AI can lower some barriers to education, but the evidence is thin and it does not reach the largest one: access to a connection. UNESCO's 2026 Global Education Monitoring Report counts 273 million children, adolescents, and youth out of school, and the International Telecommunication Union (ITU) estimates that 2.2 billion people were still offline in 2025. The best controlled evidence for AI so far is a one-year trial of a phone-based math tutor in Ghana, which reported a modest learning gain.
Key takeaways
- UNESCO reports 273 million children, adolescents, and youth out of school, about one in six, a rise for the seventh consecutive year.
- The ITU reports that 2.2 billion people remain offline, most of them in low and middle income countries.
- A Ghana trial of a chat-based math tutor on WhatsApp reported an effect size of 0.37 over eight months, with results covering only the first year.
- In Canada, 2021 data showed 50/10 Mbps coverage of 99.3 percent in urban areas, 59.5 percent in rural and remote areas, and 42.9 percent on First Nations reserves.
- An AI tool can only remove a barrier if the learner already has a device, a connection, and a language the tool handles.
The size of the access problem
Two counts frame the question. UNESCO's 2026 GEM Report summary, dated September 7, 2026, states that 273 million children, adolescents, and youth are out of school, one in six worldwide, and describes the figure as a rise for the seventh consecutive year. The report also notes that progress has slowed across almost every region since 2015, with a sharp slowdown in sub-Saharan Africa.
The second count comes from the ITU. Its Facts and Figures 2025 reports that almost three-quarters of the world's population is online, while 2.2 billion people remain offline, most in low and middle income countries. Gender and urban-rural divides, it says, continue to narrow but endure.
What the strongest trial found
Owen Henkel and colleagues studied a conversational math tutor delivered through WhatsApp in Ghana. About 1,000 students in grades 3 to 9 at 11 schools took part. Schools, not individual students, were assigned to treatment or control. Control students kept their regular math lessons. Treatment students had the same lessons plus two 30-minute sessions a week with the tutor over eight months. The paper reports an effect size of 0.37 on math growth scores (p below 0.001). It runs on basic phones over low-bandwidth networks, which the authors present as a reason it could scale.
The authors caution that the results cover only the first year and should be read carefully, and they call for more research on the conditions needed for success. A trial in one country with a single tool says little about other subjects, languages, or school systems.
Barriers and where AI fits
The table sorts common barriers by what the cited sources support. It is a reading aid, not a ranking of tools.
| Barrier | Can AI help? | What the cited evidence says |
|---|---|---|
| No connection or device | No, it depends on them | 2.2 billion people offline (ITU, 2025) |
| No school place | Not directly | 273 million out of school (UNESCO, 2026) |
| Too few tutors for practice | Possibly | Ghana trial: effect size 0.37 after eight months, year one only |
| Rural and remote service gaps in Canada | Only after the network gap closes | 59.5 percent rural coverage at 50/10 Mbps in 2021 |
| Languages the tool does not handle | Not shown in these sources | No data gathered here |
The Canadian picture
For Canadian learners, the barrier that comes first is often the network. The Auditor General of Canada's report on connectivity in rural and remote areas, tabled on March 27, 2023, found that Canada met the government's 90 percent goal for 2021 at 90.9 percent overall, yet coverage at the 50 Mbps download and 10 Mbps upload target stood at 99.3 percent in urban areas, 59.5 percent in rural and remote areas, and 42.9 percent on First Nations reserves. The government's later targets were 98 percent by 2026 and 100 percent by 2030. Newer figures were not retrieved for this article, so the 2021 numbers should be read as a dated snapshot.
Checks before relying on an AI tool for access
- Does the tool work on the devices and connections that learners actually have?
- Has it been tested in a controlled trial with a comparison group, and for how long?
- Does it support the language and curriculum of the learners?
- Who sees the data a student types, and where is it stored?
- What happens when the connection drops or the tool gives a wrong answer?
FAQ
Can AI replace teachers where schools are scarce?
The cited trial tested a tutor added to regular math lessons, not a replacement for them. No source reviewed here shows AI replacing schooling.
Does AI help learners who are offline?
Not unless the model runs on the device itself. The tutoring tool in the Ghana study depended on a messaging connection, which is why connectivity data matter to any access claim.
Is the Ghana result likely to apply in Canada or the United States?
It may not. The trial covered grades 3 to 9 in Ghanaian schools and one subject, and the authors limit their claims to the first year. Canadian and US schools differ in resources, language, and curriculum.
Why does the Canadian rural data come from 2021?
The most recent primary document retrieved for this article was the 2023 Auditor General report, which uses 2021 coverage data. More recent regulator figures may differ.
Related reading
- Impact and potential of AI in education
- Challenges and limitations of AI in education
- A decade of AI in special education
- Why AI is considered beneficial
More in AI in education.
Last verified: October 2026
Sources
- What does the 2026 GEM Report say about access and equity in education?, UNESCO Global Education Monitoring Report, September 2026
- Measuring digital development: Facts and Figures 2025, International Telecommunication Union, November 2025
- Effective and Scalable Math Support: Evidence on the Impact of an AI-Tutor on Math Achievement in Ghana, Henkel O, Horne-Robinson H, Kozhakhmetova N, Lee A, arXiv:2402.09809, 2024
- Report 2: Connectivity in Rural and Remote Areas, Office of the Auditor General of Canada, March 2023