Eighty-seven percent of higher education institutions in Latin America and the Caribbean already use artificial intelligence in at least one area of operation—74% in teaching and learning, and 57% in data analysis and research. Only 26% have a specific formal strategy for AI. And, according to the report itself, even fewer have policies, governance structures, and monitoring and evaluation frameworks in place.

The figures come from a survey conducted by UNESCO IESALC and the United Nations University, involving 200 institutions from 19 countries in the region. The survey reveals an ongoing transformation: the adoption of AI is advancing much faster than the institutional capacity to govern it. For students, the effect is immediate—one in two students reports confusion about what they can and cannot do with AI.

There is one question that these figures do not directly answer, but that should be on the agenda of every university administration: Who is making the decisions that determine how AI is transforming higher education?

To a large extent, the teacher.

The decision was made

When a university does not establish an AI strategy, it does not mean that nothing is happening. It means that many decisions lack sufficient institutional guidance.

For each course, the professor must decide what students can delegate to the machine and what they must do on their own. They must distinguish between legitimate assistance and academic fraud, verify the quality of what AI produces, and determine how to assess learning that can no longer be inferred simply from the work submitted.

They have to do all of this while answering students’ questions, negotiating expectations with colleagues, and interpreting rules that often come into play after the technology has already been introduced into the classroom.

The absence of an institutional policy does not prevent these decisions from being made. It merely shifts the responsibility for making them to the teacher.

It is important to make a distinction: not having a specific formal strategy for AI does not necessarily mean having no policy or guidance at all. The regional survey itself identifies institutions with broader digital strategies and others with partial guidelines. The problem is that institutional governance is not keeping pace with the spread of the technology.

The result is a form of involuntary decentralization. Each professor develops local solutions to problems that are also institutional in nature. In one course, the use of AI may be encouraged; in another, prohibited; in a third, tolerated without clear criteria. Students at the same university encounter different rules, which are not always justified by distinct pedagogical objectives.

See also:

• School culture fosters innovation based on human relationships

This is not to argue that all disciplines should adopt the same practices. It is about recognizing that pedagogical differences must result from conscious decisions, not merely from a lack of coordination.

Perception data help us understand the scope of the phenomenon. In a Digital Education Council survey on Latin American higher education, 79% of the faculty members surveyed reported already using AI in their teaching, and 94% expected to continue using it. The results indicate widespread adoption among participants, although they do not allow us to conclude that there is no resistance or concern regarding specific uses.

The same survey points to a disconnect between students' experiences and the institutional integration of technology: only 30% felt that the institution's use of AI met their expectations.

AI is therefore already integrated into the practices of a significant portion of the academic community, with expectations that are not always met and institutions that still need to develop their responses.

The question is no longer whether the university will decide what to do about AI. The university is already facing the consequences of not having made a sufficient decision.

The Job Nobody Named

There is a recognizable pattern to how teachers make these decisions. It is worth naming this pattern, because what goes unrecognized is unlikely to be accounted for, compensated, or incorporated into criteria for career advancement.

The teacher reviews what the computer has produced before presenting the material to the class. He or she anticipates errors, identifies oversimplifications, and decides what needs to be reworked. The teacher designs activities in which students must formulate problems, justify their choices, and demonstrate what they have learned. The teacher sets limits on the delegation of tasks and upholds those limits when they are challenged.

It translates general institutional guidelines into specific pedagogical decisions. And it revisits part of that work when the tools, conditions of use, or learning objectives change.

This isn't simply about learning how to use a new technology. It's about exercising professional judgment regarding the relationship between knowledge, activity, and learning.

We can call this activity “Pedagogical Orchestration of AI” (OPIA): the process of determining how people and artificial systems can collaborate without transferring to the machine the intellectual operations that students need to develop.

Orchestrating does not mean monitoring every response from a system, nor does it mean adding a layer of supervision to all activities. It means designing situations in which the use of AI serves a pedagogically justifiable purpose, determining what should remain under human responsibility, and creating conditions to assess what has actually been learned.

There is evidence that helps us understand the importance of this work. A study by Wang and Zhang, published in 2026 in the *International Journal of Educational Technology in Higher Education*, investigated the relationship between collaboration with AI systems, cognitive surveillance, and learning. The results highlight the importance of questioning, verifying, and critically evaluating the responses generated by these systems, rather than simply using them as tools.

The research does not show that a single approach works across all disciplines or that the teacher is the only possible source of guidance. But it underscores a crucial point: the use of AI, by itself, does not lead to deep learning. What matters is what people do cognitively with what the technology offers.

This approach cannot be mandated by a regulation. It must be developed through activities, guided by feedback, and supported by pedagogical decisions. And this work takes time.

If the university expects its faculty to redesign learning experiences, monitor outcomes, mentor students, and develop new forms of assessment, it must recognize these activities as part of academic work. The time required for these activities cannot continue to depend solely on the individual willingness of those who teach.

The problem isn't just that teachers work harder. It's that a significant portion of the work required to foster learning remains invisible in the systems that allocate time, resources, and professional recognition.

See also:

• “Courage to Transform”: Bett Brasil Sets Its Central Theme for 2027

The difference between adding and redesigning

I have been using a distinction to separate two movements that are often conflated in the debate on AI.

An AI-enabled university adds tools to an existing institutional framework. It purchases licenses, publishes guidelines, offers introductory training, and incorporates new resources into courses that continue to operate, for the most part, with the same structures for course load, assessment, and career advancement.

An AI-native university, on the other hand, intentionally redesigns its academic processes by treating AI as part of the cognitive infrastructure available for teaching, learning, conducting research, and producing knowledge.

These terms are ideal types, not rigid categories. A single institution may make significant progress in some programs while remaining stuck in outdated structures in others. What distinguishes the two movements is not the amount of technology acquired, but the depth of the institutional changes that accompany it.

An AI-native university does not organize its mission around technology. It organizes learning and the production of knowledge, viewing AI as an infrastructure that can expand capabilities but can also create dependencies, errors, and new inequalities.

The difference between adding and redesigning lies in administrative and pedagogical decisions that often seem secondary: how the workload is distributed, what counts toward career advancement, how technology purchases are evaluated, and what evidence is required to demonstrate learning.

It is also in this context that the risk of “digital Taylorism”—discussed by Silvio Meira and colleagues in 2025—manifests itself. The promise of efficiency can lead to the fragmentation of intellectual work into monitorable tasks, the automation of parts of the evaluation process, and the reduction of the teacher to an inspector of products and processes designed by third parties.

Technology is introduced to lighten the workload. Next, the organization redefines what it considers productive. Finally, the time saved can be converted into cost savings, without examining whether learning has improved. This is not an inevitable outcome. It is a management choice.

In Brazil, the discussion takes on added urgency due to the national guidelines for the use of AI in education, which were approved by the National Education Council in September and are still pending approval by the Ministry of Education. The guidelines emphasize the need for institutions to take responsibility for the use of this technology.

The existence of national guidelines is important. But no set of guidelines, on its own, can reshape the organization of academic work. This transformation depends on the decisions that institutions make and the conditions they provide to ensure those decisions are carried out.

What should I do now?

Four administrative changes can help turn diagnosis into action. None of them requires the acquisition of new technology.

1. Recognize pedagogical orchestration in the teaching load

The work assignment spreadsheet must account for the time spent redesigning courses, reviewing activities, developing criteria for AI use, and analyzing evidence of learning.

This can begin with hours formally allocated to pedagogical redesign projects, accompanied by defined objectives and deliverables. The important thing is for the institution to stop relying on unrecognized overtime to carry out a transformation that it itself considers strategic.

2. Include the redesign of learning in the criteria for career advancement

If the academic career track recognizes publications and other research outputs but does not adequately recognize the reconstruction of learning experiences, it creates a contradictory incentive.

The redesign of a course can involve sophisticated intellectual work, including theoretical grounding, experimentation, evaluation, and documentation of results. It should be recognized for its quality and demonstrable contributions, not merely for the number of activities that have been redesigned.

This is not about replacing research with pedagogical innovation. It is about recognizing that the quality of education also requires intellectual output and systematic work.

3. Assess the academic impact before purchasing technology

A technical procurement assessment cannot be limited to price, information security, and ease of integration. It must ask: What learning problem are we trying to solve? How does the tool change the work of teachers and students? What decisions does it automate? What new assessment tasks does it create? How will we know if learning has improved?

These questions are not obstacles to innovation. They are prerequisites for ensuring that the purchase of technology is an educational decision, not merely an infrastructure decision.

This assessment must also consider what the institution provides to the vendor when it adopts a technology. Who controls the data produced by faculty and students? What decisions are made by models that the university cannot examine? Is it possible to audit the results, migrate the data, and terminate the contract without compromising essential academic activities? AI governance does not end with information security: it includes preserving the institution’s ability to understand, question, and, when necessary, replace the systems on which it depends.

Finally, the institution should evaluate not only the cost of the leave, but also the implementation costs, training needs, the impact on academic work, and the expected outcomes.

4. Assess learning, not just the final product

We don't necessarily need to do away with exams, written assignments, or other academic work. We need to stop treating them as sufficient evidence in every situation.

The issue is also epistemological: what can we say about a student’s knowledge based on work produced with the assistance of AI? The answer depends on the purpose of the activity and the available evidence. In some cases, it is important to assess what the student can accomplish without assistance; in others, it is their ability to use AI critically, verify results, and justify decisions. The assessment must explicitly state what it aims to evaluate, rather than assuming that the final product, on its own, answers that question.

Monitored portfolios, oral presentations, successive drafts of a project, justifications for decisions, and reflective discussions can reveal aspects of the process that the final product does not show. The choice of assessment tools should depend on the competencies to be developed and the actual conditions of each course.

This requires planning, time, and clear guidelines. It also requires that students understand what is expected of them, including when AI is permitted, recommended, or required.

The goal is not to make assessment more difficult. It is to make it better able to demonstrate what students can do, what they understand, and what decisions they can justify.

---

Listening to faculty and students is essential for understanding the problems and assessing the consequences of the choices made. But participation does not mean shifting all responsibility for decisions onto them. It is up to the institution to establish criteria, clarify priorities, provide the conditions for implementation, and be accountable for the results. Pedagogical autonomy must be supported by a governance structure that makes it possible, rather than being used as an excuse to leave each faculty member to solve institutional problems on their own.

The four changes share a common principle: institutional accountability must be accompanied by concrete conditions for implementation and evidence that allows for the evaluation of their results.

This is the direction we are taking at CESAR School, through an initiative organized around seven interdependent pillars. The results of this effort will become apparent in the coming months.

A denial, to wrap things up

Of all the pitfalls in this process, the one that worries me the most is the easiest to disguise: using “efficiency” as a euphemism for cost-cutting.

Efficiency is important. Reducing unnecessary tasks, expanding access to knowledge, and freeing up time for intellectually meaningful activities are legitimate goals. The problem arises when resource conservation becomes the dominant criterion, without examining what happens to learning.

It is possible to reduce the time spent on an activity and improve its results. It is also possible to reduce costs by shifting work to students, automating decisions that would otherwise require judgment, or reducing pedagogical oversight. Both approaches may be presented under the same banner of innovation.

It saves teachers time, reduces the cost per student, and increases scale. But if the institution fails to verify what its students have actually learned, efficiency comes to be measured by the ability to award diplomas, not by the quality of the education they represent.

Therefore, the key question is not whether AI will allow us to do more with less. It is what we want to do better, for whom, and under what learning conditions.

The adoption of AI in Latin American higher education has advanced largely through the initiative of professors, researchers, and students, while institutional strategies have lagged behind. This does not mean that every individual decision has been the right one, nor that faculty autonomy eliminates the need for coordination. It means that the transformation is already underway and that institutions need to play their part in guiding this process.

When they finally do, they'll have a choice.

They may recognize pedagogical judgment as a central part of academic work, create conditions to redesign learning, and evaluate the results rigorously. Or they may use technology to standardize activities, cut costs, and strip teachers of the discretion that has made adaptation possible thus far.

The first option requires investment, coordination, and a willingness to overhaul existing structures. The second may seem faster and simpler.

But a university does not demonstrate its capacity for innovation by the number of tasks it automates. It demonstrates it through the quality of the knowledge it produces and, in particular, through the relevance of its degrees in the job market it serves.

The teacher made the decision first. Now, the decision must be made at the institutional level—and learning must continue to be the guiding principle.

About the author:

Article originally published HERE. 

Share on social media:

Sources and references:

• UNESCO IESALC and UNU-IAS. AI Implementation in Higher Education in Latin America and the Caribbean (2026). Survey of 200 institutions in 19 countries. iesalc.unesco.org
• Digital Education Council. AI in Higher Education — LATAM Survey 2026. Findings on the use of AI by faculty and students’ perceptions.
• Wang and Zhang (2026). A Study on Human-AI Partnership, Cognitive Monitoring, and Deep Learning. International Journal of Educational Technology in Higher Education.
• National Education Council. Opinion CNE/CP No. 15/2026 on guidelines for the use of artificial intelligence in Brazilian education.
• Meira, S., Neves, A., and Braga, R. (2025). Digital Taylorism. TDS Company.