Last month McMaster University linguistics professor Victor Kuperman\u00a0was awarded a 2026 Killam Fellowship by the\u00a0National Research Council of Canada\u00a0to create AI-powered tools for adult literacy. Here, he explains the stakes.<\/p>\n
AI for Everyone? Not with the current, widening literacy gap. As governments and employers rapidly embed AI into everyday decisions, Canada faces a pressing question about productivity gains: who is actually equipped to use these tools safely and effectively.<\/p>\n
Imagine a typical Canadian bus carrying 40 passengers, each holding a copy of this newspaper and turning to this article at the same time.<\/p>\n
Everyone on board will likely be able to read these words and process these sentences.<\/p>\n
But at least 16 of them won\u2019t be able to understand them well enough to form conclusions about what they have read or to place it into the context of what they already know.<\/p>\n
More than 40 per cent of adult Canadians are functionally illiterate, as international surveys show, and the percentage is even higher among non-native speakers of English or French.<\/p>\n
They struggle to act reliably on written instructions, to make connections between different texts, or to integrate text information with related charts or tables\u00a0\u2014 that is, they lack the foundational skill of literacy required to succeed in today\u2019s increasingly technological society.<\/p>\n
The lack of foundational skills such as literacy among adults is often cited as one of the causes of the productivity crisis in Canada, and its impact on the economy is enormous. At a time when Canada is searching for productivity gains, we are overlooking one of the most fundamental constraints: a workforce that cannot fully access, interpret, or act on information. A 2020 report from Deloitte Canada estimated that an increase of one per cent in average adult literacy scores would boost Canada\u2019s GDP by what would be $95 billion in today\u2019s dollars.<\/p>\n
But the impact goes far beyond macroeconomic statistics. At the individual level, it diminishes the quality of life.<\/p>\n
Every passenger on that bus should care about improving literacy. It\u2019s a pathway to a more qualified and productive workforce; healthier, happier and more engaged citizens; and more cohesive communities and societies.<\/p>\n
The person who lacks foundational literacy can\u2019t sufficiently integrate socially or culturally, can\u2019t contribute fully to political and community life, or even use the health care system effectively. Without foundational literacy, it\u2019s impossible to meet one\u2019s full potential at work or at home.<\/p>\n
In this sense, everyone misses out when so many struggle to process written material.<\/p>\n
Governments and companies are rapidly embedding AI into workplaces, public services, and education systems. As billions are invested, critical questions are being overlooked: who is actually equipped to use these tools safely and effectively? And can AI solve a problem it fundamentally depends on?<\/p>\n
It seems sensible to think so, since generative AI appears to be doing a great job both reading and writing for us. And agentic AI is getting increasingly better with planning and implementing complex workflows, as well as interacting with external tools, requiring only minimal verbal input from us.<\/p>\n
But AI alone is not enough. If anything, the burden is on us to understand the answers AI generates, the options it offers, and the knowledge it is collecting and summarizing for us. All that requires literacy. In practice, this means workers misinterpreting AI-generated instructions, patients misunderstanding health information, or citizens relying on outputs they cannot fully evaluate.<\/p>\n
Today\u2019s AI does not provide legal or ethical safeguards against suboptimal, mistaken, or harmful responses.<\/p>\n
It is not immune to misinformation or disinformation. Nor is it flexible enough yet to account for the needs and practices of specific demographic, social, ethnic, or linguistic groups.<\/p>\n
This puts humans in a dual role as both users and judges of the quality and reliability of the information AI provides. Without the literacy to question and interpret outputs, AI risks amplifying \u2014 not reducing \u2014 existing inequalities. Far from replacing literacy, AI raises the bar for what literacy must enable: not just reading text, but evaluating, contextualizing, and making informed judgments.<\/p>\n
Not only does functional illiteracy make this complex role difficult to perform, it also hinders a large proportion of the population from adopting or efficiently using the very AI tools that are designed to simplify and improve everyone\u2019s lives.<\/p>\n
For decades, functional illiteracy has been widening the digital divide that separates those who can access communication technologies from those who can\u2019t.<\/p>\n
In the AI era, illiteracy threatens to perpetuate the vicious cycle that pushes large groups of populations farther behind and deepens the inequalities in Canadian society. What was once a gap risks becoming a rapidly widening divide \u2014 one that accelerates with every technological leap.<\/p>\n
An AI strategy without a literacy strategy is, ultimately, an inequality strategy.<\/p>\n
A partial solution\u00a0\u2014 one that is easy to name but far from easy to implement\u00a0\u2014 would be to apply AI tools to improving the literacy levels of Canadians.<\/p>\n
Computer-assisted language learning is a natural domain for such implementation, and it has indeed been extremely quick in incorporating AI into the creation and delivery of instructional materials.<\/p>\n
Consider, for example, second-language learning apps such as Duolingo, which rely on AI for their entire instructional workflow, from assessing the proficiency of their users to creating language exercises and boosting learning through repeating words and phrases based on the user\u2019s progress.<\/p>\n
Similarly, automatic translation tools are extremely helpful for navigating our multilingual world, and summarization tools help us cope with an influx of information.<\/p>\n
While such tools can be helpful for individuals, it does not look like the solutions we have are moving the needle towards better literacy at the national scale. An authoritative survey did not show any change in adult literacy skills between 2012 and 2023. This suggests we need to explore new avenues and invest more effort in closing the digital divide.<\/p>\n
I see untapped potential for AI tools to provide personalized education through what\u2019s called \u201ctext adjustment.\u201d<\/p>\n
Imagine an app that can take an English text provided by the user and adjust it to that person\u2019s proficiency level. The app would make a quick assessment of the user\u2019s English proficiency as well as asking for the user\u2019s first language.<\/p>\n
The input text could then be rendered into an adjusted English text that retains the original content but adjusts the level of complexity to bring it into closer alignment with the user\u2019s optimal proficiency.<\/p>\n
As cognitive science shows, setting that level is challenging. It needs to encourage the learner to progress but not be so difficult that it discourages them.<\/p>\n
Training AI to adjust texts to individual proficiency levels while accounting for the progress users make over time can lead to the gradual scaffolding of literacy skills in English, while helping people with endless daily tasks that require text understanding, like filling out forms, reading lease agreements, or following instruction manuals.<\/p>\n
Text adjustment technology could also take the edge off a common problem with motivation and interest in reading-skill acquisition.<\/p>\n
I am a beginning learner of Spanish, but the only materials available to me for improving my reading at this stage are either Spanish books for children or rudimentary vocabulary exercises in textbooks or language learning apps.<\/p>\n
What would boost my learning is an app that would take the Spanish text of my choosing\u00a0\u2014 be it news items, weather reports, or short fiction\u00a0\u2014 and adjust the text to a level that enables me to understand interesting content while strengthening my skill.<\/p>\n
Knowing the user\u2019s first language would also help an app to create adjusted texts that use words and structure that are common to both English and the user\u2019s first language.<\/p>\n
<\/p>\n
McMaster University linguistics professor Victor Kuperman.<\/p>\n
Georgia Kirkos\/McMaster<\/p>\n
For example, \u201ccommence\u201d is a relatively rare word in English that would not work well in simplified English text for Mandarin speakers. The word\u2019s form and meaning would be very familiar, though, to speakers of Romance languages. In this way, English texts adjusted by the AI app for Spanish speakers would differ from those adjusted for speakers of other languages.<\/p>\n
The question is no longer whether AI will transform how we live and work, but who will be equipped to participate in that transformation.<\/p>\n
My research focuses on developing and validating text adjustment technology. Even if it turns out to be successful, as I hope it will be, it will not be a cure-all for improving adult literacy. At best, it may move a few more passengers to that group on the bus who can comprehend this article.<\/p>\n
No technology will ever remove the need to support literacy in our institutions, communities and families.<\/p>\n
We must maintain and increase our investment in language education for children, adults, and newcomers; align teacher training and school curricula with the best evidence-based practices; rigorously investigate the causes of functional illiteracy; support literacy-positive informal education; and simply read to and with our children.<\/p>\n
Literacy is critical for the well-being of everyone on the bus. Any shortfall hurts everyone\u2019s progress. If literacy was the foundation of past economies, it is the gatekeeper of the AI economy. Before AI can work for everyone, literacy must.<\/p>\n
Victor Kuperman is internationally recognized for his work in literacy, the cognitive science of reading, eye-movement research, and language learning using computational and experimental methods to investigate language-related issues of direct social relevance. <\/p>\n