Has the Access Problem Been Solved? Teachers and AI’s Invisible Threshold

When we talk about what artificial intelligence might bring to education, we often assume that teachers and students already have access to these tools. We tend to think that anyone with an internet connection and a computer or smartphone is starting from the same place. Yet having access to a technology is not the same as being able to benefit from it.

At its most basic, the digital divide refers to inequalities in people’s ability to access digital technologies, use them and gain tangible benefits from that use. When the concept first emerged, it was mainly concerned with who did or did not have access to the internet and digital devices. Today, it encompasses more than connectivity. It also includes which tools people can access, how effectively they can use them and whether that use leads to meaningful outcomes in areas such as education, income, professional development and civic participation.

Artificial intelligence makes this divide more complex. Two people using the same application may not have the same opportunities. One may have access to a more capable paid model while the other remains within the limits of a free tier. One may receive high-quality results in their own language while the other has to correct errors constantly. One may have the time, training and institutional support needed to experiment with these tools while the other is trying to squeeze them into an already demanding workload. Both are recorded in the statistics as having “access,” yet the quality of that access is not the same.

The digital divide is therefore not merely a matter of technology policy. It directly affects the future of education, teachers’ working conditions and the learning opportunities available to students. As AI tools become more deeply embedded in education, differences that seem minor today may develop into much larger inequalities over time. Teachers who can use these tools more effectively may gain an advantage in lesson planning, student feedback, developing materials for different learning needs and accessing professional knowledge. Teachers without the same opportunities may be left behind, and that difference will inevitably shape the learning experiences available to their students.

At the global level, the question is whether AI will reduce existing inequalities or reproduce them in a new form. Differences in income, language and infrastructure between countries are already being carried over into AI use. English-speaking countries with strong digital infrastructure and greater purchasing power not only gain earlier access to these technologies but also benefit from better-performing models and a wider range of resources produced in their own languages.

For Türkiye, the issue matters for another reason. Internet access is now available to most of the population, yet access to paid models, the quality of Turkish-language outputs, teachers’ workloads, differences in school infrastructure and institutional support have not been addressed to the same extent. We can no longer ask only, “Do teachers have access to AI?” We must also ask, “Which AI tools can they access, under what conditions and with what level of quality?”

This is where a study published in January 2026 makes a striking claim. The study set out to identify the factors that shape teachers’ AI literacy and reached an unexpected conclusion.

According to the study, which involved 270 teachers, neither age nor gender made a significant difference. Education level and computational-thinking skills also lost their explanatory power when all variables were considered together. The same was true of AI anxiety.

Two variables remained: how much teachers enjoyed using AI and how willing they were to use it.

The researchers drew a fairly bold conclusion from these findings. In their view, the contemporary digital divide is no longer primarily a structural problem rooted in access; it has become psychological and emotional. The new divide, they argue, stems from differences in confidence, curiosity and the enjoyment people derive from using AI.

It is a powerful and highly quotable claim. That is precisely why it deserves closer scrutiny.

When we look at how the study measured the digital divide, a much narrower picture emerges. The researchers used only the physical access dimension of the digital divide scale: Do you have a device? Can you connect to the internet?

The digital-divide literature, however, approaches access on three separate levels. The first level concerns physical access, the second concerns digital skills and patterns of use, and the third concerns the tangible outcomes people gain from that use. Measuring only the first level and then concluding that the divide as a whole has become psychological is like inspecting only the entrance to a building and declaring the entire building accessible.

Where the research was conducted and who participated also matter. The study was carried out in Israel through an online consumer panel. The researchers themselves acknowledge that this method may have overrepresented participants who were already more comfortable with technology.

During the same period, Israel ranked first in the world for per-capita AI use in Anthropic’s September 2025 Economic Index, with usage approximately seven times higher than expected for its population. In other words, the finding emerged in a country where physical access had largely been secured and among a group already likely to have a close relationship with technology.

This is where one of statistics’ less exciting but decisive rules comes into play: If a variable is nearly the same for everyone, it cannot explain differences between groups.

The fact that access did not appear to be a determining factor in this study does not mean that access has ceased to matter. It means only that access did not vary enough within this particular sample. Those are not the same claim. The difference between them is the difference between repeating a study’s conclusion and understanding the conditions under which that conclusion was produced.

When we turn to Türkiye, it may be tempting to say, “We have not crossed that threshold yet.” The data, however, does not fully support this. According to TurkStat’s 2025 survey, internet use among people aged 16 to 74 reached 90.9 percent. At least at the first level of the digital divide, connectivity is beginning to lose its place as the central problem.

In that sense, Türkiye resembles the picture described in the study: The threshold has largely been crossed.

The real question is what becomes available once we step through the door.

The first dimension is economic. More capable models, long-context conversations, access to up-to-date information, file uploads and higher usage limits are largely locked behind subscriptions priced in foreign currency. A teacher using a free version and a teacher paying for a subscription may appear to be saying the same thing: “I use AI.”

In reality, they are not using the same AI.

The divide has not disappeared; it has moved from outside the tool to within it. Statistics measuring internet connectivity cannot capture this difference because both teachers are counted as having “access.” Yet the model capabilities, processing capacity and usage limits available to one may be very different from those available to the other.

The second dimension is language, an inequality that receives far less attention in Türkiye.

As an agglutinative language, Turkish operates differently from the languages on which large language models have primarily been developed. A recent study suggests that tokenization quality in Turkish—how a model breaks the language into units for processing—can in some cases matter more than model size. Simply increasing the number of parameters does not automatically improve performance in Turkish.

A teacher in Istanbul and a teacher in Tel Aviv or Boston may therefore use the same model without receiving support of the same quality. The name and version of the model may be identical, yet the user experience is not.

For someone who works with language and evaluates AI-generated text line by line, this is not an abstract technical matter. The work itself lies in recognizing where linguistic nuance disappears, which patterns seep into Turkish from English and which sentences appear grammatically correct yet still sound wrong in Turkish.

This difference cannot be explained by the variable of “enjoyment.” On the contrary, output quality directly shapes how much someone enjoys using AI. A tool that constantly needs correcting, makes language sound artificial or fails to understand what the user wants may produce fatigue rather than curiosity.

The third dimension is institutional.

For a teacher to bring AI into the classroom, having an account is not enough. They also need institutional permission, time, appropriate infrastructure and a clear framework defining what they can do and within which boundaries.

In Türkiye, this gap has been recognized at the policy level. The Ministry of National Education’s “Artificial Intelligence in Education Policy Document and Action Plan” came into effect in June 2025. Its forty action items include in-service training and AI literacy programmes for teachers.

UNESCO’s 2024 AI Competency Framework for Teachers, meanwhile, notes that only a small number of countries worldwide have defined the AI competencies teachers need and developed national teacher-training programmes around them. In this respect, Türkiye is not lagging behind.

Even so, there is a considerable distance between a policy document and forty minutes in a crowded classroom. That distance cannot easily be closed through training programmes alone while teachers’ workloads, class sizes, curriculum pressures, lack of technical support and limited preparation time remain unchanged.

None of this invalidates the study’s central finding. On the contrary, that finding is both valid and important: Motivation may matter more than many people assume.

AI literacy cannot be built through technical instruction alone. People need to be curious about the tool, willing to experiment with it, unafraid to make mistakes and able to learn how to adapt it to their own needs. This is a finding that should be taken seriously in teacher training.

Yet the statement “Teachers who enjoy using AI have higher AI literacy” contains a silent precondition:

First, that enjoyment must be possible.

That requires sufficient access, a tool capable enough to be useful, a reliable connection, adequate time and a model that produces satisfactory results in the user’s own language. It is difficult to enjoy a tool that you can barely access, that only half understands you and that you are trying to use in the few minutes between classes.

Enjoyment does not replace access; it comes after access.

When we reverse this order, we begin to interpret structural shortcomings as personal reluctance. It becomes easy to say that teachers are not curious enough, are afraid of technology or are resistant to change. Meanwhile, the cost of the tool, its lower-quality performance in Turkish and teachers’ lack of time disappear from view.

The claim that the digital divide has become psychological is appealing partly for this reason. Psychological problems appear cheaper to solve: organise a workshop, design a more playful training module or develop a programme intended to change teachers’ attitudes.

Structural problems require money, infrastructure, licences, investment in language technologies and time.

The way a research finding circulates is not determined solely by how accurate it is. It is also shaped by the solutions it demands and whose interests those solutions serve.

The real risk is not that the study is wrong. The real risk is that it will be cited to make claims about areas it never measured.

When a divide becomes invisible to the instrument designed to measure it, it has not closed.

It has simply stopped being counted.

References

Anthropic (2025). Anthropic Economic Index: Uneven Geographic Adoption, September 2025 report.

Bayram, M. A., Fincan, A. A., Gümüş, A. S., Karakaş, S., Diri, B. & Yıldırım, S. (2025). Tokenization Standards for Linguistic Integrity: Turkish as a Benchmark. arXiv:2502.07057.

Deshen, M., Harari, R. & Aharony, N. (2026). Teachers’ Artificial Intelligence (AI) Literacy: An Exploratory Study. Smart Learning Environments, 13(7).

Ministry of National Education of the Republic of Türkiye (2025). Artificial Intelligence in Education Policy Document and Action Plan (2025–2029), June 2025.

Scheerder, A., van Deursen, A. & van Dijk, J. (2017). Determinants of Internet Skills, Uses and Outcomes: A Systematic Review of the Second- and Third-Level Digital Divide. Telematics and Informatics, 34(8).

TurkStat (2025). Household Information Technology Usage Survey; reported by Anadolu Agency.

UNESCO (2024). AI Competency Framework for Teachers.