Flipping the Script: How AI can be used for Social Justice in Latin America

Artificial Intelligence (AI) is often perceived as a double-edged sword. On one hand, it's praised as a revolutionary technology that can solve complex problems at lightning speed. On the other, it's feared as a force that could deepen inequality and reinforce the power structures of big corporations. In truth, AI itself isn’t inherently good or bad; it’s a tool. What matters is how we use it, and how we train it. And in the communities of emerging economies, especially in Latin America, we're using it wrong.

The Problem: Misrepresentation in AI Datasets

Current models that utilize demographic datasets do not truly represent the reality of these communities. In these models, people are labeled based on categories that don’t reflect their life experiences. For example, we often use nominal categories such as “formal" or “informal” workers (i.e. workers inside or outside the formal or legal system), or classify individuals as “high income,” “medium income,” or “low income.” How can we deploy these socioeconomic categories in countries where a significant part of the economy operates informally and where income categories cannot be accurately measured?

In many Latin American countries, most of the population participates in the informal economy, which operates outside of official regulations and is not captured by traditional data collection systems. As a result, relying on conventional classifications and metrics is both ineffective and misleading, as they fail to accurately reflect the realities of these communities. Furthermore, established AI models are trained on demographic data derived from these traditional systems, meaning the AI-generated "solutions" fail to address the unique challenges faced by these populations.

AI is shaping the future. We must rethink how we collect data, label data, and how we design AI tools that work for everyone, not just the privileged few. Misrepresentations in AI keep deepening social inequalities. Instead of solving problems, AI tools that are meant to help communities often make the same issues worse for underserved groups. Rather than driving social progress, AI risks becoming another tool that widens the gap between the rich and poor.

A Framework for Change: 

1. Collecting Data That Matters

If we want AI to serve Latin America, we must start with the right data. This means that we need to find ways to collect accurate, community-specific data where data collection is difficult. This goes beyond simple surveys; it’s about capturing human experiences and identities in digital settings. This requires a collaborative research approach in which community members actively participate in identifying issues, creating appropriate labels, and gathering data that reflects their realities. By working alongside researchers, these individuals ensure that their real identity is represented in the new datasets. 

One example of this participatory research approach is the Mesa de Urbanización Participativa y Rotativa (MUP) in Buenos Aires, Argentina. Developed by the Secretaría de Integración Social y Urbana (SECISYU), this initiative was instrumental in the transformation of Barrio 31 from an informal settlement into a neighborhood. Through multiple meetings, community leaders and household heads engaged in discussions about public spaces and housing projects, allowing them to collaborate on a shared vision. Initially, they encountered misunderstandings regarding data interpretation; however, these meetings empowered community leaders to take charge of organizing data related to families. This process was not linear and included significant discussions and initial resistance. Yet, as researchers demonstrated persistence, community members began to work together, fostering mutual understanding and collaboration.

2. Completing Data That is Missing

A critical challenge in demographic datasets is the presence of missing or incomplete labels, which creates gaps in the data. This lack of comprehensive information presents a significant obstacle when building AI models tailored to these regions.

To address this issue, machine learning can be employed to complete these datasets in a way that reflects the unique circumstances of these communities. Rather than replicating external or generic datasets, probabilistic models such as Gaussian Processes or Bayesian Neural Networks can be used. These models are particularly effective with small datasets, allowing for the generation of accurate predictions despite limited data availability. By utilizing these models, missing data can be predicted based on community-specific information, ensuring that AI-generated outcomes are rooted in the lived experiences and local realities of the people involved.

This approach doesn't simply fill in data gaps, it provides context-aware predictions aligned with the social and economic complexities of the community. In this way, AI solutions are grounded in the realities of underserved populations, rather than relying on external data sources that fail to represent their specific contexts.

Changing the AI Narrative: A Tool for Social Good

It’s time to shift the narrative around AI from fear and skepticism to one of opportunity, transforming communities for the better. Too often, AI is viewed as a destructive tool, but what’s missed is its immense potential to drive positive change in cities and communities, if leveraged correctly. Instead of perceiving AI as a threat, we should embrace it as a powerful tool to improve lives by designing smarter public services, economic stability, and enhancing transportation systems, all driven by real and localized data.

To make this happen, urban planners, technologists, and community organizers must collaborate. We need an interdisciplinary approach where the people affected by AI are directly involved in its development. This collaboration will allow us to flip the script and make AI work for us rather than against us, ensuring it becomes a tool for inclusion rather than exclusion.

In the end, AI is only as good as the data we feed it. If we want to create solutions that genuinely help people, we need to adopt new frameworks for data collection and AI training. We can ensure that the cities of tomorrow are designed for everyone, starting with building a foundation that truly represents the communities we seek to serve.

Sofia Chiappero, a Fulbright Fellow from Argentina, is currently a candidate for a Master's degree in City Planning at MIT. Sofia's true passion lies in social entrepreneurship and immersing herself in the world of data analysis and economic development. Since 2019, she has been working with informal settlements in South America, acquiring insights into the challenges that communities encounter within the framework of fragile economies, political conflicts and misrepresentation of community identities through inadequate data, policies, and social narratives. Her previous works have ranged from establishing a Community-Centered Urban Lab in Argentina, facilitating dialogue between communities and government initiatives, to co-founding her own startup with a team of women from the Global South, aimed at helping marginalized communities at risk of displacement in safeguarding their livelihoods.

While at MIT, she has explored various departments, fueling her passion for innovation and problem-solving. She has also been actively involved in the startup world through MIT Design X and the MIT PKG Social Challenge accelerators, contributing to real-world projects. Additionally, she collaborates with underserved communities on economic development initiatives through the Just Money Program at MIT's Community Innovators Lab (CoLab). Recognizing her dedication and impact, she has been selected as a MAD Design Fellow at the MIT Morningside Academy of Design.