Language and algorithms: how AI has been used to curb gender-based violence since its inception – International experiences

“Promoting the use of technologies such as artificial intelligence and natural language processing will help improve our response capacity and enable us to generate evidence-based statistics for more informed decision-making.” Miranda Mejía, Director of Costa Rica's 911 Emergency System.

In recent years, artificial intelligence has gained prominence in fields as diverse as economics, art, and television. In the field of statistics and data analysis, its application is beginning to focus on the recognition of linguistic and contextual patterns. These systems examine testimonies, documents, and speeches to detect signals that, in a conventional reading, might go unnoticed. Previously, many of these traces were beyond the reach of traditional methods; today, thanks to advances in artificial intelligence (AI) and natural language processing (NLP), it is possible to address these voices from a different perspective: through models trained to identify violence, patterns, and risks.

This is the focus of the “Saving Lives” webinar, an initiative promoted by the UNODC-INEGI Center of Excellence and Costa Rica's 911 service, which brought together emergency service professionals and statistical offices from Latin America on June 16 to explore how AI can support the generation of statistical information on violence against women from 911 calls, as well as other experiences using other sources of information.

This event was divided into three main parts, where the Statistical Offices of Chile and the Dominican Republic, as well as the Center of Excellence, were able to share their experiences and knowledge on the subject. During the first presentation, Costa Rica's 911 Emergency Service shared an overview of the scope of the services provided and the main statistics on incidents and calls handled on a daily basis. Subsequently, Ignacio Agloni, from the National Institute of Statistics of Chile, gave a detailed introduction to machine learning (supervised and unsupervised) and deep learning.

Following this, Pablo Guevara, from the UNODC-INEGI Center of Excellence (CoE), presented the principles of Natural Language Processing (NLP), highlighting how call texts are interpreted and how key entities and patterns that are not evident to traditional human analysis are identified.

Another key session led by Adriana Oropeza of the CdE presented the results of the first characterization of incidents related to “payday” loans between 2015 and April of this year in Costa Rica, based on administrative records from 911. In addition, she presented a pilot project in which an Artificial Intelligence model (BERTopic) was applied, thematic clusters were constructed through activities such as database cleaning and tokenization, supported by the use of libraries such as SpaCy and NLTK for Python.

Subsequently, Klaus Lehman from the National Institute of Statistics of Chile shared his experience in the automated classification of crimes using BETO artificial intelligence in the National Urban Survey on Citizen Security (ENUSC). This process contributed to reducing human labor by 68.4%. For its part, the National Statistics Office (ONE) of the Dominican Republic showed how NLP techniques classify administrative records on building permits from unstructured narratives.

Leidy Rodriguez from the National Statistics Office of the Dominican Republic showed how NLP techniques are used to classify administrative records on building permits from unstructured narratives and characterize the properties to be built.

Finally, the UNODC-INEGI Center of Excellence presented the results of the automated analysis of 911 calls to detect specific cases of violence against women, explaining each phase of the model in detail: from data selection and cleaning to algorithm implementation and validation.

This model analyzes the texts of 911 call transcripts in order to separate them into two groups, those with and without violence against women. This proposal establishes a new technological option capable of identifying the level of risk to victims, taking advantage of the timely information obtained as a result of incident reporting. Currently, the Center and Costa Rica's 911 service are beginning work to replicate this methodology.

The closing session of the webinar was devoted to the ethical aspects of artificial intelligence. Pablo Guevara, from the UNODC-INEGI Center of Excellence, raised urgent issues: How is personal data protected? What biases can models introduce? How can we prevent automation from reinforcing existing inequalities? The session highlighted the need to implement anonymization mechanisms, monitor algorithmic biases, and constantly validate the models used.

The “Saving Lives” webinar highlighted that NLP can be a very useful tool for characterizing different phenomena, through methodologies with a solid conceptual framework, and by carrying out activities with care for ethical aspects and documenting the development and implementation of models.