I remember the day I first experienced artificial intelligence (AI) in education. It was the late ’90s, and I was a maths teacher at a school in London described as having “challenging circumstances.”
I looked at my students, eyes glued to their computer screens, ears covered by headphones, all working in silence using an adaptive maths program. While marvelling at the rare moment of tranquillity, I also wondered: If the students were learning from the technology, what was the point of me being there? After all, I was a teacher, and a rather good one (if I do say so myself!).
Using the student learning outcomes data generated by the educational technology (edtech) program, I split the class into two groups based on their competencies: I taught half of them, while the others worked quietly on their computers, and then we switched. I kept using the data generated by the program to inform my teaching, and over time the students’ results improved. Technology in service of the teacher and the students contributed to our maths results being a top 2% value-add in England!
Fast-forward to today: Generative AI technology is mind-blowing, with great potential to be harnessed for teaching and learning. In a world where seven out of 10 children born in a low- and middle-income countries cannot read by age 10, AI could help address the dramatic learning equity gaps. But this work needs careful thought.
I have experienced many edtech pitches over the years, but only a few have impressed me. Most are either too focused on the technology or on reaching large numbers of students, without considering the pedagogy, or they fail to address the challenges and barriers that underserved students face in accessing and using technology, particularly in low- and middle-income contexts.
Furthermore, it is all too rare to see any evidence of impact on learning outcomes. (If you don’t show impact, I will assume there is none!)
A few edtech solutions combine evidence and the best of human expertise and wisdom with the benefits that technology can afford, particularly in data analysis. And with the recent advances in AI, especially in natural language processing, speech recognition, and computer vision, I have seen pitches that have blown me away—they genuinely make me wish I were teaching again.
Here are three problems that students and teachers across the globe face, along with AI-based solutions that could help, if they are implemented with careful thought, informed by data, and focused on equity:
For a glossary of technical terms. See below.
Problem 1: Many students don’t have access to high-quality learning resources that are tailored to their needs, interests, and learning levels.
Students may struggle to learn from their starting point and at their own pace, receive timely and constructive feedback, or struggle to find motivation and support for their learning goals. If they fall behind, they are likely to stay behind and drop out, failing to acquire important life skills.
- Personalized adaptive learning solutions adjust the difficulty, content, and presentation of learning activities based on the student’s performance, preferences, and progress. These platforms have been shown to improve learning outcomes, especially among low-performing students in sub-Saharan Africa and South Asia. Existing products include Mindspark in India, onebillion in Malawi, and EIDU in Kenya.
- AI tutors offer one-on-one instruction and guidance for students, using natural language processing and dialogue systems to simulate human interactions. These tutors have been shown to increase student engagement, confidence, and achievement. Rori in Ghana and Khanmigo in the U.S. are already assisting students in this manner.
These tools need to be able to assess students’ current learning levels and chart the most effective and efficient learning trajectory; ensure that the content aligns with the national curriculum and is relevant and engaging; and, importantly, be sensitive to bias. The issue of how children’s data is gathered and used, and what can be diagnosed from it, requires serious consideration and established guidelines.
Watch: Bill Gates and Tonee Ndungu discuss AI education in Africa
Problem 2: Some teachers may not have the training, knowledge, and experience they need to be effective in their roles.
Up to 40% of teachers in sub-Saharan Africa cannot demonstrate proficiency in the subjects they teach. Teachers may also face challenges such as large class sizes, diverse student needs, high curriculum demands, and administrative responsibilities. They may also lack opportunities for professional development, feedback, and collaboration or may lack the confidence to try out new methodologies in the classroom.
- AI-enabled teacher coaches can help teachers develop and practice their skills in a simulated environment, using natural language processing and computer vision to create realistic scenarios and characters and get feedback. These coaches have been shown to enhance teacher self-efficacy, competence, and performance in high-income countries. Some models use speech recognition software so teachers can record their lessons and get feedback. TeachFX and Loquat Learning are products to watch.
- AI lesson-planning support programs can help teachers create and customize high-quality learning resources, using natural language processing and semantic analysis to match the resources with curriculum, standards, and student needs. These tools have been shown to save teachers time, increase student engagement, and improve learning outcomes in countries including Kenya and India. Programs in use include Teacher.AI in Sierra Leone, EIDU in Kenya, TeleTaleem in Pakistan and Oak National Academy in the UK.
- AI assistants can help teachers automate and streamline administrative tasks such as grading, recording attendance, and reporting, using natural language processing, optical character recognition, and machine learning to process and analyze student work and data. These assistants have been shown to reduce teacher workload, provide instant feedback, and monitor student progress. ConveGenius and Smart Paper in India are both being implemented on a broad scale.
Teachers will only use AI-based tools that help solve problems and make their lives easier, so the tools must be user friendly, relevant, and lead to impact. As with student-facing tools, teacher-facing tools must be accurate, engaging, and unbiased, and they must take into account the evidence on how adults, and teachers in particular, learn and apply that learning in the classroom.
Problem 3: It is expensive and time consuming to develop high-quality educational content, as well as assessment and evaluation tools, in multiple local languages and contexts.
It is often expensive for governments to produce teaching and learning resources such as student textbooks and lesson plans in multiple local languages. Governments and educational systems may also lack the data and insights to make informed decisions and policies for improving educational outcomes and equity. Technology can generate good first drafts for experts to review.
- AI translators can translate educational content into many local languages, using natural language processing and machine translation to produce accurate and fluent translations. These translators can increase access, equity, and inclusion for learners and educators, and they are being tested in countries including South Africa, India, Mali, and Senegal.
- AI assessors can create, deliver, and score assessments in various formats, such as verbal, handwritten, and multiple choice, using natural language processing, speech recognition, and optical character recognition to evaluate student responses and provide feedback. These assessors have been shown to improve reliability, validity, and efficiency of assessments. Automated assessment platforms in use include Wadhwani in India and EGRA-AI in South Africa.
- AI evaluators can analyze and optimize educational outcomes using machine learning and data mining to generate insights, recommendations, and predictions based on student data and performance. These evaluators have been shown to enhance student retention, completion, and success in higher education in countries including the United States, Australia, and Japan.
These emerging tools need careful research, development, and testing. They need to have established benchmarks and include quality assurance mechanisms to ensure that translations are accurate and pedagogical principles are followed. Assessments must be accurate and available and cover enough languages, subjects, and age groups. AI-enabled evaluations in the Global South are at a nascent stage, so new tools need to be tested against human evaluators for accuracy and efficiency.
Principles for guiding AI in global education—and beyond
The opportunity to improve educational outcomes through AI is substantial. But we must all keep asking who is and isn’t benefiting from these technological advances. The Bill & Melinda Gates Foundation has established an AI Ethics and Safety Committee that is working alongside our program teams, including Global Education, to address these issues. As the Global Education team develops strategies and funds solutions that incorporate AI, our team is focusing on the following five principles:
- Equity. AI systems should be designed and implemented with equity and inclusion in mind, ensuring that they do not exacerbate existing inequalities or create new ones.
- Data privacy. AI systems must respect and protect the data privacy and security of students and educators and follow ethical and legal standards and norms.
- Ethics. AI systems for education must adhere to ethical principles and values—such as fairness, justice, dignity, and human rights—in their development and deployment.
- Evidence and accuracy. AI solutions must be evaluated and validated for their quality, and their accuracy and reliability must be assured for different contexts, languages, and domains.
- Impact. AI solutions should be monitored and measured for their impact on educational outcomes, such as learning gains.
Engaging with AI or not is no longer a choice. AI is being used already. The world ignores it at the risk of deepening educational inequities in ways we do not yet understand. So let’s put systems and processes in place to ensure that students from all walks of life can benefit from these innovations.
Let’s also work with teachers to enable their use of this technology so they can get the greatest impact for themselves and their students. We have an opportunity to revisit, on a broad scale, what I did in the ’90s, by asking, “How can I use this technology to serve my students and myself most effectively?” I believe it is the integration of technology into teachers’ work that will have the greatest impact.
Computer vision. A computer’s ability to recognize people and objects in pictures, illustrations, and videos the way humans see and understand images.
Dialogue system. A computer system designed to converse with humans.
Generative AI. A computer system designed to produce something new based on “training” provided by large data sets (like text or images).
Machine learning. A computer system’s ability to learn without being explicitly programmed to do so, typically using a data set such as numbers, photos, or text.
Machine translation. A computer system’s ability translate text from one language to another using machine learning.
Natural language processing. A computer’s ability to process and produce language in the way humans do.
Optical character recognition. A computer’s ability to convert images of text into editable text that a machine can read.
Semantic analysis. A computer’s use of context to make sense of words or phrases that have multiple potential meanings, to enhance the accuracy of natural language processing.
Speech recognition. A computer’s ability to recognize speech and process it into writing.
Read next
AI equity: Ensuring access to AI for all
The first principles guiding our work with AI
