Can AI Help Struggling Students?

MILLIONS OF STUDENTS ARE LAGGING BEHIND. CAN AI HELP?

If a student hasn't mastered foundational skills like reading, it is hard for them to build other skills. Most education systems expect students to learn to read by grades 1 or 2. By grade 3, students need to read to access their curriculum. Children who cannot read by grade 3 fall behind and struggle to catch up, often irreparably.

For millions, schooling is not producing key foundational skills. Around 70 percent of 10-year-olds in low- and middle-income countries (LMICs) are unable to read and understand a simple text. The pace of classroom instruction is often determined by the need to cover an overly ambitious curriculum rather than by the pace of student learning. This means teachers may ignore students who are falling behind. Nearly 48 percent of teachers, from across nine LMICs, believe that there is little they can do to help a student learn if he or she comes unprepared from previous grades. In nearly all countries, it is poor students who are most likely to lag behind.

Theoretically, artificial intelligence (AI) can support lagging students. But does this happen in practice? In fact, AI is already transforming education at an unprecedented pace, offering game-changing opportunities to personalize learning experiences, support teachers in their daily tasks, and optimize education management. And this promise only grows as AI capabilities continue to expand. But alongside the promise, there are tremendous fears and skepticism about AI. So, what does real-world experience and evidence say about the true implications of using AI for struggling students? Below we summarize the latest evidence.

THERE IS EVIDENCE THAT AI CAN HELP!

A pilot in Nigeria used generative artificial intelligence (AI) to support learning through an after-school program. First-year senior secondary students who were randomly assigned to participate in the program significantly outperformed their peers in all areas, including English, which was the main goal of the program. The learning improvements were striking, about 0.3 standard deviations. This is equivalent to nearly two years of typical learning in just six weeks.

Hence there is proof of concept that generative AI, when implemented thoughtfully, can function effectively as a virtual tutor. The Nigeria study further shows that when properly guided, AI can help students engage more deeply with the material. Instead of asking students to summarize texts, which AI can do instantly, teachers can guide them to critique AI-generated analyses, compare different AI interpretations, or use AI as a brainstorming partner while developing their own ideas.

References: World Bank 2022; Pritchett and Beatty 2012; Sabarwal et al. 2022; World Bank 2018, World Development Report: Learning to Realize Education's Promise; World Bank 2024, AI Revolution in Education; https://blogs.worldbank.org/en/education/From-chalkboards-to-chatbots-Transforming-learning-in-Nigeria

An AI-assisted program in Ecuador helped struggling higher education students close learning gaps in math, achieving significant academic gains. Ecuador's Ministry of Higher Education, with World Bank support, launched an AI-assisted program in math to address learning gaps among tertiary students. After using the platform for 16 consecutive weeks, their mastery of the math curriculum increased from 25 percent to 68.7 percent, the equivalent of a full year of schooling. What's more, these gains were achieved at the cost of just $18 per student.

Many education systems are exploring AI-powered tutoring systems that adapt to individual needs and provide real-time feedback. Singapore, for example, is aiming to mainstream the use of AI to support personalized education and help teachers better address the needs of students, especially those with special needs. Similarly, South Korea is using AI to personalize homework and learning activities based on individual educational levels and learning behaviors, while Finland is integrating AI into its national education system with a strong emphasis on ethics and equity. In Iowa, a new law requires students deemed not proficient, about a quarter of them, to get an individualized reading plan. As part of that effort, the state's education department spent $3 million on an AI-driven personalized tutoring program. When students struggle, a digital avatar intervenes.

Some countries are using AI more for teacher training and administrative support rather than for direct student learning. AI-powered feedback systems, such as TeachFX in the United States, offer targeted professional development by analyzing classroom discourse data and generating pedagogical insights, enabling teachers to refine their instructional practices. A randomized controlled trial with math and science teachers in Utah found that automated feedback can improve teaching practices. Teachers receiving weekly AI-generated feedback on their use of focusing questions, which probe students' thinking and press them to explain their reasoning, increased their use of these techniques by 20 percent in five weeks.

Despite these promising signs, the use of AI in education is still very limited, even in high-income contexts. In the US, only 18 percent of K-12 teachers used AI for teaching by the end of 2023-2024, primarily to adapt instruction and generate learning materials. Most teachers using AI relied on virtual learning platforms, adaptive learning systems, and chatbots. Another study on the use of AI in US education shows that while 97 percent of ed-tech leaders see AI as beneficial, only 35 percent of districts have an AI initiative. The most promising applications are productivity and personalized learning, but concerns include cyberattacks, cyberbullying, false information, and data privacy risks.

BUT THERE ARE SOME RISKS

Mainstreaming AI support in education could leave under-resourced students further behind. For instance, take the case of Peru. In Lima's private schools, students engage directly with AI-powered chatbots. Meanwhile, in rural Cajamarca, many classrooms still lack electricity, let alone internet access. This stark contrast isn't just true for Peru; it can be seen all over the world.

AI-powered tools, such as ChatGPT, also raise concerns about academic integrity. While schools are adapting to the growing use of AI, there are still significant concerns about the effectiveness of AI-detection tools and the potential for false accusations of plagiarism. Moreover, an over-reliance on AI-detection tools may create an atmosphere of mistrust between teachers and students, undermining the role of the teacher-student relationship in the learning process.

The evidence on the effectiveness and risks of AI interventions in school settings is still very limited. While the potential for transformative impact is significant, there is currently a lack of rigorous evidence regarding their effectiveness and scalability in real-world contexts. To ensure responsible adoption and optimal outcomes, these innovations will require further evaluation and validation through carefully designed pilot studies and impact assessments.

HOW SHOULD EDUCATION SYSTEMS THINK ABOUT AI?

AI-driven solutions can positively impact human capability development for students and teachers alike. This area of work aligns perfectly with the Human Capability Initiative's theme of Harnessing the Ways We Learn. This theme emphasizes innovations in learning science that prepare creative, critical, lifelong learners for an ever-changing world.

To do this well, AI should make the student-teacher interaction more effective and not replace it. It is essential to keep the human element at the center. Additionally, AI-driven solutions for education must be context-specific and tailored to the student population and education priorities of each country.

For responsibly integrating AI in education systems, three actions are critical:

1. Skilled and digitally literate teachers. Teacher selection, training, and continuous professional development must equip teachers with the skills to meet evolving demands, including AI and digital literacy.

2. Investment in an enabling digital ecosystem. The most important ingredients are affordable high-speed internet access, appropriate devices, adequate school infrastructure, and robust digital platforms. However, institutions must carefully evaluate their specific context and needs before investing in these solutions.

3. AI curriculum and advancement of research in AI. School curricula should equip students with the skills necessary for an AI-driven world, ensuring alignment with both national priorities and global standards. At the same time, national investment in AI research for education is essential. Collaborating with higher education institutions and the private sector to support AI research and development can drive innovation and enhance AI expertise.

Additional references: https://www.worldbank.org/en/news/feature/2022/02/10/en-ecuador-aprender-matematicas-es-mas-facil-con-inteligencia-artificial-nivelacion-remediacion-academica; Lake 2023; https://www.courthousenews.com/ai-is-a-game-changer-for-students-with-disabilities-schools-are-still-learning-to-harness-it/; https://documents1.worldbank.org/curated/en/099734306182493324/pdf/IDU152823b13109c514ebd19c241a289470b6902.pdf; https://www.rand.org/pubs/research_reports/RRA956-21.html; RAND Corporation; https://www.naesp.org/resource/opportunities-and-risks-in-with-ai-use-in-schools/; https://blogs.worldbank.org/en/latinamerica/inteligencia-artificial-ia-privilegio-para-pocos-oportunidad-para-todos

Shwetlena Sabarwal (Lead Economist, World Bank) and Saloni Bhatia (Consultant, World Bank).