AI Education

Equity in LMICs

What is the purpose

In the last section we saw how the lack of teacher support contributes to low-quality education in classrooms. It is tempting to jump straight into the excitement of AI tools and imagine their possibilities: producing useful insights for decision-making, strengthening teaching, and boosting productivity. For schools in low- and middle-income countries (LMICs), this could translate into real opportunities such as relieving overstretched teachers, supporting students who struggle with basic literacy, and compensating for the outdated textbooks and learning materials.

But these benefits are not automatic. Most AI tools are not built with everyone in mind. For example:

Language bias privileges English and a handful of dominant languages while neglecting African, Asian, and Indigenous languages.
Cultural mismatch means examples and explanations are designed for U.S. or European classrooms, not for local realities.
Infrastructure barriers leave out schools without stable internet, reliable devices, or the funds to pay for subscriptions.

Used without care, AI could risk deepening the very gaps it claims to solve. In other words, AI has the potential to advance equity in LMICs classrooms but only if we are clear about how it is built, who it is designed for, and what barriers stand in the way of fair access.

How to use this page

This Section Has Two Parts

How AI Systems Work : A step by step overview of how AI systems are built, to help you understand where inequities can emerge in the process.
Barriers to AI in Education for LMICs : A deeper look at the specific barriers, divided into two areas: inequitable data and inequitable access.

How AI is Built and how it impacts equitable education

What is AI

AI is not a single technology but a broad field of study focused on building systems that can simulate aspects of human intelligence. Over the years, researchers have tried many approaches to make computers “intelligent.” The most common approach used today is called machine learning. In machine learning, computers are given instructions on how to process large amounts of data, look for patterns, and then use those patterns to make predictions or generate outputs.

How AI is Built Step by Step (Language Model Example)

Step 1 : Collecting Data

In the next few steps we will use a language model as an example. Language models are systems like ChatGPT that generate text based on patterns they have learned from huge amounts of text data. But where does that data come from?

The answer is: almost anywhere text exists. It can include internet pages, online books, social media posts, and especially user interactions. For example, ChatGPT (and similar tools) often analyze and improve from the conversations people have with them.

The important point is that the data that goes in shapes the answers that come out. At present, most of the data used to train large language models is in English and reflects content from Western, high-income countries.

Step 2 : Analyzing the Data

After the data is collected, the language model is trained to recognize patterns in text. It does this by repeatedly going through the data and trying to predict the next word in a sentence. Each time it makes a guess, it compares its prediction with the actual word and adjusts its internal parameters to do better next time. Over billions of these predictions, the model gradually becomes more accurate at predicting the next likely word in its training data.

Step 3 : Evaluating the results

After the language model becomes more accurate at predicting the next likely word based on its training data, the next step is evaluation. This means checking whether the model’s answers are actually useful, safe, and relevant. Evaluation can include both tests on standard tasks and feedback from real people, who compare responses and judge which ones are more helpful or appropriate. That feedback is then used to adjust the system, making it more reliable for real-world use.

Step 4 : How it Produces answer

In the end, what we have is a model that generates answers by predicting one word at a time until a full response is formed. It may sound simple, but behind it are the steps of collecting data, analyzing patterns, and making predictions, which are the foundation of any AI system. However, predicting each word in real time is extremely computationally demanding. That is why these models usually run on powerful internet based services or on specialized devices with large amounts of memory and processing power.

Why does this mean in Equity?

From these steps, we see that everything begins with the data. If most of the data is Western in language and culture, the results produced may not reflect the realities or needs of people in low and middle income countries. The language may not match what they speak, and the cultural references may not fit their context. On top of that, using these models requires significant computing power and internet infrastructure, which are not equally available everywhere. This means many communities face disadvantages both in how useful the outputs are and in their ability to access the tools in the first place.

Barriers to Equitable AI in Education

Inequitable Data

Most training data comes from Western and English sources, leaving out many local voices. See how new projects are building models with their own languages and knowledge.

Learn More

Inequitable Access

AI often requires strong internet, powerful devices, and money, which limits who can use it. See how smaller models are being designed to run anywhere at lower cost.

Learn More

Page updated

Google Sites

Report abuse