The AI Education Paradox: Why Better AI Requires Better Learners
The educator’s challenges in the age of LLMs
By Rafael Izbicki, Federal University of São Carlos.
Large Language Models have quickly become part of everyday professional routines. They’re fast, efficient, and often surprisingly capable. I use them regularly myself, including to help brainstorm and refine the very ideas in this text. Their impact on productivity is undeniable. Yet, as an educator and parent, I’m increasingly concerned about their role in education, particularly in preparing students for the job market.
The Rising Bar for Human Value
Here’s the tricky part: As LLMs get better, users need more knowledge and skill to add real value beyond what the model already gives. In other words, the more sophisticated the tool, the more sophisticated its user must be to contribute beyond its outputs, or even to craft better prompts.
Consider my field of statistics. Tasks that once required advanced training — fitting models, generating clear visualizations, writing analysis summaries — can now be automated by LLMs with increasing accuracy. This democratization of technical skills is powerful, potentially enabling people with minimal training to produce useful results.
But there’s a catch: not everyone can have a job simply because they can use these tools. To remain valuable, professionals must excel where LLMs fall short. For a statistician today, real value isn’t about figuring out a decent model for the data and running it; it’s about being the skeptical detective who questions it. It’s about spotting the subtle flaw in the code that an LLM misses, or challenging how the data was collected in the first place. Most importantly, it’s about having the wisdom to ask questions the machine would never think to.
What keeps professionals uniquely valuable is mastery of first principles: the habit of questioning data sources, assumptions, and incentives (including those embedded in LLMs’ training data). Deep foundational understanding helps prevent human expertise from becoming obsolete.
The Paradox
This creates a paradox in education in the era of AI.
On the one hand, students need stronger foundational training to generate insights beyond what a layperson with an LLM could produce. They must go beyond mere procedure execution to understand why some approaches work, when they fail, and how to evaluate results critically. They need to read and interpret code, debug subtle errors, and carefully reflect on the choices made by the AI and their implications.
On the other hand — and here’s the irony — LLMs often undermine the very learning process that builds these foundations. The temptation to use these tools as shortcuts to learning is overwhelming. Students increasingly rely on LLMs for programming assignments, solving math problems, and writing entire reports without engaging with the underlying reasoning. The result is surface-level understanding with minimal actual learning. Genuine learning requires effort; there’s no way around.
Finding the Balance
This doesn’t mean LLMs cannot be a good tool for education. Used the right way, they can help students dig deeper into topics, make sense of tricky concepts, and work through their ideas. But they’re only beneficial when integrated into a learning process that still demands effort. Without these elements, learning doesn’t occur.
Critical Questions for Educators
As educators, we face critical challenges: How do we design instruction that strengthens foundational understanding? How do we promote responsible AI use while preventing misuse? Most importantly, what core skills and thinking habits must we preserve at all costs?
These questions haunt me every time I plan a lesson. How can I teach students to code properly when I know ChatGPT can deliver a working script in seconds? How do I convince them that the frustration of debugging a bug is actually where the learning happens? How do I convince them not to hand off their calculus homework to DeepSeek? What do I need to teach?
These questions have no easy answers as far as I can tell. But if we don’t confront them directly, we risk training a generation of professionals who may be outpaced by the very tools they were taught to depend on.
We must teach students to become irreplaceable by mastering what machines cannot (yet!) do, while the presence of those same machines can make that mastery harder to achieve. This is the tension of modern education.
