The 98 Percent Trap: When Technology Measures Learning Better Than It Understands It

What if the most dangerous number in education is not zero, but 98?

A nearly perfect score feels reassuring. It suggests accuracy, progress, and control. It gives a school administrator a metric to display, a product company a claim to advertise, and a parent a simple answer to a complicated question: Is this working?

But a number can be precise without being meaningful. A screen can record every tap while missing the hesitation behind it. A classroom application can produce a clean dashboard while weakening the very capacities education is supposed to develop. A wearable can promise an impressive measurement while leaving us unsure what, exactly, has been measured and whether it matters.

The deeper problem is not technology itself. It is our growing habit of confusing measurement with understanding.

The seduction of the measurable

Educational technology and consumer wearables appear to belong to different worlds. One sits in a classroom, the other on a wrist. Yet both participate in the same cultural promise: invisible human processes can be converted into visible data, and visible data can be used to improve behavior.

The promise is compelling. If a student struggles to write by hand because of dysgraphia, a word processor can remove an irrelevant barrier. If a student with dyslexia needs assistance to demonstrate what they know, specialized tools can make assessment fairer. If a teacher can see how each student is progressing, instruction can be adjusted with greater sensitivity.

These are not trivial benefits. They reveal a crucial distinction between technology that expands access and technology that merely increases activity. A tool can give a student a legitimate way to express knowledge that would otherwise remain hidden. It can help a teacher notice patterns that would be difficult to track through memory alone. In such cases, the device is not replacing judgment. It is making better judgment possible.

The danger begins when the metric becomes the purpose.

A product announcement that includes an apparent claim ending in the number 98 illustrates the psychological force of quantified performance. The available claim is incomplete, so we do not know whether 98 refers to accuracy, health, battery performance, or something else. That incompleteness is instructive. Even without knowing the unit, the number feels persuasive. It borrows authority from precision.

This is how data often operates in modern life. The number arrives before the question. We see 98 and instinctively ask whether it is high, rather than asking: Ninety eight of what, measured how, under which conditions, and for what purpose?

In a classroom, the same confusion appears when time spent in an application is treated as evidence of learning, when correct answers are treated as evidence of understanding, or when a polished performance is treated as evidence of independent mastery. The interface makes progress legible, but legibility is not the same as development.

Easy skills, difficult capacities

The most important educational abilities are often the least convenient to measure. They develop slowly, unevenly, and through repeated encounters with difficulty. They include forming an argument, recognizing an unstated assumption, revising an idea, listening carefully, tolerating ambiguity, and making connections across subjects.

These are unconstrained skills. They do not have one correct form or a clear endpoint. They deepen over a lifetime. A person can become a better writer, thinker, or listener for decades without reaching a final level that can be displayed on a dashboard.

Technology tends to perform best where the task is already structured. It can provide immediate feedback on spelling, support text to speech, organize information, display examples, or adapt a sequence of exercises. Such functions are useful precisely because they narrow the problem. The tool knows what counts as an answer, or at least what kind of assistance it is designed to offer.

But the most valuable forms of learning often require the student to confront what cannot be narrowed so easily. Why is this argument weak even though its facts are correct? What evidence would change my mind? Which detail in this story deserves attention? How should I explain an idea to someone who sees the problem differently?

A student can complete a hundred interactive exercises and still avoid these questions. The activity may be energetic, personalized, and constantly rewarded. It may even generate a remarkable success rate. Yet the student may be practicing compliance with a system rather than developing judgment outside it.

This is the central paradox of educational technology: the better a system is at making performance smooth, the more carefully we must ask whether it is preserving the productive friction through which deeper ability grows.

A calculator can help a student explore mathematical relationships, but it can also conceal whether the student understands estimation. A grammar tool can help a student communicate, but it can also prevent the student from noticing why a sentence is unclear. A writing assistant can remove the mechanical burden for someone with a disability, but it can also encourage every writer to outsource the struggle that teaches structure and voice.

The answer is not to romanticize struggle. Unnecessary struggle is not inherently educational. Asking a student with dysgraphia to prove writing ability through handwriting may measure motor difficulty rather than composition. Requiring a student with dyslexia to decode every word without support may test a barrier rather than knowledge.

The relevant question is more precise: What kind of difficulty should remain, and what kind should technology remove?

The barrier test

A useful way to answer that question is to divide difficulty into three categories.

The first is incidental difficulty. This is friction that blocks a person from demonstrating the capacity we actually care about. For a student with dysgraphia, the physical act of handwriting may interfere with the assessment of reasoning and expression. For a student with dyslexia, decoding may obscure comprehension. Assistive technology can remove these barriers without lowering the intellectual demand of the task.

The second is developmental difficulty. This is the challenge through which a skill is built. Revising a paragraph, comparing competing interpretations, or explaining a solution in one’s own words may feel slow and frustrating. Yet removing all of that friction can eliminate the practice itself.

The third is productive uncertainty. This occurs when the learner must make a choice without being given an immediate answer. It is the uncertainty of deciding which evidence matters, generating a hypothesis, or realizing that an initial interpretation does not hold. Productive uncertainty is difficult to quantify, but it is where independent thought begins.

The mistake is to treat all three categories alike. Some schools use technology to remove developmental difficulty, because smoother completion looks like improvement. Other schools preserve incidental difficulty, because tradition makes the barrier appear morally valuable. Both errors confuse the instrument with the goal.

A better principle is the barrier test:

Use technology to remove obstacles that conceal competence. Preserve challenges that create competence.

This principle makes room for both compassion and rigor. It supports assistive tools while resisting the assumption that every moment of frustration is a defect in the learning environment. It also changes how teachers should interpret data. A dashboard can show where a student is struggling, but it cannot by itself explain whether the struggle is a barrier, a developmental necessity, or a sign that the concept has not yet been understood.

That interpretation remains a human task.

Why supervision matters more than the screen

The best use of classroom technology is not autonomous. It is guided, observed, and connected to conversation. A teacher watches what the student does with the tool, asks why a choice was made, and compares digital performance with work produced elsewhere.

Suppose a student receives an excellent score on a reading application. A weak interpretation says: the student has mastered the material. A stronger interpretation asks several questions. Can the student explain the answer without the prompts? Can the student apply the idea to a new passage? Can the student disagree with the program’s framing? Can the student identify why one answer is better than another?

The point is not to distrust every score. It is to place each score inside a portfolio of evidence. Digital results are one signal among several, alongside discussion, drafting, observation, transfer, and reflection.

This resembles the responsible use of a wearable device. A number on a wrist can prompt curiosity about sleep, movement, or recovery. It cannot substitute for asking how a person feels, what changed in their routine, or whether the measurement is reliable in that context. Data can begin an inquiry. It cannot finish one merely by being numerical.

The same logic applies to a claimed performance figure such as 98. Before celebrating it, we need the surrounding structure: the definition, the testing conditions, the comparison point, and the practical significance. A result may be excellent in a laboratory and irrelevant in daily life. It may be accurate for one population and misleading for another. It may track a convenient proxy rather than the underlying experience.

The classroom version of this problem is known as proxy drift. We begin by wanting students to understand. Because understanding is hard to observe directly, we measure completion, speed, or correctness. Eventually, the proxy becomes the target. Students learn to maximize the visible indicator, and the institution congratulates itself when the indicator rises.

Proxy drift is not caused by bad intentions. It is caused by convenience. The thing we can count gradually replaces the thing we care about.

A different model of progress

We need a model of educational technology that distinguishes between three layers of performance.

The first layer is access. Can the learner enter the task, perceive the material, communicate, and participate? Technology can be transformative here. Enlarged text, speech recognition, alternative input, translation, and adaptable presentation can turn exclusion into participation.

The second layer is assisted performance. Can the learner complete the task with support? This layer matters because support can reveal ability that conventional methods hide. It also gives teachers information about which tools enable meaningful participation.

The third layer is independent transfer. Can the learner use the underlying skill in a new setting, with different materials, and without the exact prompts supplied by the tool? This is where durable learning becomes visible.

A system that improves access but never leads toward transfer may be useful accommodation, but it is not sufficient education. A system that produces high assisted performance while weakening independent transfer is more troubling. It creates the appearance of progress while making the learner increasingly dependent on the environment that generated the score.

This three layer model also prevents a cruel misunderstanding. Independence should not mean doing everything unaided. Human beings are always supported by language, tools, teachers, colleagues, and institutions. The goal is not purity from assistance. The goal is agency within assistance: knowing what support is needed, using it deliberately, and understanding the work well enough to adapt when circumstances change.

A student who uses text to speech to access a difficult passage may be exercising more agency than a student who reads it unaided but cannot explain it. A writer who uses speech recognition to compose a sophisticated argument may demonstrate more genuine mastery than a writer whose handwriting looks impeccable but whose ideas are thin.

Technology should therefore be judged not by how little help it provides, but by whether the help expands the learner’s range.

Key Takeaways

1. Ask what the number represents before reacting to the number. Whether the figure is 98 or 48, identify the unit, conditions, population, and practical meaning. Precision without context is often just persuasive decoration.

2. Apply the barrier test. Remove friction that hides a learner’s competence, especially when assistive technology allows a student to demonstrate knowledge more fairly. Preserve difficulty that develops reasoning, expression, and judgment.

3. Treat digital results as signals, not verdicts. Pair application scores with conversation, open ended work, revision, observation, and transfer to unfamiliar problems.

4. Measure independence through adaptation, not isolation. A learner is becoming more capable when support increases agency and flexibility, not merely when support disappears.

5. Audit the proxy. Every time a classroom metric improves, ask whether the underlying capacity improved too. If students are optimizing the score rather than developing the skill, the measurement has become the curriculum.

The future of learning will not be decided by choosing between screens and paper, devices and books, data and intuition. The real choice is between two philosophies of technology.

In one, the learner is adjusted to fit the system. Success means producing clean signals, completing standardized actions, and approaching a target number. In the other, technology is adjusted to fit the learner. Success means gaining access, receiving appropriate support, and gradually expanding the ability to think and act beyond the tool.

The first philosophy worships the metric. The second uses the metric as a clue.

That distinction matters because education is not primarily the production of correct responses. It is the cultivation of people who can recognize what deserves a response, construct one when no template is available, and revise it when reality proves them wrong.

A number such as 98 can tell us something. It can never tell us everything. The mature question is not whether technology delivers impressive measurements. It is whether, after the measurement disappears, the human being has become more capable of seeing, choosing, creating, and understanding.

That is the standard no dashboard can replace.