By Javier Surasky
.png)
Two Texts, One Problem, Several Dimensions, and Many Agreements
In August 2026, two works were published on the
relationship between artificial intelligence and education: the Report of
MIT’s Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training,
and Manual urgente sobre IA en el aula. Miradas y análisis de la esfera
digital desde Abya Yala, by Mariana Ferrarelli and Natalia Corvalán,
published by the Center for the Study of Digital Society at EIDAES-UNSAM.
The texts were written for different contexts: the MIT
report responds to the needs of a research-intensive university and devotes
considerable attention to its residential experience, while Ferrarelli and
Corvalán’s work is addressed primarily to secondary school teachers and
situates its analysis in Abya Yala, exposing the inequalities of the region’s
educational systems and the material conditions under which digital
technologies are expanding.
Those different points of departure explain much of
the emphasis that follows.
There is one common point that organizes much of both
texts: the idea that the use of AI must be justified by educational purposes.
This shifts the discussion toward the learning an activity is meant to produce,
rather than toward how to operate a model more effectively or how to prompt it
correctly.
Returning to the Objectives of Education
The MIT Committee reviews practices that have served
for years to teach and assess, from problem sets and written assignments to
take-home exams, all of which can now be completed with the assistance of
generative systems.
One possible reaction, they tell us, is to move more
assessments into the classroom. But the report itself points out the limits of
that solution: in-person tasks constrained by short time frames may be poorly
suited to forms of learning that require extended elaboration. This leads the
Committee to favor oral exams and work developed outside the classroom that
must then be discussed in person (Massachusetts Institute of Technology [MIT],
2026, pp. 11–13).
The report turns to backward design to organize these
decisions: learning outcomes are determined first, and only then are activities
and forms of assessment designed, with AI policy incorporated into that
process, on the premise that “intentional teaching promotes intentional
learning” (MIT, 2026, p. 6).
The authorization or restriction of a tool should
therefore be explainable in terms of what students are expected to learn.
The Manual urgente reaches a similar position
through the notion of the sociotechnical assemblage, arguing that AI
applications are part of a structure involving models and data, infrastructure,
human decisions, economic interests, and regulation. This broadens what counts
as a pedagogical decision and leads Ferrarelli and Corvalán to maintain that
the use of AI requires “decisions anchored in the teacher’s educational
purposes and not in the technical possibilities of the tool” (2026, p. 10).
Once again, the question of use appears only after the educational problem one
is trying to address has been defined (Ferrarelli & Corvalán, 2026, p. 10).
The two formulations come from different conceptual
frameworks: while MIT relies on the design of learning experiences, the Manual
urgente also examines the technological mediations that intervene in them.
Those different starting points nevertheless lead to a similar and practical
conclusion: the mere availability of a technological function has no
pedagogical value in itself.
This position has direct implications for the rules
governing the use of AI in education. MIT rejects a uniform policy for all
subjects since, for example, an introductory course may need to protect certain
practices while students are acquiring basic knowledge, whereas in more
advanced stages of training, the use of AI tools may be part of the expected
learning. The MIT Committee proposes that instructors and departments make
these decisions within a common institutional framework and explain their reasons
in relation to the objectives of the course (MIT, 2026, pp. 7, 16–17).
The Manual urgente faces a similar problem, but
on a different scale: the authors argue that “the incorporation of these
technologies cannot be left to individual decisions or to the logic of trial
and error” (Ferrarelli & Corvalán, 2026, p. 44). Situated criteria require
institutional reference points capable of guiding the decisions of each
educational community.
What Is Delegated When We Automate
Both texts also pause over the tasks that are
delegated.
MIT refers to the principle of augmentation, not
automation, to examine what happens when a system performs part of the work
that previously belonged to the student, affecting the learning process in such
a way that the student may arrive at a correct answer without developing the
capacities they were supposed to acquire in order to solve it. In response to
this, the report is clear in stating that there is such a thing as a misuse of
AI, which “should be used to augment and enhance curiosity, creativity, and
learning, not automate them” (MIT, 2026, p. 7).
Moreover, during the Committee’s consultations, signs
emerged of changes in academic sociability: variations in attendance at office
hours and participation in discussions, as well as a decline in in-person study
groups. Although the latter are presented as “anecdotal,” they seem relevant to
me, all the more so if we consider that the document is still working with
partial evidence and with recent phenomena (MIT, 2026, pp. 3, 9).
The Undergraduate Research Opportunities Program
offers a concrete case: some faculty members were considering using AI agents
for tasks that had previously been assigned to students involved in research
projects. This is an understandable choice from the perspective of immediate
productivity, but unacceptable from a formative view of students’ capacities,
since it closes off their possibility of gaining experience through
participation in research work and contact with more experienced researchers
(MIT, 2026, pp. 14–15).
This clash of priorities leads MIT to speak of the
need for an educational social contract, defined as “a shared understanding
among students, instructors, and the Institute about the purposes of education”
(MIT, 2026, p. 8). Within that contract, rules for the use of AI are integrated
insofar as they are part of preexisting relationships among those who teach,
those who learn, and the institution responsible for organizing that experience
(MIT, 2026, p. 8).
Ferrarelli and Corvalán, by contrast, recover the
paradigm of care in order to examine related questions: the Manual urgente
situates this perspective within pedagogical currents and educational policies
that predate the expansion of AI, and from there they analyze discourses that
value technologies for the automation they enable or for the time savings they
promise. They also show how educational assessment incorporates considerations
related to learning conditions and to the responsibilities of institutions
toward those who participate in them (Ferrarelli & Corvalán, 2026, pp.
33–34).
This framing explains the manual’s distinction between
teaching with AI and teaching for AI. The first refers to learning situations
in which these technologies perform some pedagogical function; the second
points to the formation of criteria for understanding the algorithmic systems
that intervene in educational and social contexts. The distinction establishes
that knowledge of a tool is one part of literacy, but that it must necessarily
be accompanied by an understanding of the decisions embedded in its design and
of the effects its use may produce (Ferrarelli & Corvalán, 2026, pp.
36–37).
Materiality Changes the Problem
Materiality occupies different places in the two
works: the MIT report includes privacy, intellectual property, inequalities of
access, and corporate concentration among the issues an institutional policy
should consider, and also devotes attention to environmental impact (MIT, 2026,
p. 27).
In the Manual urgente, these elements are part
of the sociotechnical definition adopted from the outset, and an entire chapter
studies the material infrastructure of AI, its environmental impacts, and the
human labor present in chains that often appear to users as automated processes.
This integrated analysis builds on previous research on the materiality of AI
and on the human labor that sustains its operation (Ferrarelli & Corvalán,
2026, pp. 26–29).
The regional perspective also incorporates the
geographic distribution of technological capacities. To do so, the authors draw
on the Atlas de inteligencia artificial para el desarrollo humano en América
Latina y el Caribe, coordinated by Gustavo Beliz, which points to the
concentration of data centers and computing capacity outside the region and
connects that distribution to the margins available for developing
technological policies of one’s own (Beliz, 2025, as cited in Ferrarelli &
Corvalán, 2026, p. 34).
The educational incorporation of AI thus takes on a
specific meaning, since schools in the region operate under very different
conditions in terms of infrastructure and resource availability. For that
reason, the adoption of AI is analyzed together with the institutional
capacities needed to sustain it and subject it to educational criteria
(Ferrarelli & Corvalán, 2026, pp. 33–34).
At MIT, these elements form part of the
responsibilities that accompany an institutional policy; in the UNSAM text, by
contrast, they enter into the explanation of how these technologies are
constituted.
That distinction affects the field of decision-making:
an institution can define rules of use for teachers and students, but the real
conditions of that use depend on providers and infrastructures over which the
institution has almost no control.
From Teaching Criteria to Institutional Capacity
The MIT report proposes reviewing assessments,
reconsidering some grading systems, increasing certain in-person activities,
and establishing explicit policies for each subject. These measures would be
accompanied by structures responsible for gathering evidence and revising the
rules adopted as technologies change and their effects become better understood
(MIT, 2026, pp. 12–18, 28). This “continuous review” occupies an important
place in the report, and the Committee assumes that the decisions adopted in
2026 will probably have to be modified. This leads it to recommend procedures
capable of producing evidence and correcting policies as their results become
clearer (MIT, 2026, p. 28).
Ferrarelli and Corvalán find a similar logic in some
Argentine regulatory processes, such as the creation of the Artificial
Intelligence Observatory of the Federal Council of Education, where they speak
of “progressive regulation” as a process of producing evidence and developing
criteria that accompanies a field that remains unstable (2026, p. 44).
Both works assume that educational institutions are
already making decisions about AI, even when they lack explicit guiding
policies: the MIT report responds through a program of experimentation and
revision, while Ferrarelli and Corvalán’s work insists on keeping open the
discussion about the conditions under which the technology is incorporated.
This leads them to write one of my favorite sentences in their work: “in a
scenario where technologies are often presented as inevitable, recovering the
question becomes a countercultural gesture” (2026, p. 47).
Conclusion
MIT observes a university that needs to reorganize
educational practices in response to generative systems already integrated into
academic life. The Manual urgente situates those decisions within a
technological structure whose infrastructure, resources, and capacities are
distributed unequally.
Taken together, the two works pose a question that
goes beyond the authorization or prohibition of tools, and that I find much
more interesting than either extreme: What real margin do educational
institutions still have to define the conditions under which they accept the
intervention of these technologies in teaching?
With more similarities than differences, despite their
different realities and starting points, both reports invite us to complicate
simple views, to think about the double functions of AI in education, and to
move beyond “technology in the classroom” in order to place both within
situated, unequal, and complex societies. Too often, those societies seem to
expect schools to act as islands immune to social conflict. And that, I would
say, might perhaps be the greatest tragedy that could happen to us.
References
Ferrarelli,
M., & Corvalán, N. (2026). Manual urgente sobre IA en el aula: Miradas y
análisis de la esfera digital desde Abya Yala. Colección Miradas del CESDi.
Massachusetts Institute of Technology. (2026, August
13). Report of MIT’s Ad Hoc Committee on AI Use in Teaching, Learning, and
Research Training.