Johnny Chang's year-end review of AI in education cuts through the hype to reveal a stark reality: while technology races ahead, the human infrastructure required to wield it responsibly is dangerously lagging. This piece stands out not for listing new tools, but for synthesizing a global disconnect between student adoption and institutional guidance, backed by fresh data from the World Bank and major consulting firms.
The Global Adoption Gap
Chang opens by noting the sheer velocity of technical progress, citing the launch of OpenAI's GPT-4o and o-series models as bringing "unprecedented accuracy and speed." Yet, the narrative quickly pivots from the capabilities of the software to the uneven distribution of access. Chang highlights a World Bank report based on focus groups in ten countries, which found that while students are eager to use AI for coding and writing, "there are still some regions where students face challenges with high internet costs and low connectivity affecting their access to AI."
This observation is crucial. It reframes the conversation from one of academic integrity to one of fundamental equity. Chang points out that students in STEM fields have emerged as "early adopters," signaling a need for comprehensive AI education across all disciplines, not just technical ones. The argument here is that the digital divide is no longer just about having a device; it is about having the bandwidth and literacy to compete in an AI-augmented world.
The Trust Deficit and the Source Problem
Perhaps the most striking finding Chang presents comes from a joint report by EY and TeachAI, which surveyed over 5,000 Gen Z respondents. The data reveals a troubling disconnect in how young people learn about the technology that will define their careers. Chang quotes the report directly: "The two most common sources Gen Z turns to for AI information are social media (55%) and news articles/media (35%). Unlike these self-directed sources of education, educators and colleagues/employers were much lower in comparison, at 14% and 12%, respectively."
This statistic suggests that schools are failing to lead the conversation, leaving students to navigate a complex landscape via algorithms designed for engagement, not accuracy. Chang further notes that trust in AI is not uniform; "Those in the Middle East, Africa and India have a higher trust in AI, whereas North American respondents have the most distrust." This regional variance complicates the idea of a universal "best practice" for AI integration in the classroom.
That perception is coming from conflicting messages from educational institutions around Generative AI as a tool to "cheat on assignments" versus Gen AI as a tool to use in their education but not rely on … How do educators help prepare students for the kind of critical pivoting that Gen AI tools of the future are going to ask of them? I think this is an incredibly powerful way to prepare the workforce of the future. And this survey shows that's not happening. Yet.
Chang attributes this quote to Gina Neff, an expert at the University of Cambridge, and the framing is devastatingly effective. It exposes the paralysis of institutions that cannot decide if AI is a cheat code or a calculator. Critics might argue that schools are right to be cautious, fearing that premature integration could erode foundational skills. However, as Chang implies, the current "wait and see" approach is arguably more damaging, leaving students unprepared for a workforce that is already pivoting.
From Theory to the Classroom Floor
Moving from global surveys to specific case studies, Chang details how institutions are attempting to operationalize these tools. He highlights an Arizona charter school, Unbound Academy, which has been approved to offer "AI-only online classes" for grades 4–8. This represents a radical departure from traditional schooling, offering just two hours of academic instruction daily through an AI-driven model. While Chang presents this as an innovation, the implications for human interaction in education are profound and warrant scrutiny.
Conversely, Chang points to a more balanced approach at UCLA, where a comparative literature course will utilize a custom AI system called Kudu. This system is designed as a "closed-loop" where the AI only draws from professor-approved materials, preventing misuse while allowing the professor to focus on "critical thinking and primary source analysis." Chang notes that the AI-generated textbook is available for a low cost and supports accessibility features like audio readers. This contrast between the fully automated Arizona model and the human-supervised UCLA model illustrates the spectrum of possibilities currently being tested.
The Research Verdict
To ground these anecdotes in evidence, Chang reviews a meta-analysis of 69 experimental studies published in Computers & Education. The findings are nuanced: ChatGPT usage in university language education "boosts academic performance, enhances motivation, and fosters higher-order thinking, while reducing mental effort for students." However, Chang is careful to note that the study does not find a significant effect on self-efficacy, and warns of "small sample sizes and potential biases in post-intervention assessments."
Chang also includes a student opinion piece that captures the double-edged sword of this efficiency. The student argues that while AI reduces the load on tedious tasks like formatting, "in the long term, on the other hand, it increases dependence on these tools." The example given is stark: a computer science student who relies on AI to generate Python code may eventually be "unable to write basic Python syntax because they are used to generating syntax automatically." This is the core tension Chang weaves throughout the piece: efficiency versus mastery.
Bottom Line
Chang's review succeeds by refusing to treat AI as a monolith, instead mapping the friction between rapid technological adoption and the slow, uneven pace of educational reform. The strongest part of the argument is the data showing that students are learning about AI from social media rather than educators, a gap that threatens to widen inequality. The biggest vulnerability remains the lack of long-term data on how AI dependence impacts deep cognitive skills. As the year closes, the most urgent task for the education sector is not just adopting new tools, but resolving the conflicting messages that leave students navigating this future alone.