Mirror, Not an Answer
The study’s value is showing that adoption and governance can move at different speeds and raising questions about training, strategy, student participation, monitoring, and public investment. It cannot answer Taiwan’s legal, linguistic, institutional-data, or Indigenous-education arrangements. The useful lesson is to ask the same questions locally, not paste 87% and 26% onto Taiwan.
The Final Object of Governance Is Trust
When a university can explain where artificial intelligence is used, who is responsible, how data is handled, how students appeal, and how errors are corrected, trust can be grounded. Trust is not asking people to believe technology; it is knowing how to question, exit, and hold someone accountable. The contrast between 87% and 26% says governance cannot wait for adoption to finish.
AI Literacy Includes Choosing Not to Use It
If a course treats AI use as the only sign of modern competence, students without tools, students who will not upload data, and students who need foundational practice are penalized. Literacy should include when not to use a tool, how to preserve human reasoning, protect others’ data, and state limits. Choosing not to use AI can be responsible judgment.
Governance Must Hold Disciplinary Differences
Medical, language, engineering, arts, and Indigenous research data carry different consequences, so one campus rule cannot replace disciplinary judgment. A university policy can set minimum rights and appeal standards, while departments and communities add contextual data, assessment, and professional safeguards. This is not fragmentation; it translates common responsibilities into context.
Rebuild Learning Relationships through Transparency
Students need to know not only whether a tool is allowed, but why, where it is used, who sees data, how errors are handled, and what alternatives exist. Teachers also need institutional support rather than sole blame after an incident. When transparency becomes course dialogue, AI governance becomes part of democratic learning rather than only an administrative document.
Make Governance a Learnable Institution
Each AI pilot should leave a record that the next class, teacher, and community can inspect: whether the purpose was met, which errors appeared, who was excluded, what data was deleted, and how rules changed. This is not paperwork for its own sake. It prevents each update from starting with guesswork and shows students that institutions can hear feedback.
Continue asking by role
- University teacher: Ask about evidence, limits, and feasible action.
- Student: Ask about evidence, limits, and feasible action.
- Academic and information governance leader: Ask about evidence, limits, and feasible action.
- Indigenous education worker: Ask about evidence, limits, and feasible action.
Sources and further reading
Role-Driven Inquiry
Yuan Media AI | Guided Questions by Role
Select a role first, then review the complete questions from that perspective; clicking a question will display Yuan Media AI’s response in real time.
Select a role
University teacher
Examine the article's evidence, governance boundaries, data rights, and actionable steps from the perspective of a University teacher.
Choose a question
AI use and content-safety disclosure
This article draws on official and research sources. Evidence, limitations, and analysis are distinguished.