We need artificial intelligence developed in Latin America

Only that which is named exists. And who names what exists? For Luciana BenottiA computational linguistics academic at the National University of Cordoba, Argentina, is as clear as he is biased. “30-year-old white males living on the West Coast of the United States” speaks Country. Essentially from Silicon Valley, where decisions are made based on broad language models that dominate tech conversation, such as ChatGPT. The problem, says this scientist, who started programming when she was six, is that artificial intelligence thinks like a minority, because when it comes to platforms like ChatGPT, it’s not just about neural networks and language, but and about an entire value system that can fall into discriminatory or sexist prejudices. Therefore, he says, it is best to have variety.

The relevance, says Benotti, is that using masses of internet data comes with the risk of replicating bias. “They often internalize existing prejudices. This could lead to results that reinforce stereotypes such as “Mapuche are drunkards” or “women in the kitchen,” says the Cordoba-based academic. “There is a lot of work going on in our field of research to reduce these biases and reconcile these models in terms of the values ​​of the global North,” he says. Country.

This diversity invariably involves the creation of broad language patterns (or large language models) Latin Americans. There are some initiatives, but they are few so far. For example, Latin LLM stands out, a variation of ChatGPT developed at the National Center for Artificial Intelligence of Chile. Part of the problem—and the need—is saying Jocelyn Dunstanresearcher in computer science at the Pontifical Catholic University of Chile, is that Latin America is never the main market. This has led to a huge gap between the amount of resources for English and Spanish language models. But this gap can be bridged through collaboration. “This will help us move faster,” he says. Country.

Spain is also starting to do the same. The Alia project was developed using databases in Spanish and other official languages ​​of Spain. Among its advantages he names Diary Eric Delgado, director of client development at IBM Spain, believes that any Spaniard will be able to contribute to its development using knowledge in their field, and this difference seems decisive compared to ChatGPT, which is trained on data from the Internet, but also has legal force. disputes over the use of site materials without consent New York Times and other media.

Latinos rule in AI development

Last September, the Economic Commission for Latin America and the Caribbean (ECLAC) presented the results of the 2nd edition of the Latin American Index of Artificial Intelligence (ILIA 2024), a study that assesses the level of preparedness of member countries regarding artificial intelligence. .

The ranking, which assumes a maximum score of 100 points, puts Chile in first place with 73.07 points; Next comes Brazil with 69.30 points; Uruguay – 64.98 points; Argentina – 55.77 points; Colombia – 52.64 points and Mexico – 51.40 points.

The most important aspect of progress in the region, ECLAC emphasizes, is the implementation of gender policies that guarantee the equal participation of women in research and development in the field of artificial intelligence. “Science will improve when it becomes more diverse,” Benotti tells the magazine. Twelve.