This article—Part I of a two-part series—asks a simple yet vital question: In an era when scientific and technological progress drives prosperity, what should be the top research and development priorities for Latin America?
Beyond intellectual curiosity, there is a real purpose in this exercise: contributing to a common cause in order to produce a quantum leap in scientific and technological development in Latin America, a region known for lagging behind most of the rest of the world precisely in that area.1
A similar exercise has been tried by a variety of institutions and analysts several times. Just for illustrative purposes, RELAI—a network of leading public CTI agencies across Latin America—recently surveyed its members, institutions that shape or implement national R&D and innovation policies, to identify the key scientific and technological fields they deem most important.
Source: Cuello, H., A. Glennie and N. Zolho. 2022.Innovation Agencies in Latin America and the Caribbean: A characterization of the RELAI Network. IGL/NESTA-RELAI-IDB. p.20. Washington DC.
The priorities emerging from that survey reflect the views of Latin American decision-makers. They offer a snapshot of how unique socio-economic contexts, current scientific and technological strengths, as well as innovation potential combine into a list of critical areas that could significantly boost the region’s economic development and quality of life. Yet, while all the areas identified are important, deciding how to rank them remains a challenge. Traditionally, national STI agencies in Latin America and beyond have used prioritization frameworks to choose which projects or research lines to fund. However, the rationale behind selecting specific criteria and their assigned weights is rarely explained in depth. What’s more, addressing these complex questions at a regional level is an even more complex task.
In this article, we explore a different approach. We introduce a key preliminary step: developing a prioritization framework. The idea is simple—before selecting R&D fields to prioritize, we must systematically define the criteria for making that choice and determine how to weigh them.
To tackle this task, we develop an AI-assisted prioritization framework. Think of it as a decision-making tool that can be applied to projects, various initiatives, or entire R&D fields.
In Part II of this series—set to appear in a couple of weeks—we will apply this AI-assisted framework to a list of R&D fields developed through a more traditional inductive method. The result will be a robust, prioritized list that serves as a starting point for dialogue among experts, authorities, and stakeholders, paving the way for informed decision-making at both national and regional levels.
Using AI to develop a region-wide decision-making framework in R&D
To begin with, we prompted DeepSeek R1 to produce a prioritization framework for investments in R&D in Latin America, taking into account common sense criteria that are regularly considered by public agencies in similar exercises around the world, such as relevance for the socio-economic context, alignment with current trends in science and technology, complementarity with current efforts and strengths, and a few others. In addition, we educated DeepSeek into the larger issues of the underperformance of Latin America in science and technology, as well as in the basics of the economic structure of the countries of the region.
Before presenting the outcome, it is worth looking at the reasoning and method of DeepSeek R1. This LLM allows for the person prompting it not only to receive a response, but to witness with some detail how it went about step by step to accomplish the assigned task. An excerpt of the reasoning when we asked for a prioritization framework for region-wide investments in R&D in Latin America follows:
Okay, I need to help the user create a conceptual framework for prioritizing science and technology investments in Latin America. The user mentioned several key issues: underinvestment in R&D, low patents, few PhDs, lack of leading institutions, etc. They want a framework that considers centrality to global trends, relevance to local issues, complementarity with existing efforts, and other criteria.
First, I should outline the main criteria. The user mentioned centrality, pertinence, complementarity. Maybe add others like feasibility, capacity building, and scalability. Then, determine how to weight these criteria. The user suggested using a weighted system or alphanumeric order. A scoring system with weights seems more practical for decision-making.
Next, how to apply these criteria. For each potential area, score them on each criterion, multiply by weights, sum for a total. This gives a quantifiable way to compare. Need to explain each criterion clearly. For example, centrality would involve looking at global trends like AI, biotech, climate tech. Pertinence would align with Latin America's issues like inequality, biodiversity loss, agricultural challenges.
Complementarity requires researching each country's existing investments. Maybe use a matrix to map existing efforts vs. proposed areas. Feasibility considers infrastructure, political support, talent. Scalability is about potential for broad impact. Capacity building ensures long-term sustainability by developing local talent.
…
The framework should also include steps: define priorities, collect data, score, compare, validate with stakeholders.
…
I need to make sure the framework is actionable. Maybe suggest a weighted scoring model with percentages.
…
Also, address challenges like data collection for complementarity. Suggest reviewing national science agency websites, policy documents, and databases like RICYT. Emphasize stakeholder engagement to validate scores and ensure buy-in.
Finally, structure the answer with sections: criteria, weighting rationale, application steps, tools, example, and challenges. Keep it clear and practical for policymakers.
After several human/LLM iterations in which interaction led to the addition of some criteria not initially considered, the process above led to the following prioritization framework (slightly edited for clarity):
1. Core Criteria for Prioritization
· Each criterion is assigned a weight (e.g., 1-5) based on strategic importance. and total scores determine priority.
· In the following table, each criterion is weighted to reflect its strategic importance. Total weights sum to 100%.
2. Application of the Framework
Step 1: Define Priority Areas
· Identify candidate fields (e.g., agritech, tropical medicine, renewable energy, AI for biodiversity).
· Step 2: Data Collection
· Global Trends: Use bibliometric databases (e.g., Scopus) and reports (e.g., WIPO, OECD) to identify high-growth fields.
· Regional Challenges: Map priorities using UN SDGs, national development plans, and regional needs assessments (e.g., Amazon deforestation, urban informality).
· Complementarity: Audit existing initiatives via national S&T agency websites (e.g., Argentina’s MINCyT, Brazil’s MCTI) and regional platforms (e.g., RICYT).
Step 3: Scoring & Ranking
Projects/fields are scored (1-10) on each criterion,
• Stakeholder Validation: Engage governments, academia, and industry to refine weights/scores.
• Dynamic Prioritization: Reassess annually using updated data.
As clearly articulated in the LLM’s proposal, a typical selection framework has three moving parts: i) prioritization criteria (such as those listed above); ii) a system of relative weights, embodying how a particular criterion is more or less important vis a vis the others; and iii) a scoring scale that makes possible assigning points to any option under consideration, be it a project, an initiative, a field of knowledge. This three-leg architecture was not indicated to LLM, but it emerged out its reasoning process and/or its review of information.
The overall list of decision criteria seems to be all-encompassing in comparison to those used by a variety of national and international agencies for the same purpose. The initial prompts to DeepSeek mentioned relevance for the needs of Latin America, as well as consistency with current trends in scientific research and technological development. It also highlighted the need to be vigilant so that eventual decisions made at the regional level would not duplicate programs or funding already in place in single countries as well as the importance of assessing links of eventual scientific priorities with leading or future economic sectors. The LLM response incorporated these criteria, assigned them the highest weights, but appropriately added feasibility, equity concerns and still others to the mix.
Other key aspects of the suggested framework are worth highlighting:
i) Suggesting specific sources for collecting data relevant to complete the prioritization exercise;
ii) stressing that the exercise must be understood as complementary to existing national and international efforts in Latin America, and suggesting specific tools for minimizing overlaps;
iii) Connecting with the SDG framework;
iv) having into account the possibility of linking to private sector investment in R&D (including specific suggestions about how to do that);
v) emphasizing the need for stakeholder validation, therefore opening the framework to be enriched and confirmed by governments, industry and academia; and last, but not least,
vi) the use of examples (drawn from Latin American contexts) to better illustrate key features of the proposed prioritization framework.
All these features make the product of our interaction with the LLM a convincing tool in which specialized knowledge of science and technology in Latin America and international experience with priority-setting in R& D converge.
The final touch
Next, looking at the first table above, containing the decision criteria, the issue of excessive complexity and therefore eventual difficulty of its application emerged. Thus, for the sake of practicality, it seemed desirable to develop a streamlined table of criteria and weights that, by merging the main considerations behind the full list of criteria previously shown, would become easier to understand and to apply in practice, without a significant loss of comprehensiveness. Prompting DeepSeekR1 to accomplish the task of streamlining the previous list produced the set of merged criteria and weight shown in the following table:
To further streamline, you could:
1 Provide evaluators with a pre-populated rubric (e.g., “Score Global Trends from 1-10 based on WIPO’s 2024 Tech Trends report”).
2 Use a traffic-light dashboard to visualize overlaps with existing initiatives.
After the precedent output, DeepSeek was told of a concern: The framework seemed to have lost an explicit connection with its original intent, namely, to be applied to the selection of R&D priorities. No clear mention of compatibility with research trends or quality was visible. The response led to the final version of the prioritization criteria:
The LLM has now settled, through recurrent human interaction, on a refined and short list of decision criteria and relative weights, plus an implementation protocol full of useful and tailor-made instructions (for Latin America), so we have produced a prioritization framework ready to be applied when it comes to making Latin America-wide decisions about investments in R&D. Should decision-makers prefer the more detailed list of 10 criteria shown earlier, they would of course be free to adopt it.
The pair of brief case studies elaborated as illustration of the LLM output are suggestive of how this would work in practice. They look cogent as an exercise:
Case Study: AI for Healthcare
• Global Trends: 9/10 (AI-drug discovery is a global priority).
• Regional Impact: 8/10 (addresses tropical diseases and rural health gaps).
• Research Excellence: 8/10 (builds on Brazil’s Fiocruz, Mexico’s UNAM research).
• Global Trends (7/10): CRISPR is a global priority if less "hot" than AI/quantum.
• Research Excellence (7/10): Builds on EMBRAPA’s (Brazil) agritech research but needs talent upgrades.
• Total: 8.15/10 → Top Priority.
The closing step in this process is, going back to Step One in the suggested framework (see above), to “Identify candidate fields” to be assessed in their priority, producing and ordered list of R&D fields of study on which international cooperation, coordination and funding should converge.
But this is the issue to be addressed in the second article of this series.
I will leave you here with a short poll about what we, along AI, accomplished, in this first installment of the series. You can also offer comments on this article.
More specifically, in this case the question is being considered as a part of the process leading to the eventual creation -estimated by mid-2026- of INKA (Initiative for Knowledge in the Americas), an international initiative aimed at developing a comprehensive regional instrument for advancing science, technology and innovation at unprecedented scale. Every country in Latin America has and will continue to have its own set of scientific priorities. But INKA will be built on the assumption that there are commonalities among countries, and that some of the tasks required to achieve effective results and global prominence will be facilitated by international collaboration. For more details, click, INKA.
