Artificial Intelligence and Public Safety in Latin American Democracies

Abstract

Artificial intelligence is revolutionizing various areas of public policies. Its application in the fight against crime permeates contemporary political discourse and appears in the public security proposals of presidential candidates. Ensuring that AI is used responsibly and ethically is essential to maximize its benefits and mitigate the risks it poses to civil liberties in Latin American democracies.

Recent breakthroughs in artificial intelligence (AI) are revolutionizing a number of policy domains. Public safety is no exception, with AI and machine learning applied to policing and law enforcement more generally. The public safety uses of AI-based technology range from systems that identify violators of traffic rules, to those that predict where future crime is likely to occur, to those that help prevent online fraud to forensic DNA testing.

Several uses of AI for public safety, defined as “the application of algorithms to large sets of data to either assist human policing or replace it,” have received considerable attention in the United States and other advanced industrialized countries. Among the most prominent uses are face recognition systems–which capture individuals’ unique facial features and sort through potentially millions of possibilities to establish a suspect’s identity–and license plate reading systems–which can capture, identify, and match license plate numbers to vehicles and their owners and find their whereabouts.

Some of these applications, including the recognition of vast numbers of images or patterns of behavior to predict illegal activity or the assignment of crime propensity scores to individuals based on risk factors, remain controversial due to ethical and privacy considerations. Although its adoption shows great potential to address crime, AI also presents important challenges for liberal democratic contexts in which the protection of civil liberties is paramount.

Despite these concerns, many Latin American countries have quickly turned to adopting AI technologies for law enforcement purposes. Proponents point to speed, the ability to analyze vast amounts of data that would be prohibitive for humans, and the reduction of human errors and bias as important benefits that warrant the widespread adoption of AI for public safety.

Mindful that AI-based crime-fighting technologies remain understudied in Latin America compared to the United States, this text presents a primer on the current state of AI-based crime-fighting technologies in the regino’s democracies. In the following sections, we turn to examples of AI used to enhance public safety in Latin America. Through these examples, we then illustrate their intended purposes, as well as the opportunities and challenges that they represent.

The Promise of AI for Public Safety in Latin America

AI technology to address crime holds great promise in Latin America. With only around 8% of the world’s population, the region accounts for about 30% of global homicides. Additionally, while there is considerable variation in terms of rates of violent crime across countries-–from Chile, with a rate of about 4.5 homicides per 100,000 people, to Ecuador, with a rate of about 44 per 100,000—national averages can mask considerable higher rates within countries.

In this context, AI-based technology can help law enforcement agencies carry out several tasks more efficiently and effectively. In particular, AI can assist with the collection and analysis of massive amounts of crime data and predict potential hotspots for criminal activity, allowing law enforcement to deploy resources more effectively. By identifying patterns in time, location, and behavior, AI can help resource-constrained law enforcement agencies preemptively act to deter crime.

AI can also enable the region’s police to carry out real-time monitoring to maintain public safety. AI tools can help monitor surveillance cameras, social media, or other open sources of information to detect suspicious activity and alert authorities as events unfold. Similarly, AI can be leveraged for facial recognition and biometrics toward the identification of suspects or missing persons more efficiently, particularly in crowded or chaotic environments.

On the forensics side, AI tools can play an important role in evidence management and analysis. They are valuable to organize and sift through vast amounts of digital evidence, improving the efficiency of investigations and judicial proceedings. In doing so, AI has the potential to help address human biases and guide decision-making based on objective data to lead to more transparent and just outcomes.

AI Use Examples in Latin America

Because of the region’s pressing concern with violent crime, Latin American governments have looked to incorporate technological advancements in AI towards public safety. While the region’s adoption of AI for public safety has not been linear–rather, there has been considerable trial and error–several examples of its adoption are encouraging. As the number of cases increases, and as governments and publics become accustomed to the technology, its adoption is likely to become more generalized.

Governments at all levels have incorporated AI for public safety in the region. In Colombia, the National Police published a national strategy to provide local governments with resources to purchase unmanned drones and surveillance cameras to prevent and detect crime. To connect the vast amounts of criminal data it held and the various points of information it had to make decisions, the government hired an external software from Amazon Web Serices to organize and store data from several sources. Apart from analyzing results quickly, the software also leveraged a series of tools including Nuvu’s XCrime app to aggregate and analyze information as well as predict potential crimes.

In Chile, the National Insurance Association developed AI software to identify and report stolen vehicles. The software turns cellphones into license plate readers, which police departments can easily use. It automatically identifies stolen vehicles by scanning license plates and comparing them to a database in seconds. The program was initially rolled out in 60 municipalities and expanded to 345 in 2022.

Efforts have taken place at the state and local levels as well. In the state of Mexico, the United Nations Office of Drug and Crime (UNODC) partnered with the country’s statistics agency INEGI to develop a program that leverages deep neural networks to identify key words and phrases from previous 911 calls suspected of domestic violence or crimes against women. At the municipal level, local governments have also made efforts to incorporate AI–especially AI-assisted surveillance programs–to address crime.

The government of Benito Juárez, one of Mexico City’s 16 boroughs, adopted a public safety strategy in 2018 to address crime through AI-based public surveillance tools. The municipality increased the number of police units and created a wide surveillance network comprised of the government’s cameras and private cameras given to private citizens as part of public safety kits. Footage is linked to a Control and Command Center (C2) which leverages facial recognition and license plate technology to identify and track suspects once a report is filed. The program also leveraged citizen participation by giving them access to Blindar BJ – an app where citizens can report crimes and track police unit locations. In 2021, the program was expanded to the neighboring borough of Álvaro Obregón.

In 2018, the municipal government of Tlajomulco de Zúñiga, in the state of Jalisco, also sought to improve public safety by integrating AI-based technology into surveillance systems, including monitoring public areas with video surveillance cameras and offering citizens publicly-available information on crime incidence. The effort involves 564 cameras for face recognition and license plate readers to monitor behavior, identify patterns, and report suspicious activity.

In Brazil, the government of the city of São Paulo inaugurated in 2022 the use of an AI-based facial recognition system to monitor Line 3 of the subway system through 14,000 cameras in 18 stations. The city is also expanding its biometrics-based surveillance beyond the subway through the Smart Sampa program which will link ~20,000 cameras equipped with facial recognition to a monitoring center and to other agencies’ databases for tasks ranging from identifying subjects to locating ambulances. Although the project has received criticism over potential biases, the city’s mayor announced the center’s launch in April and expects to fully implement the video surveillance system by the end of 2024.

In Uruguay, as Montevideo started experiencing higher crime rates, the city’s police department adopted in 2016 a predictive policing tool named PredPol. By leveraging historical crime data, the software generated crime predictions maps to inform police efforts to deter crime or respond quickly to emerging threats. PredPol was adapted to Montevideo from the tool first used by the Los Angeles Police Department in California to generate crime predictions based on crime type, location, and date/time. At a cost of US$ 140 million annually and with a proprietary algorithm to which governments are not privy, this software helped to streamline police deployment and preemptively address crime. However, the government ended the program after finding no differences between zones that used the predictive software and those that did not.

Challenges for the adoption of AI for public safety

While these examples show the promise of AI-based technology for policing and the enthusiasm it has generated among governments as a potential aide in addressing rising levels of crime, the use of AI in law enforcement poses several important challenges. Facing public pressure to deliver public safety results, governments have moved quickly to adopt AI technologies toward this end, but they often do so without proper legislation in place to protect privacy and civil liberties, promote transparent procurement processes, and ensure the sustainability of AI. In this section, we discuss some of the main challenges facing governments in the adoption of AI technology for public safety, actions taken to address these shortcomings, and opportunities.

Regulatory gaps

AI Regulation is crucial to prevent its misuse and anticipate unintended consequences. While most Latin American countries lack AI legislation in general, and for public safety purposes in particular, some are in the process of generating regulatory frameworks (e. g., Argentina and Brazil) based on multilateral regulatory regimes such as the European Union’s AI Act or with support of the Inter American Development Bank. However, as regulators grapple with safeguarding rights without stifling economic growth, debates on how to approach AI regulation continue. For example, Brazil has two regulatory legislative proposals: the first places fewer restrictions on AI, as it aims to create “a decentralized system and restricts government intervention,” while the second includes principles modeled after the European Union’s AI Act. While AI legislation is less common, some countries such as Chile, Brazil, and Colombia have published national AI strategies that provide guidelines to advance AI adoption.

Additionally, international guidelines are also influencing the creation of AI frameworks in the region. For example, Brazil aligned its national AI strategy with several OECD principles. Further, several Latin American countries recently signed the Santiago Declaration that encourages more active participation in AI deliberations and shows a commitment to creating governance frameworks tailored to Latin America’s own needs.

Privacy and Civil Liberties

Although AI can be an effective tool to address crime, it also carries important risks for human rights. For example, one common use is to monitor public spaces to identify potentially harmful behavioral patterns and prevent crimes. Although this can be beneficial, surveilling public spaces to predict crime has deeper implications, as this constant monitoring can turn everyone into a potential suspect even when no crime has been committed. In other words, guilt is anticipated and estimated instead of giving people the benefit of innocence until proven guilty.

In this context, an important concern is the potential use of AI to undermine the protections afforded to citizens under liberal democracy. In particular, AI can be used to monitor citizens’ lawful activity, as is the case in authoritarian regimes, such as China. Facial recognition, license plate readers, or cell phone location transmitters can be used to track individuals, even when they have not committed a crime.

A few examples from the region are illustrative. In Argentina, the police mistakenly logged the wrong variables to identify a suspect and instead jailed an innocent man for six days. In Mexico City’s Blindar Benito Juarez Program, some users noted the lack of a privacy disclosure notice when downloading the government’s app, raising concerns over users’ privacy. In Ecuador, there are reports that the government’s intelligence agency relies on the ECU-911 crime surveillance system imported from China to spy on journalists and politicians for political advantage. These incidents show how AI technology can undermine civil liberties if governments put it to the wrong use.

Procurement and Corruption

Government procurement processes in Latin America are often opaque, and the acquisition of AI-based technology is no exception. Governments in the region have been quick to adopt AI for public safety because of the rapid growth of organized crime, but transparency in the acquisition processes has lagged. A report by AccessNow found that some foreign firms that supplied services in Latin America denied selling surveillance tools, rephrased their purpose, or deflected accountability to end users.

In the case of Mexico City’s Blindar BJ and Alvaro Obregon, the lack of transparency over funds’ allocation has raised eyebrows. As reference, the government invested MX$385 million (about US$19.6 million) in a span of three years, but it has faced allegations of inflated equipment valuation. While market research estimated that each camera cost around MX$2,700 (US$159), the borough government budgeted them for MX$35,000 (US$2,000).

Technology Adoption and Maintenance

Apart from requiring more robust regulatory frameworks, Latin American countries also tend to lack sufficient infrastructure to ensure technologies function properly. In Bogotá, Colombia, the city government adopted a system to identify and predict crime based on AI. With a significant investment of $11M, the system raised concerns about human rights and insufficient police to act on the information generated, but in particular it was criticized because an estimated 22% of its cameras were not working properly. Further, differences in the software used across AI platforms limited information sharing and efficiency–as was the case with the softwares for the public bus rapid transit system Transmilenio and for the central command center for the city’s surveillance cameras, C4.

Similarly, in Mexico City, the Blindar Benito Juárez program faced important implementation challenges. In particular, neighbors complained that cameras were not recording, making it difficult to present evidence when crimes took place. These challenges are likely to shape perceptions of the usefulness of AI for law enforcement if left unaddressed.

“Black-Box” Decision Making

The algorithms employed in the AI software are typically not made known to government agencies. In most cases police departments relying on the technology will have little knowledge about how input data are weighted or the extent to which biases can be present. In the case of Montevideo’s adoption of AI, high costs (US$140 million per year), dependence on a fixed training data set, and biases in the reporting of historical crime data were important concerns. Since the algorithm used historical police data to feed the algorithm, critics noted the model could direct attention to already policed areas, creating a biased feedback loop. Further, given the software’s proprietary nature, authorities did not have access to the software’s algorithms. Similarly, with Brazil’s Smart Sampa project, human rights organizations have highlighted its potential to incarcerate Black and low income individuals more frequently than other individuals.

AI in the public eye

As Latin America’s use of AI-technology for law enforcement increases, so does its presence in the public domain. In particular, AI-based technology to fight crime is entering contemporary mainstream political discourse, including presidential candidates’ public safety proposals. For example, during Panama’s 2024 presidential election, several candidates proposed leveraging AI technology to prevent crime. President José Raúl Mulino proposed additional training and procurement of new AI technology to prevent crime during his campaign, and former president Martín Torrijos, advocated during his campaign for the use of “… technology and surveillance and monitoring systems with artificial intelligence” coupled with increased police force units.

Similarly, candidates in Mexico’s presidential and mayoral elections have also campaigned on the use of AI to curb crime, ranging from installing the highest number of surveillance cameras to expanding public safety programs that have leveraged AI technologies such as surveillance cameras and real-time police unit tracking. As the right-of-center candidate in the 2024 presidential election put it, “we are all in on the use of technology and AI to address crime.” Although these efforts seem promising, critics are concerned that candidates place more emphasis on implementing or expanding programs without also considering the policies required to regulate these technologies and ensure their responsible use.

Despite Latin American countries’ controversial history of government surveillance on civil society and political opponents, there does not seem to be a strong opposition to the technology. In fact, recent polls show citizens may have a positive outlook towards AI as well. In 2023, Ipsos, a global market research firm, surveyed over 22,000 individuals across 31 countries, including four in Latin America: Mexico, Colombia, Chile, and Argentina, asking about public perceptions of AI.

Globally, around 67% of respondents strongly or somewhat agreed that they had a good understanding of AI. Comparatively, Latin American averages were generally higher (Mexico – 75%, Colombia 73%, Chile 70%, and Argentina 67%). Similarly, while 54% of respondents worldwide strongly or somewhat agreed that AI products and services provide more benefits than drawbacks, the share of respondents sharing this view was higher in Latin America: 73% in Mexico, 65% in Colombia, 59% in Chile, and 57% in Argentina. Notably, attitudes toward AI among Latin American countries generally were more favorable than those in European counterparts.

Conclusion

The adoption of AI-based technologies for public safety in Latin America’s democracies offers significant promise in addressing the region’s high rates of violent crime and improving law enforcement efficiency. From predictive policing to facial recognition and real-time surveillance, AI tools have the potential to transform crime-fighting efforts by enhancing data analysis, resource deployment, and forensic investigations in a region where drug trafficking and organized crime more generally are on the rise. However, the rapid integration of these technologies also presents substantial challenges for liberal democracies, including regulatory gaps, privacy concerns, and issues of transparency in government procurement. The risk of civil liberties being undermined, coupled with the lack of proper infrastructure and maintenance, underscores the need for a more cautious and regulated approach to AI implementation.

As AI becomes more prominent in political discourse and public safety strategies, governments must prioritize establishing robust legal frameworks that balance innovation with human rights protections. Ensuring that AI is used responsibly and ethically will be key to maximizing its benefits while mitigating the risks it poses to civil liberties in Latin American democracies.

References

A otro nivel. Plan de gobierno 2024-2029 Martín Torrijos. (2024).

Access Now. (2021, August 10). Made Abroad, Deployed at Home.

Access Now. (2023, April). Remote biometric surveillance in Latin America. Are companies respecting human rights

App “Blindar Benito Juárez” te permite ver cámaras de seguridad; así funciona. (2021, April 8). El Heraldo de México.

Asociación de Aseguradores de Chile. (2022, June 22). Asociación de Aseguradores firma convenio con municipios para detectar y denunciar autos robados.

Benito Juárez presenta su centro digital de vigilancia. (2021, April 7). El Universal.

Cosme, M. (2021, November 17). Lía Limón solicita 5 mil 784 mdp de presupuesto 2022 para Álvaro Obregón. El Sol de México.

Cumbre Ministerial y de Altas Autoridades de América Latina y el Caribe. (2023, October 24). Declaración de Santiago.

Díaz, O. (2023, November 13). Miguel Hidalgo, Álvaro Obregón y Azcapotzalco piden aumento de presupuesto para el 2024.

EGA. (2024). Artificial Intelligence. Latin America’s Regulatory and Policy Environment.

Freitas, H. (2024, July 4). Smart Sampa: Prefeitura inicia programa de câmeras com reconhecimento facial em São Paulo. O Globo.

Gaudín, A. (2017, May 5). Uruguay Tries Preventative Policing with a High-tech Twist.

Gobierno de Jalisco. (2021). Plan Municipal de Desarrollo y Gobernanza. Tlajomulco de Zúñiga.

Gobierno de la Ciudad de México. (2023, March 28). Síntesis informativa. Alcaldía.

IA al servicio de la seguridad pública: el éxito de la Policía Nacional de Colombia prediciendo crímenes en AWS. (2023). AWS.

Ipsos. (2023, July). Global Views on A.I. 2023. How people across the world feel about artificial intelligence and expect it will impact their life.

Joh, E. (2018). Artificial Intelligence and Policing: First Questions. Seattle University Law Review, vol. 41, 1139-1144.

Kessel, J. (2019, April 26). In a Secret Bunker in the Andes, a Wall That Was Really a Window. The New York Times.

Lum, K. (2016, October 10). Predictive Policing Reinforces Police Bias. HRDAG.

Manjarrés, J., & Newton, Ch. (2024, February 21). InSight Crime’s 2023 Homicide Round-Up. Insight Crime.

Mari, A. (2023, July 13). Facial recognition surveillance in São Paulo could worsen racismo.

Mascellino, A. (2022, December 9). Brazil deploys ISS facial recognition to secure São Paulo metro. biometricupdate.com.

Ministerio del Interior, Ministerio de Defensa Nacional, Consejería Presidencial para la Seguridad Nacional. (2020). Política Marco de Convivencia y Seguridad Ciudadana 2019-2022. Dirección de Antinarcóticos.

Municipio de Teno. (2024). Municipio de Teno instala lectores de patentes para detectar autos robados.

Naundorf, K. (2023, September 15). Un escándalo en Buenos Aires revela los peligros del reconocimiento facial. Wired.

Nuvu. (2023). Policía Nacional de Colombia.

OECD.AI Policy Observatory. (2021). National AI policies & strategies.

OECD.AI Policy Observatory. (2022, September 6). Brazilian AI Strategy.

Plan de Gobierno del candidato a la presidencia José Raúl Mulino. (2024).

Rangel, L. (2024, March 7). Tecnología en propuestas de seguridad de candidatas: uso de datos biométricos y videovigilancia pone en riesgo derechos humanos. El Sabueso.

Sin transparencia el funcionamiento de las aplicaciones de seguridad de la BJ. (2021, February 25). DDM Benito Juárez.

¿Sistema de videovigilancia de la Ciudad ¿en cuidados intensivos? (2022). Concejo de Bogotá.

Tenorio, G. (2018, May 6). Santiago Taboada hace público su Plan de Gobierno 2018-2021 para la alcaldía de Benito Juárez. Periódico Leo.

Tlajomulco presenta modelo de vídeo vigilância. (2018). Tlajo.

UNODC México. (2023, January 30). Inteligencia artificial para detectar y prevenir la violencia contra las mujeres.