Here are 3 lessons for using AI to aid learning in low-resource settings

23 mar. 2019 |

Technology has enormous potential to democratise learning: promising access to education for children in every corner of the globe.

According to the UN, 617 million youth worldwide lack basic maths and literacy skills and 57 million primary age children remain out of school.

But can artificial intelligence really help a generation of young learners in low-income, low-resource countries gain literacy?

While AI-based systems have shown great success in learning gains, most development and evaluation of such systems has been limited to WEIRD (Western, Educated, Industrialised, Rich, Democratic) places.

These technologies have the potential for enormous impact if they can serve the next billion learners in the resource-poor Global South.

But, according to Amy Ogan, Assistant Professor and educational technologist at Carnegie Mellon University, EdTech is not a silver bullet or a one-size-fits-all solution to attaining the UN’s Sustainable Development Goal 4.

Speaking at the Global Education and Skills Forum 2019, Ogan outlined the following barriers that need to be addressed in order to harness the power of advanced EdTech for all learners.


Lesson 1: Understand the context and design to fit it

This goes beyond translating EdTech programmes into students’ home languages and even the differences between what a long-division sum looks like in different countries.

Ogan says it’s about learner autonomy and motivation and how AI deals with those. She gives the example of presenting a student with three problems to choose from - versus one problem that’s been chosen by their family.

“Most learning science will tell you students will opt to choose for themselves, and in European and North American settings it works.”

But in Confucian-heritage culture, she says, which includes such families as China, Vietnam, Singapore, Korea and Japan, family choice significantly improves students’ learning.

AI-based systems modelled on American learners can predict with 76% accuracy the help they need and their learning outcomes, but when used in Chile, where there is three times as much classroom collaboration, the AI doesn’t work.

“Much of the help-seeking is happening offline, so the models driving AI get it wrong,” says Ogan, stressing the need to work with learning science principles built on students in their context.


Lesson 2: Take physical infrastructure into account

In some settings, this will involve the availability of electricity and other infrastructure - which learners might not have access to on a daily basis.

So instead of equipping all children with a tablet, for example, it’s better to work with the tech they already have at home.

In much of Africa, the number of people with mobile phones outnumbers the number of people with electricity, so in Côte d’Ivoire, an innovative voice-based AI system, called Allo Alphabet, gives children lessons through the family mobile.

“There are unexpected situations one must account for to provide technology on a regular basis, so physical infrastructure must be planned for and investigated ahead of time,” says Ogan.

She uses the example of an innovative idea in Tanzania, where Wi-Fi was so patchy, motorcycle drivers were hired to take learning around with them, on USB sticks.


Lesson 3: The importance of human infrastructure

The human involvement of teachers is a critical part of the learning environment for these technologies to succeed, says Ogan.

Teachers are able to provide students with the support they need above and beyond technology, such as motivation, but they are not considered in the design, so there is no interface between them.

Peers and siblings are also a critical part of the support structure, for example, fellow students can explain maths problems in a way algorithms can’t, so by building peer support into technology, we harness it.

In certain low-resource contexts, parents don’t necessarily have the skills being taught by the system, so is there value in having parental support when parents don’t know the content?

Yes, says Ogan. In the Côte d’Ivoire project, 58% of children received support from their parents when using the Allo Alphabet system, even if their parents were illiterate themselves. These children answered 10% more of the questions correctly.  

As one parent commented to Ogan: “Everybody needs to bring their grain of salt” to help children learn.

She adds: “In low-resource settings, sometimes we forget the resources that are there.”

The Next Billion Prize supports EdTech entrepreneurs trying to find solutions in resource-poor settings. This year’s prize will be awarded at the end of the 2019 Global Education and Skills Forum.