Education Is Falling Behind Because We’re Teaching Around AI

Recently, a colleague asked me, “Do you think AI is destroying education?” I paused…not because the question was difficult, but because the premise was off. AI isn’t disrupting a healthy education system; it’s exposing one that was misaligned from the start. The uncomfortable truth is that many of the problems people blame on ChatGPT, Large Language Models, and automated assessment were baked into our classrooms long before students typed their first prompt.

Like putting a spotlight on an old stage, AI has simply made the cracks visible. If we want to stop the stage from collapsing, we have to repair the underlying structure, not just dim the lights.

When AI Pulls Back the Curtain

Over the past two years, I’ve watched educators scramble to ban, police, or detect AI-generated work. The instinct is understandable—no one wants their class turned into a contest of who can trick an algorithm better. But this focus on cheating obscures a deeper problem: many of our most common teaching practices never measured what we thought they measured.

I saw it first in my own corner of the world. In EMS and healthcare training, we pride ourselves on rigorous assessments and licensure exams. Yet behind the scenes, cut scores are often inherited rather than justified. As a Public Safety Group post noted, “Arbitrary cut scores, commonly 70 % in typical letter grading, are a harmful standard, as they lack detailed information about the difficulty of the items and their importance” . Why do so many programs pick 70 % as the dividing line between safe and unsafe practice? It’s tradition—not validation. These thresholds tell us little about competence, much less about the professional judgment we claim to assess.

It gets worse. The reliance on percentage grades across education is pervasive and problematic. A 2013 ASCD analysis explained that percentage grades are “difficult to defend from a procedural, practical, or ethical perspective” (Yes, it’s an old study but still relative) . In fact, a century-old study found that 147 high school teachers assigned scores ranging from 64 to 98 to the same paper . Modern replications show the same spread . When teachers can’t agree by more than thirty points on a single essay, the difference between 69 % and 71 % becomes laughably arbitrary. Yet that line decides who passes and who fails.

AI didn’t invent this inconsistency. It just makes it impossible to ignore.

The Illusion of Objectivity: Passing Scores Without Proof

Ask any instructor why a 70 % is considered “passing” and you’ll hear some variation of, “That’s just how it is.” We rarely interrogate the origins or validity of these thresholds. But cut scores and letter grades are judgments, not facts. In high-stakes exams, psychometricians use methods like Angoff and Nedelsky to align cut scores with minimal competence. Unfortunately, that rigor is rare in everyday classrooms.

The Public Safety Group explains that determining a cut score “is a critical and difficult task” and that arbitrary cut scores can be harmful . Good standard setting requires subject matter experts to define what a minimally competent performer should know and to calibrate each exam item accordingly . In most education settings, though, cut scores are borrowed from unrelated disciplines or chosen for convenience. A 70 % threshold may separate pass from fail on paper, but it lacks any empirical tie to actual competence.

Percentage grading systems themselves amplify the distortion. As the ASCD article notes, a 100-point scale “identifies 60 or more distinct levels of failure and only 40 levels of success” . This skew means that a single bad test can tank an average, while there are few ways to recover. In the same analysis, researchers found that modern teachers’ scores on identical papers still range from 50 to 96 . If well-trained educators can’t agree within 46 percentage points, what does a pass/fail line really tell us?

The uncomfortable consequence is this: our beloved grade cutoffs are often more about tradition and convenience than about student learning or safety. AI didn’t create this. It exposed how fragile the system already was.

Unreliable Measures, Unprepared Instructors

In healthcare education, we like to think we’re better. We talk about item discrimination, distractor analysis, KR–20 reliability. But in practice, many exam administrators can’t explain what a discrimination index is or why it matters. Too often, cut scores are set once—usually at 70 %—and left unexamined for years. Exam items are rarely reviewed for difficulty, bias, or alignment with objectives. It’s no wonder that AI-generated answers can sometimes pass; the tests measure rote recall more than clinical reasoning.

Outside of healthcare, the problem is broader. Vocational and technical education instructors are often recruited from industry, not academia, because they know the trades. According to a World Bank education report, there are “no clear paths for becoming a teacher in TVET” and those who come from industry “sometimes lack training in pedagogy.” Teaching hands-on skills requires different approaches from teaching academic subjects , yet many vocational trainers never receive formal instruction in how to teach.

Meanwhile, educators who do go through pre-service programs receive pedagogy but not enough industry experience . They’re prepared to lecture, but not to guide apprentices through messy, unpredictable real-world tasks. The World Bank notes that high-quality vocational teachers need a mix of pedagogical and technical training, including interactive techniques like simulations and project-based learning . They also need support to develop assessments that actually measure practical competencies . Without that training, instructors fall back on what they experienced as students: lectures, worksheets, and multiple-choice tests.

In other words, our teacher preparation pipelines don’t equip instructors for the challenges they face, particularly in vocational settings. When AI shows students can generate passable essays or code snippets in seconds, it isn’t undermining good teaching—it’s revealing that we often weren’t doing much teaching at all.

Why the Myth of “Grade-Based Mastery” Persists

If the research is clear, why do we hold on to percentage grades and arbitrary cut scores? Several forces keep these myths alive:

• Simplicity and familiarity. Numeric grades and cut scores are easy to calculate and report. Schools and software vendors have built entire systems around percentages, so changing them feels disruptive.
• Illusion of precision. A two-decimal-point grade looks scientific. It gives parents and students something to fixate on. But as the ASCD study showed, percentage grades give a false sense of accuracy when teachers’ scores can vary by forty points .
• Tradition and inertia. We were graded this way, our mentors were graded this way, and changing a century-old practice seems daunting.
• External expectations. Licensure boards, employers, and accreditation bodies often require numeric scores. Even when programs know better, they feel bound to maintain thresholds.

AI pokes holes in these justifications because it can match the system’s expectations with minimal effort. Once students can produce “passing” assignments at the click of a button, the limitations of those assignments become obvious. It forces us to confront whether our assessments measure anything worth measuring.

The Hidden Costs of Clinging to Broken Metrics

Continuing to rely on fragile grading systems doesn’t just misclassify students—it carries real risks:

• Self-limiting beliefs. When students internalize that 70 % equals “barely adequate,” they may avoid subjects where they score lower, even if those areas are essential to their growth.
• Wasted teacher effort. Educators spend hours crafting assignments that AI can complete in seconds. Time that could be spent mentoring students becomes time policing the system.
• Equity gaps. Students with access to AI tools learn to leverage them; those without fall further behind. If the point of an assignment is to produce a summary rather than to think critically, AI gives advantaged students a leg up and leaves others chasing points.
• False confidence. Passing a multiple-choice test doesn’t ensure a paramedic can make split-second decisions under pressure. Graduates who have never been assessed on judgment become liabilities, not assets.
• Professional credibility. Fields like EMS claim to be evidence-based. When our assessments are built on unvalidated cut scores and inconsistent grading, we undermine our own credibility and the public’s trust.

These costs existed before ChatGPT. AI simply made them impossible to ignore.

Teaching Through AI, Not Around It

So what’s the alternative? If banning AI or doubling down on percentages isn’t the answer, what is? We need to rebuild education around competence, judgment, and evidence-based practice, not convenience and tradition.

1. Redefine Mastery Beyond Percentages

Grades should reflect what students can do, not just how many points they accumulated. Instead of arbitrary cutoffs, use criterion-referenced standards that define what it means to be competent. Subject matter experts should regularly review and update these standards based on actual job requirements, not on tradition. In certification exams, methods like Angoff produce cut scores tied to item difficulty and minimal competence . Classroom teachers can adapt similar practices by defining clear performance descriptors for each level of mastery.

2. Design Assessments That Measure Judgment

AI can regurgitate facts and assemble essays. It struggles with nuance, ethics, and context. Our assessments should take advantage of that gap. Scenario-based simulations, oral exams, collaborative projects, and reflective journals require students to make decisions, justify them, and adapt. These tasks aren’t easily automated because they involve human judgment.

3. Teach Students to Evaluate AI Outputs

Pretending AI doesn’t exist leaves students unprepared for a workforce that relies on algorithms. Instead, integrate AI tools into your assignments. Ask students to critique the AI’s response, identify its biases, improve its solution, or compare outputs from multiple models. Make the evaluation of AI part of the learning process. This builds digital literacy and critical thinking.

4. Invest in Teacher Training and Support

We can’t ask instructors to deliver modern, AI-integrated education without equipping them. TVET teachers need both pedagogical and technical training to be effective . They also need support in developing assessments that measure practical skills . For general education teachers, professional development should cover instructional design, cognitive science, formative assessment, and AI literacy. Collaborative professional learning communities can help instructors share strategies that move beyond lecture and multiple-choice.

5. Align Learning With Real-World Outcomes

Whether you’re teaching paramedics or poets, the goal is to prepare learners for life and work. Assessments and assignments should mirror the tasks and challenges students will face after graduation. If a task has no equivalent in the real world, consider whether it’s worth assigning. This alignment makes AI less threatening because the value lies in the thinking and doing, not the formatting.

A New Role for Educators: Guides in a Landscape of Abundant Information

Educators aren’t gatekeepers anymore. They’re guides in a landscape where information is cheap but wisdom is not. AI didn’t eliminate the need for teachers. It eliminated the need for mediocre assignments and places the focus on critical thinking, analysis and subject matter experts. As educators, our job is to help students:

• Question outputs: Who wrote this? Why might it be wrong?
• Identify bias: What perspectives does this model ignore?
• Recognize uncertainty: Where is the edge of the model’s knowledge?
• Apply judgment: How do we adapt a generic answer to a specific context?
• Think ethically: What are the consequences of acting on AI-generated advice?
• Develop skills: Hands-on practice, reflection, and mentorship cannot be automated.

This is deeply human work. It’s also far more rewarding than policing decimal points.

Moving Forward: Repairing the Foundation

AI has given us a choice. We can continue to patch a broken system by adding more surveillance, more point penalties, and more anxiety. Or we can use this moment to rebuild. That requires courage: to challenge tradition, to admit our current metrics aren’t working, and to invest in a better way.

As parents, teachers, and leaders, we can start by:

• Letting go of limiting labels like “70 % pass.”
• Adopting evidence-based assessment strategies that actually improve outcomes.
• Teaching students to think about how they learn—not just what percentage they need.
• Supporting professional development programs that prepare instructors to teach in modern, dynamic environments.
• Pushing our institutions to abandon outdated myths in favor of proven practices.

When we shift from myths to methods, we not only improve learning—we restore trust in education as a profession grounded in evidence. AI is not the enemy of education. It’s the pressure test forcing us to become what we should have been all along.

References

• Guskey, T. R. (2013). The Case Against Percentage Grades. ASCD. Available online. [Highlights research showing that percentage grades are difficult to defend and documenting wide variation in teachers’ scores on identical papers ].
• Vanni, N. (2015). How to Set a Cut Score. Public Safety Group. [Explains that arbitrary cut scores (often 70 %) are harmful and outlines standard-setting methods ].
• Tanaka, N., Angel-Urdinola, D., & Rodon, G. (2023). Teachers in Technical and Vocational Education and Training Are Critical for Successful Workforce Development. World Bank Blog. [Notes that there are no clear paths to becoming a TVET teacher and that industry trainers often lack pedagogical training ; emphasizes the need for a mix of pedagogical and technical training and assessment skills ].