High Tech, Low Literacy: How the AI Boom Skips Skills the Classroom Forgot

2025 is the year AI starts eating jobs.

Intel announced on 23 April 2025 it will cut 20% of its global workforce, over 20 000 positions, to fund an “AI-centric roadmap.” Google, Microsoft, and Meta are all following suit, with junior programmers going first. Independent jobs-loss tracker layoffs.fyi puts the total slashed so far this year at 28, 728. Mark Zuckerberg, boasts it’s the “Year of Efficiency” — and more “efficiency” is coming.

According to Anthropic, AI agents will be our new co-workers by next year, usurping many repetitive, automatable job roles. The World Economic Forum suggests that 86% of global bosses expect workers to possess basic digital literacy skills by 2027. Even so, 41 % of employers plan to downsize their workforce by 2030.

But it’s not all bad. History shows many new jobs will be created, and experts expect that AI will augment more roles than it will replace. Skilled tech workers will likely find clever ways to pivot and may find themselves better off in a world where they have superpowers at their fingertips. Others will probably become AI-proficient and, at least in the near-term, find ways to adapt. I’m optimistic about how AI will impact education for students, teachers, and parents alike. I experiment with AI everyday and have learned to turn curriculum design tasks that once took months into mere hours of work. Those who learn to use AI well will likely thrive.

The problem, however, is that it looks like a lot of people will struggle to learn AI. To begin with. it’s not that easy to learn. Sure, you can get it to create cool Ghibli-style cat pictures, ask it to plan your vacations, or give you answers that seem plausibly reliable. But as soon as you move to “compare and contrast the arguments in these three reports” or ask questions with uncertain or contentious answers, everything changes.

Many complex tasks demand precise prompt articulation, output evaluation, and domain-specific judgement. One needs to read, write, and think critically at a pretty high level to use AI well. Even the best models still hallucinate sometimes. One must be able to ask good questions, identify errors and inadequacies, and be creative when asking follow-ups. I have seen many teachers give up trying to use it to plan lessons because of poor one-shot AI outputs.

Nor are strong skills alone a guarantee of effective use. We need to be both aware and honest with ourselves about whether they way we use AIs helps or harms us. Even Wharton students in a recent study became over-reliant on AI tutors, weakening their retention. Imagine the challenge for people who struggle with multi-paragraph text.

And here’s the thing: most adults actually do struggle with that. In fact, more than half of American adults don’t have the skills needed to use AI effectively since they are just barely functionally literate.

When I first saw the number I was doubtful. Billionaire hedge fund manager, Ray Dalio, said in an interview that 60% of people in the US read below a grade 6 level. Perplexity, ChatGPT, and Gemini confirmed it for me, albeit with a minor revision: the number is actually 54%

This figure comes from the US National Center for Education Statistics as part of the Organization for Economic Cooperation and Development’s (OECD) Program for the International Assessment of Adult Competencies (PIAAC). The 2017 PIAAC sample revealed that130 million of the 239 million adults aged 16–74 fell at or below Level 2.

Level 2 is equivalent to a lower-middle-school reading band. What this means in plain English is that over half of US adults can read simple sentences and paragraphs but struggle to comprehend multi-paragraph texts, make inferences to locate information, or to evaluate the credibility of sources — core skills in the AI age.

Weak reading literacy skills translate into weak digital literacy skills. The OECD’s Skills Outlook 2019 shows that adults reading at Level 2 or below are more than three times as likely to score “below basic” on the Problem-Solving-in-Technology-Rich-Environments test, which measures how well people can fill in web forms, search for information online, and follow multi-step digital instructions. Not exactly advanced.

I was pretty shocked when I learned the 54% stat. How could that be? How can a country with strong high school graduation rates and most of the best universities in the world produce such poor outcomes at scale?

I decided to dig into the data and what I found is even worse. It’s not abnormal for OECD (rich) countries to have high numbers of people graduating high school with low literacy skills.

The range is from around 18% at best in Estonia to a shockingly high 67% in Hungary, the beautiful land where I live.

Crunching the numbers to see how literacy rates map onto high school graduation rates led me to create this graph in Datawrapper:

What I see here is a real problem brewing. In a world of impending automation and job displacement, it’s just not going to work if we are graduating 87% of kids from high schools in the OECD while 48% of them cannot critically evaluate information or solve problems online. It’s an abject educational failure and a crisis waiting to happen.

Curious and concerned, I dove deeper into the data. Which jobs are most at risk? This is well-known and often reported. It’s the routine-manual jobs, followed by cognitive tasks that are more efficiently done by machines.

But how much does literacy factor in? That was something I couldn’t find any good answers to, so I did the research myself with the aid of ChatGPT-o3. It appears that there is a stark correlation between jobs-at-risk due to automation and low literacy levels.

Controlling for immigration, the blue-red scatter plot combines sector-level literacy scores for native-born workers from the OECD Skills Outlook 2024 with automation-risk probabilities calculated using the widely-cited Frey & Osborne (2013) model, updated by ILOSTAT in 2023. The data underpin peer-reviewed policy papers and are considered the most reliable evidence on skills and task-automation risk.

Interestingly, computer programmers, the job role currently getting the most media attention, are only medium-risk. Though their literacy levels are surprisingly low, on average, I suspect they will be okay. It’s the people in the red who I now worry about. Tech jobs are merely the appetizers for AI. Low-literacy jobs are the main course.

Here’s my question: What happens when these people are out of work in a couple of years, robots are dancing while serving drinks, self-driving scooters are delivering pizza, agents are hiring and firing, and millions of workers need retraining… yet struggle to read?

And to be clear, it is millions — around 107 million in rich OECD alone — that we are talking about.

This isn’t good.

I spend most of my time thinking about the skills of the future — agency, wellbeing, regeneration, entrepreneurship, that sort of thing. But those skills mostly assume a level of literacy that is simply not present across the adult population in 2025. In my little bubble at REAL School Budapest, that makes sense. But at system-scale, at least two of the three Rs need a reboot.

My take-away from this? The global system of schooling has failed to prepare us for a future that is nearly here. Millions of people have been underserved by being allowed to graduate without the right levels and kinds of basic competencies. A system that gives one a false sense of capability by awarding empty diplomas does us all a grave disservice. We need a new system — and we need it fast. I have a few ideas on that.
-

This is me writing again. I am planning a series of pieces on the future of education in the AI/AGI era. The goal is to help me think more deeply about the why, what, and how of the model we are building at REAL School Budapest, and to hopefully attract the right people, as we try to be the change.