Gaps in AI education: analysis of
the AI Index 2025 and its implications for Latin America
Jorge Iván Pincay-Ponce
Universidad Laica Eloy Alfaro de Manabí, ULEAM
Manta, Manabí, Ecuador
Ana María Ponce Tomalá
Unidad Educativa Fiscal César Borja Lavayen
anam.ponce@docentes.educacion.edu.ec
Guayaquil, Guayas, Ecuador
Ingrid Edith Narea Velasco
Unidad Educativa Fiscal Vicente
Rocafuerte
ingrid.narea@docentes.educacion.edu.ec
Guayaquil, Guayas, Ecuador
Luisa Jessenia Guerrero Zambrano
Unidad Educativa Quevedo
luisaj.guerrero@docentes.educacion.edu.ec
Quevedo, Los Ríos, Ecuador
Lídice Mónica Franco-Franco
Ministerio de Educación
lidice.franco@docentes.educacion.edu.ec
Guayaquil, Guayas, Ecuador
DOI: https://doi.org/10.56124/encriptar.v9i17.012
ABSTRACT
This article analyzes the gaps in Artificial
Intelligence education in Latin America based on Stanford's AI Index Report
2025. Using documentary methodology and data triangulation with ECLAC and the
IDB, it was identified that the region contributes only 1.66% of global
research in Artificial Intelligence. The results show a contradiction between
the high demand for teacher training (81%) and the low capacity for technical
implementation. It is concluded that the educational gap is the main obstacle
to regional technological sovereignty, relegating Latin American countries to
algorithmic dependence if curricula, especially at the university level, are
not urgently reformed.
Keywords: Artificial Intelligence,
Educational Gap, Latin America, Technological Sovereignty, AI Index 2025.
Brechas en educación en IA:
análisis del AI Index 2025 y sus implicaciones para América Latina
Resumen
Este artículo analiza las brechas en educación
en Inteligencia Artificial en América Latina basándose en el AI Index Report
2025 de Stanford. Mediante una metodología documental y triangulación de datos
con CEPAL y el BID, se identificó que la región solo contribuye con el 1.66% de
la investigación mundial en Inteligencia Artificial. Los resultados muestran
una contradicción entre la alta demanda de formación docente (81%) y la baja
capacidad de ejecución técnica. Se concluye que la brecha educativa es el
principal obstáculo para la soberanía tecnológica regional, relegando a los
países latinos a una dependencia algorítmica si no se reforman urgentemente los
currículos, especialmente a nivel universitario.
Palabras clave: Inteligencia Artificial,
Brecha Educativa, América Latina, Soberanía Tecnológica, AI Index 2025.
1. Introduction
In the last decade, Artificial Intelligence (AI)
has gone from being a peripheral discipline to becoming the driving force
behind the Fourth Industrial Revolution. However, the AI Index Report 2025,
led by Stanford University, reveals a profound asymmetrical geography of
knowledge. While North America and East Asia are consolidating their hegemony
through massive investment and accelerated curricular integration, Latin
America faces the risk of a new form of exclusion: algorithmic dependence.
This problem is reflected in compelling data:
there is an 81% global consensus on the importance of AI in basic education,
but less than 50% of teachers possess the skills to teach it (Maslej et al., 2025). In the Latin
American context, this gap is exacerbated by structural infrastructure
deficiencies and an R&D investment that barely reaches 1.66% of global
output (BID, 2023; CEPAL, 2024). Although the region
has shown progress in training ICT graduates, the disconnect between
pedagogical intent and technical capacity is alarming.
To contextualize the magnitude of this
phenomenon, it is imperative to clarify two fundamental constructions. First,
Artificial Intelligence Literacy, defined as the ability to interact with
automated systems along with critical skills to evaluate their underlying logic
and ethical implications (Ng et al., 2024). Second,
Technological Sovereignty, referring to a state's capacity to develop, control,
and audit its own digital infrastructures without depending on external
hegemonic providers (Sheriffdeen Folaranmi Abiade, 2025).
Within this framework, recent studies published
by Springer emphasize that AI integration should prioritize the creation of
local language models to avoid cultural biases. However, the regional reality
reveals a persistent gap between pedagogical ambition and tangible
infrastructure (Mazzucato, 2025), which positions
algorithmic literacy as the essential foundation of technological sovereignty.
The objective of this research is to support,
through documentary analysis and data triangulation with organizations such as
ECLAC and the IDB, that these educational gaps not only limit economic
competitiveness, but also compromise the ethical and political autonomy of the
region in the face of AI models that do not reflect its social reality.
2. Materials and
methods
2.1. Research design and scope
This research employs a quantitative-analytical
approach at a descriptive level, with an ex post facto documentary design. This
design allows for the study of phenomena that have already occurred, using
records and secondary data to establish causal or correlational relationships (González Mares, 2019). The scope is
correlational, seeking to associate educational gap indicators with
technological sovereignty capacity in Latin America.
2.2. Units of analysis and documentary corpus
The Stanford University Artificial Intelligence
Index Report 2025 was used as the primary data source (Maslej et al., 2025). The procedure
consisted of the systematic extraction of quantitative indicators on education
and development, specifically selecting:
•
Chapter 1 (R&D): Volume of scientific
publications by region.
•
Chapter 6 (Policy and Governance): Incidents and
governance frameworks.
•
Chapter 7 (Education): K-12 curriculum adoption
rates and teacher preparation.
2.3. Data triangulation procedure
The data referred to in the preceding section
were subjected to a triangulation process with regional literature and
literature from global entities that analyze Latin America in their reports, in order to infer the specific implications in the Latin
American context. Triangulation with sources complementary to the AI Index
Report 2025 (Stanford University) addressed the inclusion of metrics on
productivity and the digital divide from ECLAC and the IDB (2023-2024),
allowing for a weighted comparison between global trends and the specific
infrastructural limitations of the Andean region (BID, 2023; CEPAL, 2024; Institute for Statistics, UNESCO, 2023; UNESCO,
2024).
2.4. Data collection techniques and
instruments
The quantitative content analysis technique was
used. As an instrument, an Indicator Extraction and Categorization Matrix
(MECI) were designed in digital format, which allowed the normalization of the
figures from the Stanford report and the projections of the regional organizations
to facilitate their statistical comparison.
2.4.1. Data collection techniques and
instruments
The research employed quantitative documentary
content analysis, applied to: (1) AI Index Report 2025, (2) Global Education
Monitoring Report 2024, (3) Education at a Glance
2024, and (4) Digital Transformation Reports 2023–2024. The objective was to
standardize heterogeneous metrics in different units, scales, and definitions…
to allow for regional comparison.
2.4.2. Operational Definition of the MECI
The Indicator Extraction and Categorization
Matrix (MECI) was designed as a spreadsheet structured in three levels:
·
Level 1 – Identification:
Source, Year, Region/Country, Page/Table of the report.
·
Level 2 – Original Variable:
Textual name of the indicator, original definition, unit of measurement, and
reported value.
·
Level 3 – Normalization:
Analytical category, subcategory, applied transformation, standardized value,
and methodological observations (See Table 1).
Table 1: Analytical
Categories Defined in the MECI
|
Main Category |
Subcategory |
Type |
|
AI Human Capital |
AI/Computer Science Graduates |
Structural |
|
Educational Infrastructure |
Formal AI Programs in Universities |
Institutional |
|
Investment |
STEM Expenditure (% of GDP) |
Financial |
|
Inclusion |
Percentage of Women in AI Programs |
Social |
|
Digital Capacity |
University Digital Connectivity |
Contextual |
Table 2: Presents an extract
of the matrix design. CS bachelor’s Degrees Awarded
|
Field |
Description |
|
Source |
AI Index 2025 |
|
Year |
2024 |
|
Region |
United States |
|
Original Indicator |
CS bachelor’s
Degrees Awarded |
|
Original Definition |
Total number of
Computer Science degrees awarded |
|
Unit |
Absolute |
|
Raw Value |
97,000 |
|
Analytical Category |
AI Human Capital |
|
Transformation
Applied |
Divided by total
population × 1,000,000 |
|
Standardized Value |
85 per million |
Table 3: Presents an extract
of the matrix design. CS Degrees (regional estimate)
|
Field |
Description |
|
Source |
AI Index 2025 |
|
Year |
2024 |
|
Region |
Latin America |
|
Original Indicator |
CS Degrees
(regional estimate) |
|
Original Definition |
Total CS graduates
in the region |
|
Unit |
Absolute |
|
Raw Value |
120,000 |
|
Analytical Category |
AI Human Capital |
|
Transformation
Applied |
Divided by regional
population × 1,000,000 |
|
Standardized Value |
18 per million |
Table 4: Presents an extract
of the matrix design. Expenditure on STEM Education
|
Field |
Description |
|
Source |
UNESCO GEM 2024 |
|
Year |
2023 |
|
Region |
Latin America |
|
Original Indicator |
Expenditure on STEM
Education |
|
Original Definition |
Percentage of GDP
allocated to STEM education |
|
Unit |
% of GDP |
|
Raw Value |
0.8 |
|
Analytical Category |
Investment |
|
Transformation
Applied |
No transformation
required |
|
Standardized Value |
0.8 |
Table 5: Presents an extract
of the matrix design. CS Degrees (regional estimate)
|
Field |
Description |
|
Source |
OECD 2024 |
|
Year |
2023 |
|
Region |
European Union |
|
Original Indicator |
Tertiary AI
Programs |
|
Original Definition |
Percentage of
universities offering formal AI programs |
|
Unit |
Percentage |
|
Raw Value |
65 |
|
Analytical Category |
Educational
Infrastructure |
|
Transformation
Applied |
Nominal
harmonization |
|
Standardized Value |
65 |
2.5. Ethical and validity considerations
As this research is based on publicly available
secondary data, the study complies with intellectual property regulations
through rigorous citation under APA 7th edition guidelines. Data reliability is
ensured by the data audit methodology of the Stanford Human-Centered AI
Institute.
3. Results and Discussion
In accordance with the analytical documentary
methodology employed, the results are presented organized around the three
dimensions of the AI gap identified in the AI Index Report 2025 and their
contrast with the Latin American ecosystem.
3.1. Curricular
dimension and teacher readiness: The implementation gap
The analysis of educational data revealed a
clear structural paradox: globally, 81% of computer science educators recognize
the imperative need to integrate AI into the core curriculum, but the capacity
to implement it is limited. Furthermore:
- Less than 50% of
teachers surveyed in the report's primary sources declared themselves
competent to teach AI.
- When this data is
triangulated with IDB infrastructure metrics, it is observed that in Latin
America this gap is both cognitive and technical, given that
"pedagogical will" clashes with a reality where access to
high-performance hardware for model processing is almost nonexistent in
public classrooms, placing the region in a phase of "passive
literacy".
3.2. Scientific
production and technological sovereignty
The disparity in knowledge production shown in
Figure 1 places Latin America with only 1.66% of the world's AI research
output.
Figure 1: It highlights
the marginalization of Latin America in relation to the poles of technological
power.
The 1.66% contrasts sharply with the 34.5% from
East Asia and 28.1% from North America, effectively marginalizing the region in
terms of research. Consequently, this relegates Latin America to a state of
algorithmic dependency due to the lack of proprietary foundational model
production. This scenario forces Latin American educational institutions to
import technologies that carry foreign cultural and linguistic biases, further
exacerbating the technological sovereignty gap.
3.3. The capabilities radar
Upon observing the capabilities radar (see Figure
2), which integrates data from UNESCO and the AI Index 2025, a significant
turning point is highlighted. This trend is driven by countries such as Brazil
and Mexico, which exhibit a competitive score in the generation of Information
Technology (IT) Graduates exceeding 60%. This figure even surpasses the global
average in terms of gross volume within certain subsectors.
Figure 2: Based on a
normalized 0-100 scale derived from infrastructure, teacher training, and
publication data from the AI Index 2025, a significant structural imbalance is
observed.
Figure 2 presents a comparative radar visualization
of structural AI education gaps across regions. Each axis corresponds to a
normalized indicator extracted through the Indicator Extraction and
Categorization Matrix (MECI). Raw values were obtained from the AI Index Report
2025 (Stanford HAI), UNESCO GEM 2024, OECD Education at a
Glance 2024, and CEPAL Digital Transformation Reports (2023–2024). All
indicators were normalized using Min–Max scaling to enable cross-regional
comparability.
All radar dimensions were equally weighted (w = 1)
to avoid normative bias. The visualization serves exploratory comparative
purposes rather than inferential modeling. The following tables (6-10) explain
the preparation of each dimensión, and, although not
all are used in the radar chart, they are used to support claims in this
article.
Table 6: AI Graduates per Million
|
Field |
Description |
|
Radar
Axis |
AI
Graduates per Million |
|
Primary
Source |
AI
Index Report 2025 (Stanford HAI) |
|
Year |
2024 |
|
Raw
Value – Latin America |
18 |
|
Highest
Regional Value |
120
(Asia) |
|
Lowest
Regional Value |
18 |
|
Normalization
Formula |
(X −
Xmin) / (Xmax − Xmin) |
|
Normalized
Value – Latin America |
0.00 |
|
Weight
Assigned |
1
(equal weighting) |
Table 7: STEM Investment
|
Field |
Description |
|
Radar
Axis |
STEM
Investment (% of GDP) |
|
Primary
Source |
UNESCO
GEM 2024 / OECD Education at a Glance 2024 |
|
Year |
2023 |
|
Raw
Value – Latin America |
0.8 |
|
Highest
Regional Value |
3.1
(United States) |
|
Lowest
Regional Value |
0.8 |
|
Normalization
Formula |
(X −
Xmin) / (Xmax − Xmin) |
|
Normalized
Value – Latin America |
0.00 |
|
Weight
Assigned |
1
(equal weighting) |
Table 8: Universities Offering AI Programs
|
Field |
Description |
|
Radar
Axis |
Universities
Offering AI Programs (%) |
|
Primary
Source |
OECD
2024 / AI Index 2025 |
|
Year |
2023 |
|
Raw
Value – Latin America |
32 |
|
Highest
Regional Value |
82
(Asia) |
|
Lowest
Regional Value |
32 |
|
Normalization
Formula |
(X −
Xmin) / (Xmax − Xmin) |
|
Normalized
Value – Latin America |
0.00 |
|
Weight
Assigned |
1
(equal weighting) |
Table 9: Female Participation in AI Programs
|
Field |
Description |
|
Radar
Axis |
Female
Participation in AI Programs (%) |
|
Primary
Source |
AI
Index 2025 / UNESCO |
|
Year |
2024 |
|
Raw
Value – Latin America |
22 |
|
Highest
Regional Value |
30
(Europe) |
|
Lowest
Regional Value |
22 |
|
Normalization
Formula |
(X −
Xmin) / (Xmax − Xmin) |
|
Normalized
Value – Latin America |
0.25 |
|
Weight
Assigned |
1
(equal weighting) |
Table 10: Digital Infrastructure Index (Composite Proxy)
|
Field |
Description |
|
Radar
Axis |
Digital
Infrastructure Index (Composite Proxy) |
|
Primary
Source |
CEPAL
Digital Transformation Report 2024 |
|
Year |
2024 |
|
Raw
Value – Latin America |
0.55 |
|
Highest
Regional Value |
0.92
(OECD Average) |
|
Lowest
Regional Value |
0.55 |
|
Normalization
Formula |
(X −
Xmin) / (Xmax − Xmin) |
|
Normalized
Value – Latin America |
0.12 |
|
Weight
Assigned |
1
(equal weighting) |
However, the apparent abundance of graduates does
not translate into local innovation, revealing a disconnection that can be
termed the "Talent Paradox" specifically, the gap between academic
training and R&D investment (15% vs. 70% globally). This suggests that
Latin America is cultivating talent to sustain third-party infrastructures,
primarily at the service and support levels, while remaining far from leading
the development of intellectual property. This further confirms a peripheral insertion
into the AI economy.
Furthermore, while the United States produced 40
notable AI models in 2024, the region remains predominantly a consumer of
external technology.
3.4. Feasibility Analysis and Emerging Risks
In Chapter 6, regarding Policy and Governance,
the AI Index Report points out that most global organizations still lack robust
Responsible Artificial Intelligence protocols, and only a very small fraction
actively audits for bias. Triangulation with Latin America reveals that if
major global corporations are failing, Latin American SMEs, the region’s
economic engine, are entirely blind. Given the absence of algorithmic auditing
training in Latin American universities, what remains is the integration of AI
into citizen services without defense mechanisms against unfair or
discriminatory automated decisions.
The AI Index reports that PhD graduates in AI in
leading countries have shifted nearly 70% toward the private industry, leaving
academia behind. Triangulation with Latin America reveals the opposite: due to
a lack of local industry, the scarce high-level talent either remains in
underfunded academia or emigrates. This creates a vacuum where there are no
mentors within Latin American companies to guide ethical implementation,
leaving undergraduate graduates without senior technical guidance in the region’s
labor market.
Another significant finding is that while
companies in the U.S. adopt AI to reduce operational costs, in Latin America,
the cost of importing that technology and maintaining cloud infrastructure
ultimately translates into a "technological tax." This once again
calls for a shift toward "model optimization" rather than mere usage,
ensuring that local solutions become economically viable.
3.5. Future Works
The challenges facing the region do not differ
significantly from those of Australia, Africa, or certain European countries.
As Latin America remains a region dependent on AI technologies developed
primarily in the United States and Asia, particularly China (Pincay Ponce et al., 2025), the most sensible
immediate future work is to investigate how the massive adoption of foreign
foundational models from companies such as OpenAI, Google, or Anthropic within
the Latin American educational system is displacing local teaching methodologies.
This future research should evaluate whether we are educating a generation to
be "prompt operators" dependent on the Global North, rather than
developers of sovereign solutions, while analyzing the loss of cultural
identity in AI-generated academic outcomes.
A further research path derived from the current
study would be an "audit of teacher obsolescence" in the face of
inaction by most Latin American states (Filgueiras, 2025). This would involve
a longitudinal study on the rate of skill degradation among STEM faculty, a
field more critical than others, against the velocity of Generative AI
evolution (UNESCO, 2024). This work would
serve as a benchmark for the gap between the political discourse of
"innovation" and the actual lack of minimum infrastructure, such as
servers and GPUs, in public universities, determining whether technical higher
education in the region is teaching technologies that are already obsolete
before student’s graduate.
As a final future proposal, it is suggested to
delve deeper into what this article has presented as the "mass-graduates
paradox." Research here would focus on evaluating whether current training
in basic programming and technical support in Latin America is creating a
future population of unemployed individuals, given that AI currently performs
many entry-level tasks across numerous fields of study. The eventual study must
propose a radical redesign of the curriculum toward "High-Value Human
Capital," which at a minimum addresses advanced technical AI, systems
architecture, and deep ethics; collectively, this would suggest changes in
content and forms distinct from current ones.
4. Conclusions
In
this descriptive, quantitative-analytical research, a structural gap was
evidenced, confirming that Latin America suffers from a "second order"
AI education gap. This is due to a lack of access to technology and a
deficiency in competencies for developing proprietary models, positioning the
region as a "consumer periphery."
Furthermore,
the findings reveal what this article terms the "Human Capital
Paradox." This refers to a critical disconnection between the volume of
graduates in technological fields and their capacity for impact on Research and
Development (R&D). The region cultivates human talent that ultimately
operates or supports foreign technologies due to the absence of advanced
technical training in local curricula.
The
authors contend that we are facing "ethical vulnerability by
omission." This is partly due to the scarce training in algorithmic ethics
and bias mitigation within Latin American educational programs, which generates
a dependency on governance frameworks designed in AI-model-constructing
regions. Consequently, it is expected that these models do not always account
for the cultural and social particularities of Latin America.
As
a final reflection, it must be noted that if technical training does not
transition from "consumer literacy" to the capacity for creation and
auditing of Artificial Intelligence models, the region will not only lose
economic competitiveness but will remain subordinated to an ethical and
algorithmic infrastructure designed for foreign realities. The development of
an AI with a "Latin identity" is now a matter of urgency.
5. References
BID. (2023). La
infraestructura digital en el desarrollo de América Latina.
CEPAL. (2024). Transformación
digital para el desarrollo inclusivo en ALC.
Filgueiras, F.
(2025). Artificial intelligence and governance challenges in Latin America–the
game between decolonization and dependence. In Handbook on Governance and
Data Science (pp. 198–221). Edward Elgar Publishing.
González Mares, M.
(2019). Hernández-Sampieri, R. & Mendoza, C (2018). Metodología de la
investigación. Las rutas cuantitativa, cualitativa y mixta. Revista
Universitaria Digital de Ciencias Sociales (RUDICS), 10(18), 92–95.
https://doi.org/10.22201/fesc.20072236e.2019.10.18.6
Institute for
Statistics, UNESCO. (2023). Women in Science: Interactive Data Tool.
Maslej, N.,
Fattorini, L., Perrault, R., Gil, Y., Parli, V., Kariuki, N., Capstick, E.,
Reuel, A., Brynjolfsson, E., Etchemendy, J., Ligett, K., Lyons, T., Manyika,
J., Niebles, J. C., Shoham, Y., Wald, R., Walsh, T., Hamrah, A., Santarlasci,
L., … Oak, S. (2025). Artificial Intelligence Index Report 2025 (Version
3). arXiv. https://doi.org/10.48550/ARXIV.2504.07139
Mazzucato, M. (2025).
BIG CON: How the consulting industry weakens our businesses, infantilizes
our governments, and... warps our economies. PENGUIN BOOKS.
Ng, D. T. K., Wu, W.,
Leung, J. K. L., Chiu, T. K. F., & Chu, S. K. W. (2024). Design and
validation of the literacy questionnaire: The affective, behavioural, cognitive
and ethical approach. British Journal of Educational Technology, 55(3),
1082–1104. https://doi.org/10.1111/bjet.13411
Pincay Ponce, J. I.,
Zambrano Zambrano, M. T., & Quijije Anchundia, P. J. (2025). Hacia una educación
personalizada y equitativa: Análisis de la convergencia entre tecnologías
asistivas e inteligencia artificial. Revista Científica Multidisciplinaria
SAPIENTIAE, 8(16), 536–545.
https://doi.org/10.56124/sapientiae.v8i16.034
Sheriffdeen Folaranmi
Abiade. (2025). Algorithmic Sovereignty and the New Security Dependencies: How
Foreign AI Surveillance Technologies Reshape Domestic Autonomy in the Global
Sout. World Journal of Advanced Research and Reviews, 27(2),
162–180. https://doi.org/10.30574/wjarr.2025.27.2.2845
UNESCO. (2024). Factores
determinantes en la educación STEM en la región.