A short AI tutoring pilot in Nigeria's Edo State produced unusually large learning gains, while also showing why the way AI is introduced in education matters as much as the tool itself.
What the pilot found
The program used Microsoft Copilot as a virtual tutor for students in Edo State. Students worked with the AI system twice weekly, with the project focused on English and digital skills.
After just six weeks, participating students scored 0.3 standard deviations higher on pen-and-paper tests than students in the control group. The source describes that result as equivalent to nearly two years of additional learning.
The improvement was especially notable because many of the students had never used a computer before joining the program. That means they were not only studying English and digital skills; they were also learning how to work with basic computer tools at the same time.
Girls showed the largest improvements. According to the source, girls had initially trailed boys, but their gains effectively closed the performance gap.
Why teacher guidance matters
The pilot was not simply a story about students being handed an AI system and left to learn alone. Teacher involvement appears to have been central to the outcome.
Wharton professor Ethan Mollick points to that context when discussing the results. The project was narrow in scope and lasted a limited time, and the role of teachers likely helped shape the gains.
That distinction is important because AI tutoring can create problems when it is used without guidance. The source notes that research has shown AI tutoring without teacher support can harm learning by giving students false confidence.
In practical terms, that means the lesson is not that AI replaces classroom structure. The stronger conclusion is that AI tutoring may be useful when it is built into learning with adults who can guide, correct, and frame the work.
The gains went beyond the target subjects
The program's effects were not limited to English and digital skills. Students also performed better in other subjects during final exams months after the program ended.
Attendance appeared to matter. The source says students who attended more sessions saw greater gains, which suggests that repeated participation was connected to stronger outcomes.
This broader effect matters because it points to more than narrow test preparation. The source suggests the program may have helped students build wider learning skills, not only knowledge tied to the immediate subjects in the pilot.
That is one reason the results have drawn attention. A six-week intervention is short, but the reported follow-on performance months later raises a larger question: whether guided AI tutoring can help students learn how to learn more effectively.
Promise, caution, and access
Microsoft AI CEO Mustafa Suleyman views the findings as evidence that AI could help democratize education and open new opportunities. He also frames the work as early evidence, not a finished answer.
"Yes, this is just a single pilot program, but the team's findings get me so excited about what's possible," Suleyman writes.
That caution is essential. The source makes clear that questions remain about long-term effects, how student-AI interactions may change, and what downsides could emerge. The research team notes that expansion requires more research into those questions.
AI tools in education carry both opportunity and risk. They can offer personalized support and may help struggling students, but unequal access to these technologies could widen existing educational gaps.
The implementation challenge is also institutional. Many teachers recognize AI's potential while working in rigid educational systems that leave little room for innovation. The source warns that this tension can push educators toward resistance instead of integration, a response that could ultimately hurt the students education systems are meant to support.
The Edo State pilot therefore offers a useful but careful signal. AI tutoring with Microsoft Copilot was linked to large gains in a short period, including for students who were new to computers. But the evidence also points to the conditions around the technology: teacher guidance, access, attendance, and further research all shape whether AI becomes a learning aid or a new source of inequality.