Large language models could democratize data use—but only if the foundations are in place

Around the third century, in Babylon — the modern-day Iraq — when the first use of zero-like symbol arose, it was a breakthrough. But in writing the book Men of Mathematics in 1937, the author Bell argued that almost everything useful that was produced in mathematics before the 7th century has now become accessible knowledge rather than elite expertise. This historical arc of complex breakthroughs eventually becoming foundational knowledge is now playing out at an accelerated pace in the digital age. 

In a world increasingly shaped by data, the ability to analyze and interpret information has become essential. Yet for many in low- and middle-income countries (LMICs), these skills have remained out of reach—constrained by limited training, costly tools, and scarce expertise. Generally, the use of data has remained relatively low in low- and middle-income countries according to statistical performance indicator. That’s not because data are unavailable, but because the capacity to turn it into insight is limited.

Now, Large Language Models (LLMs), a new force is democratizing this landscape. The role of data scientists once dubbed by Harvard Business Review as “the sexiest job of the 21st century” is rapidly evolving due to LLMs. For years, working with data required specialized training in statistics, programming, or data science. Tasks such as cleaning datasets, writing code, or generating visualization needtime and technical expertise. These tasks can now largely be automated. This seismic shift, particularly within data analytics and education, promises to democratize complex data skills and revolutionize how we learn and work. 

^{Source: Digital Progress and Trends Report: Strengthening AI Foundations}

A policymaker can summarize survey data with a prompt. A researcher can draft analytical narratives or explore alternative interpretations of findings. A student can generate code to visualize trends. In this sense, LLMs are not just productivity tools. They are learning accelerators. They help users move from “not knowing where to start” to actively engaging with data. For many, LLMs can transform data from an abstract concept into something quickly usable. 

This frees time for higher-value activities: designing policies, testing assumptions, and making decisions. But this also changes the nature of expertise. The role of data s is evolving from writing every line of code to supervising and validating outputs generated by AI.

The key skill in the age of AI is no longer just technical proficiency. It is also judgment.

Despite this promise, the benefits of LLMs are not evenly distributed. For many in LMICs, structural constraints limit their ability to take full advantage of these tools. Four foundational factors — connectivity, compute, context, and competency — continue to shape what is possible:
 

1. Connectivity and infrastructure

Reliable internet, electricity, and access to digital devices remain uneven. Without these basics, the most advanced tools remain out of reach for large segments of the population.

2. Compute and affordability

Accessing powerful models often requires cloud services or APIs that can be expensive. While smaller models are emerging, affordability and scalability remain concerns, particularly for governments and educational institutions working with limited budgets.

3. Context and local data

LLMs are only as useful as the data they are trained on. Many low- and middle-income countries have rich but underutilized datasets—often fragmented and undigitized without adequate governance. Without stronger data ecosystems, AI tools risk reflecting global biases rather than local realities.

4. Skills and trust

While LLMs lower entry barriers, they do not eliminate the need for skills. Users still need to interpret outputs, identify errors, and recognize bias. In many contexts, limited data literacy can lead to either over-reliance on AI or reluctance to use it altogether.

Avoiding new inequalities

There is a real risk that LLMs could widen, rather than close, existing gaps. Countries and institutions with strong infrastructure, quality data, and skilled workforces will adopt these tools faster and more effectively. Others may lag not because of lack of interest, but because of systemic constraints. This makes policy choices critical. So, how do we harness the potential of LLMs as an equalizer, a proactive and coordinated effort?

First of all, by investing in foundations. This is to expand broadband access, improve energy reliability, and ensure access to affordable digital devices remain essential. These are not just infrastructure investments—they are prerequisites for participation in the data economy.

Second, by expanding access to compute. Policies that reduce the cost of cloud services or support shared infrastructure can help widen access. Exploring “small AI” solutions that run locally on devices may also offer new pathways.

Third, we need to strengthen data ecosystems. Countries need to invest in the digitization, curation, and governance of their own data. High-quality, locally relevant datasets are essential for building AI systems that reflect local needs, languages, and priorities.

Fourth, there is a need to build critical skills. AI and data literacy should go beyond basic familiarity. Users need to understand how to question outputs, assess reliability, and apply statistical reasoning. Training programs at all levels must evolve accordingly.

And last, but not least, we need to align policy and partnerships. National AI strategies, backed by clear implementation plans and partnerships across government, academia, and the private sector, can help coordinate efforts and avoid fragmentation.

The history of technological progress suggests that today’s breakthroughs can become tomorrow’s basics. Just as zero moved from abstraction to foundation, data skills could become more widely accessible than ever before. LLMs are accelerating that transition.

But technology alone will not determine the outcome. Whether these tools reduce or reinforce inequalities will depend on the choices made today—about infrastructure, data, skills, and governance. If those choices are made wisely, LLMs could help unlock a future where more people, not just specialists, can use data to shape decisions, improve services, and drive development.

And that would be a transformation worth aiming for.

^{Further reading: 
Buwule, Robert Stalone, et al. "Data Literacy: A Catalyst for Improving Research Publication Productivity of Kyambogo University Academic Staff." Journal of eScience Librarianship, vol. 12, no. 2, 2023, p. e646.}

^{Tu, Xinming, et al. "What Should Data Science Education Do with Large Language Models?" 6 July 2023, arxiv.org/abs/2307.02792.}

^{Valverde-Rebaza, Jorge, et al. "Advanced Large Language Models and Visualization Tools for Data Analytics Learning." Frontiers in Education, vol. 9, 2024, p. 1418006.}

^{World Bank. Data Diagnostic for Kerala: Action Plan Based on a Rapid Diagnostic of Data Governance in the State of Kerala. The World Bank Group, 2023.} 

^{---. Devising a Strategic Approach to Artificial Intelligence: A Handbook for Policy Makers. The World Bank Group, 2025.}

^{---. Digital Skills: Methodological Guidebook V 2.0 for Preparing Digital Skills Country Action Plans for Higher Education and TVET. The World Bank Group, 2021.}

^{---. Digital Skills Development in EAP: Key Findings and Recommendations from the Country Studies. The World Bank Group, 2023.}

^{---. Ending Learning Poverty and Building Skills: Investing in Education from Early Childhood to Lifelong Learning. The World Bank Group, 2022.} 

^{Zhang, Linjun. "Large Language Models in Practice: A New Companion for Statisticians." IMS Bulletin, 1 Oct. 2025.}

^{The blog post was edited with the help of Copilot.}