Latin American experts collaborate to advance AI in research

A new continental collaboration around AI in science and research was the unanimous outcome of a gathering this month of experts from countries across Latin America and the Caribbean, led by the International Science Council (ISC). Only by listening to all regions can there be a proper global conversation about generative AI.

An early step in the collaboration kicked off by the experts from 11 countries will be to share and map developments and discussions, strategies and actions involving AI in science and research systems across the region, and to keep updating this resource. Hopes for collaborative AI initiatives for the region might face political and funding challenges, but the will is there.

The workshop on “Preparing National Science Ecosystems for AI – The policy perspective” was hosted by the ISC Centre for Science Futures and ILDA – the Latin American Initiative for Open Data – on 9 April 2024 in Santiago de Chile.

The discussions revealed a diversity of experiences across the region, but also commonalities and a sense of coherence.



This regional description will help to inform an alliance going forward, based on the learnings and experiences of all participants.

Collaborative work could subsequently spread to other experts in the AI space around the region, and could support other processes such as coordinating public policies or recommending joint AI-related efforts.

The workshop and the backdrop

There were 20 participants in the workshop in all, including 14 from Latin America and the Caribbean. They represented 11 countries in the region: Argentina, Brazil, Chile, Colombia, Dominican Republic, Panama, Honduras, Jamaica, Mexico, Peru and Uruguay.

All of the experts are people who are directly involved in designing and rolling out national AI strategies for science. They highlighted their countries’ main achievements in AI for science so far, and the problems they faced. How can Latin American and Caribbean countries and the region make the most out of AI?

The workshop had three main purposes. One was to hear inputs from country representatives and exchange knowledge and experience on developments around AI and science. There is currently very limited knowledge on what countries are doing to prepare their research systems for AI. A second purpose was to enlarge the global network of people the ISC Centre for Science Futures is establishing to continue future projects around AI for science.

Third, the Centre’s newly published, early-version working paper on AI and research titled Preparing National Research Ecosystems for AI: Strategies and progress in 2024 served as support for the discussions.

The paper comprises a review of literature on national policies around generative AI for science, and insights from countries around the world through case studies. It offers a framework outlining key issues to consider when planning to integrate AI into research systems.

“The workshop will help increase our understanding of strategies in Latin America and the Caribbean to integrate AI in research,” said Dr Mathieu Denis, head of the ISC Centre for Science Futures.

At a related regional workshop for Asia Pacific held in Malaysia last October, there were a similar number of countries attending but they were much more disparate in terms of their responses to AI and in the maturation of their science systems’ ability to adapt to AI.

By contrast, the Chile regional meeting uncovered a large number of points of contact and areas where countries can work together around AI and science systems.

Like so many others around the world, the Latin American and Caribbean experts felt overwhelmed by the speed of changes, and feared that their countries were falling behind in the uptake of AI in research.

But from the cases and discussions that unfolded it became clear, the International Science Council concluded, that countries in Latin America and the Caribbean are acknowledging opportunities and issues relating to AI and research, and have put in place structures – a person, group, a new institution. They are developing plans and some have started rolling them out.

“Despite perceptions that people may have in the region, countries such as Mexico, Argentina, Chile, Uruguay, Colombia and others are not lagging behind other parts of the world in terms of their acknowledgement of the issues and the development of responses,” Denis told University World News.

Shared challenges

The workshop revealed very compatible assessments of the opportunities, issues and resources required for AI to work for science in the region.

Among the key challenges presented by experts was the need to improve education and training around AI, at every level of education systems, and for nationwide digital literacy frameworks. Another was the need for clear research data policy frameworks. A third was the development of infrastructures and access to AI hardware and computing resources across Latin America and the Caribbean.

There was considerable debate around data policy and governance for research, which is often sidelined in discussions about generative AI and research, the experts agreed.

“AI is data, and good AI rests on good data,” said one participant. “But for many commentators outside talking about AI, that relationship is non-existent or they don’t see it. That is concerning, and something we need to be working on.”

There were arguments for the importance of open data for research, not only in terms of public access and not only regarding governments but also private companies and other data generators.

The discussions at the workshop confronted two approaches to data, said Mathieu Denis: Open Data versus FAIR Data. FAIR stands for ‘findable, accessible, interoperable and reusable data’. “The concepts aren’t necessarily opposites on paper, but they do lead to different priorities in research data policy and management.

“The debate is whether, first, research data should always be open by default – the open science approach – or if the emphasis should rather be put on ensuring FAIR data. And second, how to decide what datasets to open or not? This is a growing debate, in Latin America and the Caribbean and elsewhere.”

Nicholas Kee, a participant in the workshop from Jamaica, is an AI developer and advisor, start-up founder, tech non-profit director and digital security expert among other things – including his involvement in developing AI strategy for his country.

“We have a lot of problems with research data, or at least how we think about data in relation to how it is collected, how it is standardised, how it is stored, and how it is used. This makes it harder to glean key insights that could help us address problems,” such as depleting biodiversity or climate changes, he told University World News.

A particular problem for generative AI models is ‘dirty’ data that provides dirty results. “A part of us taking a step back without even looking at AI models on the whole, is to audit the research data that we currently have and to set in place best practises for collecting and processing data.” Setting standards and ethical guidelines for data and AI research is key, and must also be done at the international level, given the ongoing internationalisation of science.

There is a healthy level of scepticism within the research space about the reliability of research results obtained through existing AI models run, for instance, by the private sector. “Often there is not much academic input into the development of AI models, and so it’s hard for the academic space to trust existing AI solutions that are currently being churned out,” said Kee.

Some experts spoke of the need to address systemic biases embedded in the data, and revealed by the Large Language Models that use the data, especially regarding biases against minorities. Interestingly, the region is rich in data on minorities and bias – something the global community might consider taking advantage of to improve AI.

Mathieu Denis explained: “The issue for research is that, unless specific attention is paid to research data used, LLMs will produce ‘epistemic biases’ for the very same reasons that they reproduce biases against minorities: they will run on research data skewed towards certain (data intensive) fields, and produced in data-rich countries. Whereas good science will depend on a variety of robust datasets, from different fields and different parts of the world.”

There has been excitement recently about the proliferation of Small Language Models as a route to improving the quality of data that AI models are built on, and which might also help institutions and countries, perhaps regions collaboratively, to create AI models drawing on different languages and alternative data sources.

Imperatives for collaboration

A project to create a continental repository of excellent data relating to AI and research is a big ask, in itself, and for any region, over and above difficulties that could be encountered trying to access data from governments, publications or higher education institutions, and lack of funds.

“Piloting such collaborations around a few issues would allow demonstrating to countries and society the value of collaboration in the use of AI to address common problems,” said the representative from Colombia.

The experts agreed that a repository was a good idea, but alone it will not build pathways for productive data and AI collaborations between nations or across the continent. Coordination is needed to help to build bridges between countries and institutions that have more or fewer human or financial or capital resources, and also regulation and governance interoperability.

The experts stressed the advantages of researcher collaboration to share resources and strengthen Latin America’s nascent networks around AI and research, and also of stronger cooperation between research institutions and government agencies. The idea of a continental masters in AI was floated, as was an Erasmus-style exchange programme around AI and research.

There is potential for different kinds of bilateral as well as multilateral regional activity to integrate AI into research systems in Latin America and the Caribbean. Another idea was to create ‘birds of a feather’ groups within the region that are focussed on AI, and to use their platform to connect to the rest of the world.

A crucial point was the idea that countries could identify one or two non-controversial areas in which AI could play a pivotal role, like climate and food systems, and start collaborating. This would involve mutualising relevant datasets, developing joint infrastructures and training international teams at the regional level.

For Dr Dureen Samandar Eweis, science officer at the ISC Centre for Science Futures, it was interesting that participants, including from Mexico and Uruguay, believed countries and the region should focus on what they are strongest at, “and not necessarily follow what is happening elsewhere”, she told University World News. Similarly, there was a need for local or regional rather than foreign technologies.

The point was made that this is necessary from a scientific point of view, but would require serious collaboration between countries as none of them really has the capacity to fund AI research infrastructures and tools alone.

The next steps

There is a great deal happening very fast around generative AI and science, and the time is ripe to find out what is going on in countries across Latin America and the Caribbean and what can productively be done together.

As one expert said: “I was happily surprised to hear words like dream, creativity, and [a] very positive aspirational spirit in this discussion. Certain sentences stayed with me. ‘We need to transform our concerns into exchanges and into action. We need to figure out what we can do with what we have, and what it is that we need’.”

The group agreed to continue meeting and reflecting together on the integration of AI in Latin American and Caribbean research systems.

The ISC Centre for Science Futures was encouraged by the emergence of substantive issues around AI and science that could be worked on collaboratively, including around research data policy and management and standards for the use of AI in research.

The ISC and ILDA called on the experts to gather more people into the nascent AI and research network and to forge links with decision-makers in their countries.

“Given the complexity of the countries and region, the large and diverse numbers of actors and the multifaceted reaches of data and generative AI, there is a clear need to create an expert network that can bridge the many parallel conversations taking place around AI and research,” Mathieu Denis told University World News. “I’m confident that we will work well with this group in future.”

Email Karen MacGregor on macgregor.karen@gmail.com.