Angelina Espejel Trujillo

Es especialista en Inteligencia Artificial, con una trayectoria enfocada en el análisis crítico
de las tecnologías contemporáneas. Es Doctora en Comunicaciones y Electrónica por el
Instituto Politécnico Nacional y realizó una estancia postdoctoral en la Universidad de
Electro-Comunicaciones en Tokio, Japón.

Actualmente, se desempeña como investigadora de la SECIHTI en el Centro de Investigación en Ciencias de Información Geoespacial CentroGeo, donde lidera proyectos que integran Ciencia de Datos e Inteligencia Artificial para abordar problemáticas sociales.

Ha colaborado con instituciones públicas como el Hospital del Niño y la Mujer y el Instituto Queretano de la Mujer, además es cocreadora del Observatorio “Siempre Seguras”.
Su trabajo se distingue por el énfasis en la ética de la Inteligencia Artificial, particularmente en la detección de sesgos y la mitigación de riesgos algorítmicos, así como en el análisis de sus implicaciones sociales.

En el ámbito metodológico, ha empleado técnicas de Procesamiento de Lenguaje Natural y Análisis de Sentimiento para el estudio de la percepción social y el monitoreo de riesgos en entornos urbanos.


Sesión

11/11
12:45
15minutos
Algorithmic Bias in GeoAI: Information Poverty and Territorial Invisibility
Angelina Espejel Trujillo

The expansion of Geospatial Artificial Intelligence (GeoAI) is reshaping urban planning, territorial management, and public decision-making due to its high efficiency in the intensive processing of large volumes of data and automated responses. However, it is often assumed that these technologies offer neutral and objective solutions, yet recent literature shows that they incorporate training biases, such that GeoAI systems assimilate, reproduce, and amplify historical inequalities in territorial representation, leading to phenomena of information inequality, socio-territorial exclusion, and unequal distribution of resources, services, and political visibility.

This study examines how biases in Geospatial Artificial Intelligence (GeoAI) render marginalized communities invisible and exacerbate data poverty. This study is grounded in critical data analyses from a local perspective, allowing us to examine how the problem affects urban peripheries and rural areas very differently, thereby avoiding the generalization of the entire region as a homogeneous bloc. Specifically, the study identifies the main mechanisms of socio-technical bias present in spatial data and collaborative platforms (such as OpenStreetMap), mobility traces, and predictive models; examines their consequences for spatial justice, the distribution of services, and regional technological sovereignty; and proposes an ethical and public policy agenda to facilitate the adoption of GeoAI under the principle of spatial equity.

The problem is framed on the premise that, in practice, GeoAI systems amplify pre-existing biases due to incomplete data, unevenly distributed infrastructure, and training frameworks designed for the Global North. This situation creates a critical risk of technological importation: the use of predictive models built on foreign realities that do not represent local dynamics. Rather than treating the region as a uniform bloc, it is necessary to understand that this problem affects its constituent elements differently; for example, the models fail to account for urban informality in irregular metropolitan settlements, ignore population dispersion in rural areas, and overlook mobility patterns in marginalized territories. The situation is exacerbated when these algorithms are integrated into public planning or emergency management without prior local validation. The result goes beyond what could be considered a technical failure; it becomes a form of algorithmic violence that reinforces the exclusion from goods and services, as well as the diversion of investment and humanitarian aid away from communities most in need.

To carry out this study, an exploratory systematic review (scoping review) was conducted based on the PRISMA-ScR methodological criteria, ensuring the transparency and
replicability of the process. The literature search was conducted in a structured manner in high-impact international databases (such as Scopus, Web of Science, and IEEE Xplore), using keywords such as “GeoAI,” “spatial justice,” “data poverty,” and “Global South.” Due to rapid technical advances in this field, a primary time frame was defined, supplemented by foundational critical literature in geography and urban studies. After applying inclusion and exclusion criteria to filter the articles, the final set of scientific documents was organized and qualitatively analyzed along three thematic axes: data poverty, critical data studies, and spatial justice. This strategy responds to the fact that the GeoAI literature itself acknowledges that the field is expanding rapidly but still faces structural limitations in interpretability, as well as regional data availability and the variable quality of spatial sources.

The results identify three interconnected mechanisms of bias. The first is algorithmic redlining, through which urban analytics directs investment, disinvestment, and service improvements toward already advantaged areas, while depoliticizing public decisions, making territorial discrimination more evident, and concentrating poverty in marginalized areas. The second is representation bias in training data, as the literature, largely from the Global North, warns that geodiversity gaps and underrepresentation are a central obstacle to responsible AI outside the West. The third involves feedback loops, where biases introduced in data, algorithms, and user interaction reinforce one another and tend to exacerbate preexisting inequalities.

A particularly relevant finding is that information poverty functions as an active mechanism of territorial invisibilization. In OpenStreetMap, only 1,848 out of 13,189 urban agglomerations exceed 80% building completeness, while 9,163 cities remain below 20%, accounting for 48% of the urban population analyzed. The GeoAI review in human geography adds that OSM is a widely used but highly heterogeneous source, and in certain places it may be insufficient or even detrimental to analysis. Consequently, densely populated, peripheral, or informalized areas may fall outside the state’s operational map and the models that guide public intervention.

For Mexico and Latin America, the risk is greater due to the combination of the digital divide, unequal infrastructure, and dependence on external technical frameworks. In the region, the adoption and use of ICTs depend on human development, civil liberties, urbanization, and access to electricity, confirming that the digital divide has a territorial and structural basis. At the same time, the literature on GeoAI warns that its expansion may exacerbate inequalities not only in access but also in data sovereignty, computational infrastructure, and the critical capacity to influence these technologies.

It is concluded that algorithmic bias in GeoAI does not constitute an isolated technical flaw, but rather a structural socio-technical problem with direct ethical, legal, and social implications for spatial justice. A regional agenda for responsible adoption must prioritize equity audits of spatial datasets, mandatory local validation, controlled openness for external scrutiny, social participation, and data governance frameworks, as well as shared control and risk management. For Mexico and Latin America, this also requires strengthening their own technical capacities and producing context-specific evidence that allows for the evaluation of GeoAI before its routine institutionalization.

Ciencia de Información Geográfica y Sociedad
Sala de Video