Latam-GPT Aims for AI Sovereignty in Latin America

Chile Spearheads Regional AI Initiative

Latam-GPT is a collaborative effort led by Chile’s National Center for Artificial Intelligence (CENIA) and supported by 65 institutions across 15 countries. The project received funding from the Development Bank of Latin America and the Caribbean (CAF), the Inter-American Development Bank (IDB), and the Chilean Ministry of Science, Technology, Knowledge and Innovation.

Described at its launch as “the first model developed by and for Latin America,” Latam-GPT intends to transform the region from a passive consumer into an active player in global AI development. According to its website, the initiative represents a concrete move towards greater regional technological autonomy.

The Drive for AI Sovereignty in a Geopolitical Shift

The concept of AI sovereignty has gained traction amidst rapid AI advancements and shifting global power dynamics. With most foundational AI models originating from the US and China, concerns have grown regarding their training data’s lack of linguistic and cultural nuances relevant to the Global Majority, including Latin America.

Policymakers and researchers are increasingly considering the profound societal impacts of these models from a sovereignty perspective. Stanford’s Human-Centered Artificial Intelligence center explains that each layer of the AI technology stack involves differing interpretations of what sovereignty entails, alongside varying degrees of control. However, a universally shared definition of AI sovereignty remains elusive, leading to diverse and sometimes incompatible interpretations.

Navigating “Sovereignty” Amidst Interdependence

While the goal of full AI autonomy is appealing, it presents significant financial and capability challenges for many regions. Latam-GPT, despite its stated aim of sovereignty, is not built from scratch; it leverages Meta’s open-source LLM LLaMA and employs a continued pre-training (CPT) technique. Furthermore, the initiative utilizes Amazon Web Services’ infrastructure.

A public-facing interface, Copuchat, built on OpenAI’s API, was also developed to collect data for training the model. These details, while available in the project’s FAQs, were not prominently highlighted during the launch. Critics suggest this reliance on foreign technology creates a nuanced reality that contrasts with the project’s strong sovereignty narrative.

Questions also persist regarding the project’s sustainability, as future funding beyond initial contributions from CAF, IDB, and the Chilean government remains unclear. The long-term involvement of volunteer contributors across multiple countries also poses a potential challenge, raising concerns about the most strategic use of regional resources.

Defining the Path Forward for Regional AI

For Latin America, a clear definition of AI sovereignty is essential, alongside a strategic articulation of achievable goals. Researchers have proposed concepts like “strategic interdependence” and “managed interdependence,” focusing on deliberate decision-making and strategic alliances rather than absolute autonomy.

Rather than solely focusing on model development, the Latam-GPT consortium could advocate for frontier AI companies to incorporate their curated corpus into global models. Exploring public-private partnerships could also ensure the corpus’s accuracy and representation, urging international companies expanding into the region to prioritize local cultural nuances. An open regional conversation about what is truly feasible and strategic for Latin American AI development is paramount.