AI Literacy and Rigor Were Never Enemies

As artificial intelligence (AI) becomes a bigger part of education, the challenge is no longer whether students should use it, but how teachers can design learning that preserves rigor while building AI literacy. This three-part series explores the research, classroom practice, and practical strategies behind using AI to deepen thinking rather than replace it.

Introduction

"Striking a balance" makes it sound like a compromise, as if teaching AI literacy and protecting academic standards pull in opposite directions, and the best a teacher can do is split the difference. But actually, we want to argue the opposite. The balance is not a truce between two competing goods. It is a design problem, and when you solve it well, the two goals complement rather than oppose one another. 

We are four teachers at Colegio Nueva Granada (CNG), an international school in Bogotá, Colombia, who teach Advanced Placement (AP) Physics, high school mathematics, and sociales (Colombian social studies) to a bilingual student body that has been navigating AI tools for longer than most of their teachers have. We also serve on our school's AI Core Task Force, where we have spent the year building an institutional framework for responsible AI use while simultaneously testing tools in our own classrooms. Stress-testing the AI rules as we collaboratively designed systems has turned out to be a useful vantage point to develop real-time, practical insights on AI instruction.

Education is at a crossroads, and AI literacy is no longer a specialty to be added on once the "real" curriculum is covered. It is a basic condition of learning in every discipline, meaning it belongs in every classroom and can be taught in the context of our current courses. For us, this is not a detour from our school's mission but an expression of it. At CNG, we educate students "for leadership and service in the world of today for a better tomorrow," and to graduate deep thinkers, bold innovators, and change agents. None of those aspirations survives contact with a future in which our students use AI fluently but uncritically. Teaching them to think with these tools, rather than around them, is the work, and the result of this work provides evidence that AI does not replace the teacher-student relationship; rather, intentional AI frameworks have the potential to create more opportunities to leverage this relationship. This is the story of how we tried to build a tool to refuse the shortcuts, to insist that students access what they know and extend it, while still meeting each student wherever they happen to be in readiness. 

Where We Started

The research now emerging makes clear that the difference between AI helping and AI harming a student’s educational journey is almost entirely a question of design. This is not a new insight so much as a new application of an old one. Nearly two decades ago, Mishra and Koehler (2006) argued in their Technological Pedagogical Content Knowledge (TPACK) framework that effective technology integration does not come from technological skill on its own; it lives at the intersection where a teacher's technological, pedagogical, and content knowledge overlap. AI has not changed that relationship. It has only raised the stakes of getting it right.

In June 2025, a team at the MIT Media Lab published "Your Brain on ChatGPT," a study that used EEG to measure the cognitive engagement of students writing essays with an AI assistant, with a search engine, or with no tools at all. The students who leaned on the AI showed the weakest neural connectivity, the poorest recall of their own writing, and the least sense of ownership over their work. The researchers named the effect "cognitive debt": the way mental shortcuts feel efficient in the moment but leave lasting gaps in learning. Tellingly, the harm was not inherent to the tool. The students who began with their own thinking and only then brought AI in showed stronger engagement, not weaker. The damage came from sequence and design, from using AI to bypass thinking rather than to extend it. It’s important to note that the study was small and not yet peer-reviewed, but its central finding aligns with a growing body of work. 

This is the heart of the problem we set out to address. If there is no plan, no intentional design for how and when students engage AI, students will use these tools to produce rather than to process, and we lose the chance to have them access and apply their knowledge at a higher level. The tool will happily think for them. The question is whether we, as educators, are willing and adept at designing tasks that ask them to think instead.

International policy frameworks have begun to agree. UNESCO's (2024) AI Competency Frameworks for Teachers and for Students describe a shift from the traditional teacher-student relationship to a "teacher-AI-student" dynamic. And they insist that students be educated as critical evaluators and co-creators of AI rather than passive consumers, moving across levels they label Understanding, Applying, and Creating. The AI Assessment Scale, developed by Perkins and colleagues (2024), is now used in schools across dozens of countries, and the framework argues that banning AI has proven ineffective. The appropriate level of AI use should be matched to the student learning outcomes.

CNG's own framework draws directly on this body of work. Our Student Artificial Intelligence Level (SAIL) system, adapted from the AI Assessment Scale and aligned with UNESCO's human-centered principles, gives teachers and students a shared language for what kind of AI engagement a given task calls for, from No AI through Ideate, Edit, Evaluate, and Co-Create. The framework is explicit that AI must enhance professional and academic judgment, never replace it. Thus, the need for a professional framework along with practitioners’ professional understanding is to be developed through intentional AI use.

As teachers, we did not feel ready when AI became unavoidable in our students' lives; few educators did. But the framework gave us principles, and the pilot gave us a place to test them. What follows is what we found when we stopped asking how to keep AI out of our students' work and started asking how to design it in.

The Problem

The first fear most teachers name about AI is cheating, a concern we share. But after a year of this work, we believe it is the wrong place to start, not because integrity does not matter, but because framing the problem as cheating keeps us playing defense. It puts our energy into detection, restriction, and suspicion, and it quietly concedes that the best outcome available is a student who simply does not use the tool. It casts us as gatekeepers rather than educators who trust students' capacity. For a generation that will spend their working lives alongside these systems, that is a thin definition of success. And the predominant narrative around AI use is one of efficiency; faster planning, feedback, writing, answers. But educational frameworks are not fundamentally efficiency-based. Learning requires struggle, feedback, reflection, revision, and human judgement. And thus the most promising role of AI is not to eliminate those processes, but to allow for enhanced creativity, critique, and curiosity, leading to understanding.

The more useful question is the one the MIT study points to. If students already have access to AI, on the phones in their pockets, regardless of what any syllabus says, then the meaningful question is not whether they use it but how. The same tool, in the same hand, produces cognitive debt or cognitive gain depending entirely on the task in which it is embedded. That is not a problem that detection software can solve. It is a problem of assignment design. The design lever is sequence. Ask students to commit their own thinking first, then bring AI in to pressure test it. Even a single assignment reordered this way changes what the tool does to learning. 

This is also where a framework on paper meets its limits. CNG's SAIL system gives our students and us a shared vocabulary for AI engagement, and that vocabulary is genuinely useful, but naming a task "Evaluate" or "Co-Create" does not by itself make the task demand real thinking. The level is a label; the rigor has to be built. A teacher can assign an AI-permitted task that still lets students offload everything that matters, or design one that uses AI to push students into the higher reaches of their own understanding. The framework tells you which level you are aiming for. It does not write the lesson.

So the question we kept returning to, across three very different subjects, was narrow and concrete. Could we design AI into a task in a way that demanded more thinking from students, not less? Could a tool be built to refuse the shortcut, to insist that students access what they know, apply it, defend it, and extend it, while still meeting each student wherever they happen to be? And could it hold the content standard and the student's starting point at the same time?

We began with a question rather than a conclusion. If AI literacy and academic rigor are not opposing goals but products of thoughtful instructional design, then the real test lies in the classroom. Frameworks, research, and principles can point the way, but they cannot tell us whether students actually think more deeply when AI becomes part of the learning process.

So we built the tasks, invited students into them, and watched carefully. What happened next surprised us. Not because AI replaced good teaching, but because, when designed intentionally, it repeatedly pushed students back toward the thinking, struggle, and teacher relationships we hoped to preserve.