At AI for Good – Education at Scale, Miriam Chickering, CEO of The Frank Foundation, shared a compelling vision for how artificial intelligence can help democratize education without replacing human judgment, local expertise, or ethical responsibility.
Her presentation explored a fundamental challenge: creating high-quality educational content is not enough. For education to have meaningful, lasting impact, the systems around that content must also be ready to deliver it.
When Good Educational Content Is Not Enough
Miriam opened her presentation with two experiences from the same country.
In one case, thousands of textbooks were burned because they were considered inappropriate for the students’ ages. Later, she found another collection of books sitting unused and gathering mold because teachers had not been trained to use them.
The books themselves were not necessarily the problem. The infrastructure required to implement them effectively was missing.
As Miriam explained:
“Technology lives in the cloud, but human beings live on the ground.”
This principle shapes how The Frank Foundation approaches artificial intelligence and education. AI can support faster, cleaner, and more affordable content production, but education ultimately takes place within real communities, institutions, and cultural contexts.
Technology can only create meaningful impact when it responds to those realities.
Four Approaches to Responsible AI
The Frank Foundation’s mission is to improve the health of humanity and the planet by democratizing education. Its learners include health workers, nurses, physicians, and teachers from every country.
Through this work, the Foundation has developed four complementary approaches to using artificial intelligence responsibly:
- Atlas
- Ethos
- The Intelligent Textbook Machine
- The Adaptable Infrastructure Framework
Together, these approaches address four essential questions:
- How should people and artificial intelligence work together?
- How do we determine what should and should not be done?
- How can we produce better educational materials responsibly?
- How can educational initiatives take root within communities?
Atlas: Placing AI Inside the Human Loop
Atlas is a workflow that defines where artificial intelligence can assist, who must review and approve its outputs, and what should happen when something goes wrong.
Discussions about AI often focus on whether organizations should keep a “human in the loop” or remove people from the process to achieve greater scale. The Frank Foundation proposes a different model: AI should operate within a human-led system.
AI can propose, draft, analyze, and accelerate. People must establish the purpose, evaluate the results, and remain accountable for the final decisions.
The Foundation applied this model through the STEPS Initiative. The project initially planned to create 12 lessons, but participating teachers requested complete textbooks.
In response, the team:
- Created 52 textbooks
- Trained 160 teachers
- Reached 6,000 students across three countries
- Completed the work for approximately $100,000 per country
The materials were produced years faster and at a significantly lower cost than would normally be expected. However, speed did not replace quality control. Teachers and national partners reviewed, revised, and approved the content throughout the process.
Ethos: Adapting Ethics to Each Context
Ethos is an adaptable ethical model that helps determine what should be done, what should not be done, and which safeguards are necessary for each use case.
Speed can be critical during a crisis. However, speed without ethical oversight can harm the same communities an intervention is intended to support.
An inadequate implementation might ask:
- Did we translate the material?
- Did we distribute it?
- How many people did we reach?
A responsible implementation asks deeper questions:
- Could people understand the material?
- Was it safe?
- Was it fair?
- Was it appropriate for the local context?
- Were the affected communities involved in the process?
The Foundation’s work with UNICEF demonstrated the importance of these questions. An initial grant covered 50 lessons in 10 languages, but UNICEF needed a much larger educational library.
The project ultimately produced 367 lessons in 16 languages, reaching more than 100,000 people.
The process also generated important lessons. An initial translation into standard Swahili was tested with a small group, and participants explained that it would not adequately reach their communities. The team responded by using a specialized language model to adapt the materials into Congolese Swahili.
When conflict made in-person training in Goma impossible, the team redesigned the educational experience for basic mobile phones. This allowed people to access answers and essential information directly from their devices.
The objective was not simply to translate and distribute content faster. It was to make the education understandable, accessible, and useful to the people it was intended to serve.
The Intelligent Textbook Machine: Responsible Educational Production
The third solution is the Intelligent Textbook Machine, a software product designed to support the responsible production of educational materials.
The process begins with a country’s own learning objectives. AI can propose content, identify connections between objectives and lessons, and reveal potential gaps. However, subject-matter experts review the sources and decide whether each output should be approved, revised, or rejected.
The tool also supports localization.
A child in the Democratic Republic of the Congo should not have to learn about ecosystems exclusively through examples involving foxes and oak trees when their environment includes rainforests and gorillas.
Localization is more than translating words. It means adapting examples, references, and learning experiences to the realities of the people using the material.
Artificial intelligence can make this process faster, but it cannot accept responsibility for the outcome.
A tool can be powerful. A tool cannot be accountable.
Building Infrastructure That Allows Education to Take Root
Producing educational resources is only one part of the challenge. Ministries, teachers, local organizations, and implementation partners are still needed to deliver those resources to the people who need them.
During an Ebola crisis in the Democratic Republic of the Congo, The Frank Foundation produced 90 educational resources for the World Health Organization in fewer than three weeks.
Producing the resources quickly was an important achievement, but implementing them required ministries and hundreds of partners.
The Adaptable Infrastructure Framework addresses what Miriam described as the implementation gap: the distance between a valuable resource and the ability of institutions to deliver it effectively.
AI can perform some of the hidden work involved in production, translation, and adaptation. However, producing more resources also creates more work related to review, training, distribution, and implementation.
Human labor does not disappear. It changes shape.
Educational initiatives take root when the people closest to the work have the authority, resources, and capacity to carry it forward.
A Different Measure of Good AI
Miriam concluded by asking an essential question:
What makes artificial intelligence good?
A tool can be well designed and still be used in harmful ways. Its impact depends on the judgment of the people using it and the systems in which it operates.
Responsible artificial intelligence should:
- Serve human flourishing
- Preserve human agency
- Respect human limits
- Remain accountable to affected communities
- Strengthen local capacity rather than replace it
The presentation also challenged the idea that every application of artificial intelligence is inevitable. The ability to introduce an AI tool into a classroom, use a child’s data, or replace a worker does not automatically provide ethical permission to do so.
These are human decisions, and responsibility for them cannot be delegated to technology.
Success should therefore be measured not only by what artificial intelligence can produce, but by what people become through its use.
Does a teacher become more capable?
Does a nurse become better equipped to serve their community?
Do local institutions become stronger and more independent?
When the answer is yes, artificial intelligence can become a genuine force for good.
Miriam closed her presentation with a principle that captures The Frank Foundation’s approach:
“AI is for humans. Humans are not for AI.”
Education at scale is not simply about producing more content, reaching more people, or moving faster. It is about using technology responsibly to strengthen human capabilities, respond to local realities, and create educational systems that serve people and communities.