The AI-Powered Nonprofits Reimagining Education

Originally published in Stanford Social Innovation Review by Kevin Barenblat and Brooke James.

Have you ever heard of Kalamang? If not, you’re not alone. It’s an endangered language spoken by only 130 people and it has virtually no online presence. That’s one reason why, in 2023, it became the perfect subject for an interesting question: Could an artificial intelligence large language model (LLM) learn and translate a new language from scratch after training on only a single book?

The author of that book—Kalamang’s first-ever grammar book—was University of Oslo PhD student Eline Visser. She joined researchers from Google, Stanford University, and the University of Oxford to dig into this question. The team asked ChatGPT-4 to read Visser’s book and then translate 500 Kalamang sentences into English. A human expert outperformed the LLM in accuracy and natural flow—though it took that person 10 hours to review the grammar book and then several weeks for translation. However, on a recent retest with Google’s new Gemini 1.5, the researchers found the AI actually beat out the human translation. What does this mean? Well, if AI can almost instantly digest and then translate a completely new language, there are endless opportunities for how AI might translate the world’s knowledge into rare languages, reaching people who might never have otherwise had a formal education.

AI in education has only just arrived, but it’s already being harnessed in powerful ways. According to Forbes Advisor, 60 percent of educators use AI in their classrooms—and research shows it’s largely working as intended. Per a 2023 pulse survey from aiEDU, more than half of educators say the impacts of generative AI are either “somewhat positive” or “very positive.” Still, there’s a stigma to overcome, largely due to the fear of students using LLMs and other models to cheat.

This is a real concern, one that’s led to teachers writing up behavioral contracts about AI use and deploying AI-detection software to check student work. However, when done right, AI products augment and support, rather than replace, the work of students and teachers. AI can support teachers so they don’t feel spread so thin in a job where they’re asked to do as much as possible with few resources. AI can help students learn how to find solutions themselves rather than presenting students with easy answers, requiring students to show their work and walking them through step-by-step learning.

According to the Walton Family Foundation, 81 percent of teachers say AI has had a positive impact on education and 65 percent believe AI will be key to the future success of students—and there are nonprofits proving those teachers right. AI is being used in exciting ways to bridge educational divides, and AI-powered nonprofits (APNs) are creating a roadmap for what the future of education may hold.

Major Themes: How Education-Focused APNs Are Leveraging AI

Through our research, Fast Forward has surfaced five major themes of nonprofit use cases for AI in education:

1. Strengthening early childhood education in low-income communities: AI can tailor lesson content and pace to the needs of individual students, addressing their specific strengths and areas for improvement. AI can even help identify developmental delays, different learning styles, or learning disabilities, allowing for timely intervention and support. Identifying developmental delays and educational struggles is especially vital in low-income communities, where needs may be more likely to go undiagnosed and be unsupported.
2. Improving grammar, writing, and reading comprehension: AI can offer real-time feedback and personalized instruction. Natural language processing algorithms can analyze students' written work, identify grammatical errors, suggest improvements, and provide explanations to help them understand their mistakes. For reading comprehension, AI can recommend texts at appropriate difficulty levels, generate questions that enhance critical thinking, and track student progress over time.
3. Bringing quality education to disconnected communities: One of AI’s greatest powers is in reaching a vast number of students, crossing barriers around language and even internet connectivity. As we explain in our case study of Learning Equality below, edtech devices that aren't connected to the Internet can provide access to vast amounts of interactive learning materials, simulations, and exercises, and will soon also integrate AI-powered support tools that work completely offline. Furthermore, AI can translate material into localized languages, making education more accessible and culturally relevant.
4. Enhancing communication between teachers and parents: AI can automate routine parent updates and provide data-driven analysis about student progress, attendance, and behavior, even offering suggestions for parental involvement. What’s more, AI can bridge the gap between teachers and parents who may not speak the same language.
5. Empowering teachers: According to Gallup, 44 percent of US K-12 educators and 35 percent of college and university workers "always" or "very often" feel burned out. AI can aid overworked teachers by generating ideas for class exercises or automating routine tasks such as grading, attendance tracking, and lesson planning. The less time teachers spend on those tasks, the more they’re free to focus on classroom engagement and address complex student needs.

Three Case Studies of Education APNs

Rocket Learning: Enhancing Foundational Learning in India

Rocket Learning is deeply rooted in the personal experiences of its cofounders. One of those founders is Siddhant Sachdeva, whose mother ran an affordable school for low-income families in India. Sachdeva saw how important early childhood education could be for those young students. However, many parents didn’t participate in their children’s learning, unaware of the importance of their involvement in the education process. This was part of a larger crisis: 40 million children in India between the ages of 3 and 6 were not reaching their potential due to educational gaps.

Rocket Learning was born with the mission to equip parents and Anganwadi (rural child care center) workers with the tools to support children’s early learning journeys. By using generative AI, machine learning, predictive modeling, voice assistants, and natural language processing in its AI coach, Rocket Learning has scaled quickly. Because nearly half of India’s population has a smartphone and most families use WhatsApp, Rocket Learning decided they’d use the app to disseminate lessons and communicate with teachers and parents.

Screenshot courtesy of Rocket Learning.

Rocket Learning’s AI creates localized content in key Indian languages. It also automates grading by recognizing handwriting, uses predictive modeling to identify and retain at-risk students, and provides analytics to track each child's progress and identify those needing extra support. The content is designed to be accessible and engaging, even for parents with limited or no education who are teaching their own children. The provided lessons utilize readily available household materials and the content is bite-sized, audio-visual, and gamified to keep children engaged. File sizes are limited to under 5 MB to ensure accessibility for its users. For Anganwadi workers, the platform provides tools and training and also nudges parents to stay engaged.

Rocket Learning was part of Fast Forward’s 2021 Startup Accelerator and Growth Accelerator and, last year, they were awarded the $1.5 million AI for the Global Goals prize from Google.org. Since its inception, Rocket Learning has impacted 2.5 million children, helping them reach the top third of their class. What’s more, 75 percent of the children using Rocket Learning reach their developmental milestones by age six, compared to the national average of 57 percent.

To co-founder Vishal Sunil, Rocket Learning’s success wouldn’t have been possible without AI. "When founders use AI for good, they unleash its capacity to make a deep and real impact on society, fundamentally changing life trajectories for the most vulnerable,” he said. “To bridge educational gaps globally, founders must think big in terms of scale and use AI and technology responsibly as enablers in the process, while keeping communities at the core of their design.”

In our landscape analysis of APNs, Rocket Learning is an example of using AI for monitoring (continuously collecting and analyzing data, providing real-time insights and alerts), translating (facilitating effective communication and making services accessible to more audiences), and assessing (assessing user input and making relevant recommendations).

Khan Academy’s Khanmigo: Enabling 1:1 Tutoring and Supporting Teachers

Romeo and Juliet is one of the most commonly assigned books for high school English. So it’s no surprise that Aleksandar Tatum, a 16-year-old student at Hobart High School in Indiana, had to read the play and then write a story that changed some of the scenes. As he worked through this assignment, Tatum asked his tutor to describe Act 1 Scene 3, which he couldn’t quite remember. “I’m here to help you create a story,” the tutor said, but urged him to flip back through the pages to review that scene again. Except this wasn’t an ordinary tutor. It was an AI learning coach: Khanmigo. “Khanmigo learns the same way I do,” Tatum said. “It’s almost taking the steps with you.”

Photo courtesy of Khan Academy.

Khanmigo was built as part of Khan Academy, a nonprofit created in 2008 to provide online educational tools. Today, Khan Academy has 165 million registered users in more than 190 countries. As a student tutor, Khanmigo enables interactive learning by allowing students to debate, ask questions, and collaborate. The chatbot provides students with support without giving direct answers, guiding students through problems with prompts, questions, and outlines. Because immediate feedback is essential for learning, especially when acquiring new concepts, Khanmigo acts as a Socratic tutor, prompting students with questions and suggestions for how to think through problems step by step. To curb screen time, Khan Academy limits the amount of time students can have with the AI each day. Chat histories and activities are visible to parents, guardians, and teachers, and the platform alerts them about any inappropriate interactions.

Teachers have their own version of an AI teaching assistant, which can reduce prep time by building out lesson plans, quizzes, student groups, rubrics, and other materials. In May 2024, Khan Academy partnered with Microsoft, which donated access to its Azure AI-optimized infrastructure and enabled free access to Khanmigo for all US educators. Khan Academy plans to expand the tools to teachers in other English-speaking countries and has a pilot program underway in Brazil.

While these developments are exciting, Khan Academy founder Sal Khan has encouraged AI optimists to approach developments carefully. “I think all of us together have to fight like hell to make sure that we put [in] the guardrails. We put in—when the problems arise—reasonable regulations,” he said in a 2023 TED Talk. “But we fight like hell for the positive use cases [...] The most powerful use case, and perhaps the most poetic use case, is if AI, artificial intelligence, can be used to enhance [...] human intelligence, human potential, and human purpose.”

In our landscape analysis of APNs, Khanmigo is a prime example of using AI as a coach, dynamic AI that emulates human interaction and acts as an expert in specific fields.

Learning Equality: Taking the Online Learning Revolution Offline

In 2012, Jamie Alexandre, then a PhD student in cognitive science, spent a summer interning with Khan Academy. During that time, he reflected on when he volunteered at a Tibetan refugee center in Darjeeling. There, he encountered a lack of internet connection, which was a problem for two-thirds of the world. Alexandre wondered: How could online education be brought offline? The launch of the affordable Raspberry Pi computer sparked an idea. Alexandre and a fellow Khan Academy intern developed a prototype that enabled offline access to Khan Academy videos and exercises using the Raspberry Pi. This eventually led to the incorporation of Learning Equality as a tech nonprofit in April 2013.

Kolibri, the second generation of their platform, is an ecosystem of products designed for learning with devices that aren’t connected to the internet. Kolibri organizes and adapts nearly 200,000 open resources to meet specific learning objectives. Their offline Kolibri Learning Platform runs on many devices and systems, utilizing existing, low-cost hardware. Offline users can download Kolibri directly from Learning Equality’s website. Alternatively, organizations can distribute content via USB drives, pre-loaded devices, or peer-to-peer syncing.

Over the past decade, Learning Equality's offline edtech tools have helped more than 10 million learners in over 220 countries and territories achieve their educational goals. One of those students is Mudiyatu from Cameroon. As the first in her community to pass the baccalaureate exam, Mudiyatu wanted to attend nursing school, but didn’t have a strong enough science background. Kolibri gave her access to Khan Academy's science and math content in French, helping her to pass the nursing school exam. “There were topics on the exam that I could only answer because of what I studied on Kolibri,” she said.

Now, Learning Equality is exploring how AI can accelerate curricular alignment and lesson development. Currently, with Learning Equality’s online curricular tool, Kolibri Studio, users upload their own materials and mix and match them with other Kolibri Library resources. They can choose to align the output to their specific educational needs and their country’s curricular standards. In collaboration with UNHCR, Learning Equality is developing a semi-automated curricular alignment tool that leverages generative AI to make this process less time-consuming. Earlier this year, their AI automatically aligned 12,000 learning resources to more than 2,000 distinct learning objectives for Ugandan students. They estimated that this would have previously taken months of work and cost hundreds of thousands of dollars. With AI, it took just a few days and a couple hundred dollars of application programming interface (API) credit, plus the cost of paying experts to review the content. These materials are already being used by Ugandan teachers and youth in communities without internet connection.

Chart courtesy of Learning Equality.

In Alexandre’s view, equity needs to become a central focus for edtech founders so that technology can bridge, rather than exacerbate, existing divides. “It’s easy to slip into a pattern of serving the needs of those whom our solutions can most easily reach. But a true focus on equity means finding ways to listen to and support the needs of those who are in fact most difficult to reach,” Alexandre said. “That focus is something that makes this work so hard, but also what makes it so important.”

In the Fast Forward landscape analysis of APNs, Learning Equality is an example of using AI to organize, aligning a large corpora of data and ensuring that relevant information is easily accessible.

The Role of AI in Better, More Equitable Learning

AI’s potential to revolutionize education is being actualized by forward-thinking nonprofits around the globe. From enhancing early childhood education in low-income communities to providing offline learning solutions in disconnected regions, APNs are using AI to close educational gaps. These educational APNs are not only addressing current challenges, but setting the stage for a more equitable future and addressing challenges faced by students and educators in creative ways.

For example, just as AI can instantly translate a rare language like Kalamang, nonprofits can employ AI to translate educational content into a vast number of languages. AI can also be used to rework material into different formats for those with different learning styles or abilities. This, of course, is about more than just the act of translation. It’s about democratizing knowledge so that quality education can be accessible to all, regardless of geographic, socioeconomic, or other barriers to learning.

AI’s evolving role promises to reshape education for learners, teachers, and parents. By embracing these innovations with the right safeguards, policies, and approaches, we can ensure that quality education is accessible to all and supports, rather than replaces, the human elements of education. The journey is just beginning. With forward-thinking APNs leading the way, the impact promises to be profound and far-reaching.