The Machine That Knows Everything and Understands Nothing: AI, History Class, and the Damage We Keep Refusing to Measure
There’s a seductive logic to putting an AI in front of a history class. History, the thinking goes, is mostly facts—dates, names, treaties, battles—and an AI can recall more facts than any human who ever lived, instantly, patiently, and without ever losing its temper at the back row. Why pay a salary when a chatbot can answer every question about the French Revolution before the bell rings? Why staff a classroom with an expert when you can just tell your student to click “Next”?
The problem is that this logic misunderstands what history is, what teaching is, and what AI actually does. And we are not proposing this experiment in a vacuum. We’ve spent two decades running a quieter version of it already—handing every child a laptop or tablet—and the results are now coming in. They are not encouraging.
We already tried the smaller version of this experiment
Before we get to AI specifically, it’s worth sitting with what the neuroscientist and educator Jared Cooney Horvath lays out in his 2025 book The Digital Delusion. His argument, drawn from international assessments and a large body of cognitive-science research, is blunt: today’s children appear to be less cognitively capable than the generation before them. For most of the twentieth century, IQ scores rose steadily, generation over generation. Around the year 2000, in much of the wealthy world, that rise stalled and reversed. Tellingly, Horvath notes that in places where traditional, low-tech schooling stayed largely intact, the decline didn’t show up. And Horvath isn’t the only one reporting this research – it is backed by classroom teacher after classroom teacher, all who state how technology has impacted their classroom, and how by going back to pen and paper – it has improved.
The most ambitious technology we’ve ever pointed at a classroom is about to be the most persuasive, most authoritative, and most autonomous one yet. If the milder tools already correlate with falling attention and rigor, putting an AI in the teacher’s chair is not a careful next step. It’s flooring the accelerator.
Problem: AI will lie to you with a straight face
The most immediate AI-specific danger is the one the industry has a polite word for: hallucination. Large language models generate text that is statistically plausible, not text that is true. They have no internal mechanism that distinguishes a real event from a convincing fabrication.
In most subjects a wrong answer is at least checkable. In history, the errors are insidious because they look exactly like the truth. An AI will invent a quote and attribute it to Lincoln. It will shift a date by three years, conjure a treaty that was never signed, or describe a meeting between two figures who never met. It delivers all of this in the same calm, authoritative tone it uses for the things that are correct. Expert classroom teachers can identify and remedy these errors. An AI that isn’t sure says nothing different at all—because it is never sure, and never unsure, in the way a person is. For a fourteen-year-old who can’t yet tell a real source from a fake one, that confident wrongness is poison.
Problem: History is not a pile of facts—and treating it as one is the deeper failure
Even setting aside outright errors, the bigger danger is philosophical. History is not fundamentally about what happened. It’s about how we know what happened, whose account we trust, what counts as a cause, and why reasonable people interpret the same evidence differently. The actual skill a history class is meant to build is critical thinking: reading a primary source skeptically, noticing whose voice is missing, weighing rival explanations, and being willing to say “the evidence here is genuinely ambiguous.” Historians argue constantly. That argument is the discipline.
An AI, by design, tends to flatten all of this into a single smooth, consensus-flavored narrative. Ask it “what caused World War I” and you’ll get a tidy paragraph that sounds definitive—precisely the opposite of the messy, evidence-weighing critical thinking required of a real historian.
Many teachers and students are swayed by the “Bells and Whistles” of a computer program. A related failure Horvath called “Duolingo learning”—the confusion of engagement with education, where a slick interface keeps you tapping while very little is retained. A history bot that feels productive and answers everything instantly is the perfect machine for producing students who feel they understand the past and have actually learned to look things up.
Problem: Computer screens impact learning
Computer use in the classroom changes the way the brain functions. The first is attention. The brain can hold only one set of rules at a time; what we call multitasking is really rapid switching, and every switch costs time, accuracy, and memory. Digital environments are engineered for exactly that switching—Horvath’s framing is that learning requires stillness while the business model requires capturing your attention. He notes that students are off task for a startling share of every hour they spend on classroom devices. History, of all subjects, demands sustained attention: following a long argument, holding several causes in mind at once, reading a dense primary source to the end. It is uniquely ill-suited to a medium optimized for the next notification.
The second is transfer. Where you learn becomes part of what you learn; context gets encoded alongside content. Skills built in the narrow, uniform world of a screen don’t reliably survive the jump to a different, harder context.
There’s also a stark, concrete finding here worth flagging: reading comprehension is measurably worse on screens than on paper, especially for the kind of dense, expository text that history runs on, because the physical page gives memory a spatial anchor a scroll bar can’t. Handwritten notes likewise beat typed ones, because handwriting forces the brain to condense and process rather than transcribe. In every case, reading on paper, annotating the reading, writing and interacting with the text, improves learning.
Problem: Bias hides inside the text
An AI’s view of the past is a statistical average of its training data, and that data is wildly uneven. It overrepresents recent, English-language, internet-heavy perspectives and underrepresents nearly everyone else—colonized peoples, oral traditions, non-Western scholarship, the poor and illiterate who left few documents.
The result is a subtle but persistent skew. Marginalized perspectives get smoothed away; contested interpretations get presented as settled. And because the prose is so fluent and even-handed in tone, the bias is far harder to spot than it would be in an obviously slanted textbook. A flawed textbook can at least be criticized, revised, and replaced through a public process. A model’s biases are buried in billions of parameters that no teacher, parent, or school board can inspect.
Problem: AI doesn’t care
History deals in the heaviest material humans have: genocide, slavery, war, famine, the deliberate cruelty of people toward other people. Teaching it well means handling a student’s distress when they grasp what really happened at Auschwitz, fielding a charged question about their own family’s past, knowing when to push and when to sit in silence.
This is the point where Horvath’s research and the case against AI fuse most tightly. He argues that empathy in the classroom isn’t a soft extra—it’s physiological. When two people genuinely interact, their brain activity, heart rate, and breathing begin to synchronize. The relationship between a teacher and a student is one of the most powerful effects in all of education research. A machine has no biology to synchronize with; the connection simply can’t form. He points to the brutal dropout rates of purely online learning, and to the pandemic’s remote-schooling wreckage, as what happens when that human channel is severed. Relationships matter, teachers matter to learning the most.
An AI can generate sympathetic-sounding words about the Holocaust. It does not care that it happened. It cannot read the room, notice the kid who’s gone quiet, or take responsibility when a sensitive topic lands badly. There is no accountable adult—just an interface that will, with equal cheerfulness, narrate an atrocity and recommend a pizza recipe in the next breath. The moral weight of history requires a human who actually carries it.
Problem: The authority problem—and who profits from it
There’s the matter of how students perceive the machine. A textbook is obviously a fallible human artifact. A good teacher openly says “I might be wrong.” But a sleek AI that answers instantly and never hesitates radiates an authority it hasn’t earned. Children are primed to trust it precisely because it never falters—exactly the wrong instinct to cultivate in a subject whose entire point is learning not to trust any single source uncritically.
It’s worth being clear-eyed about why this keeps happening. Education technology generates billions of dollars. Are the engineers creating the technology educational experts? Do they understand the Science of Reading? Critical thinking? Teaching empathy? AI in the classroom is the same bet placed with much higher stakes.
Why does this matter?
Can AI help instruction? As a tool in the hands of a knowledgeable teacher, AI can draft quiz questions, suggest discussion prompts, translate a primary source, or help a student rehearse an argument. Used that way—supervised, double-checked, kept firmly in the role of assistant—it can genuinely help.
Lets be honest – most classroom teachers do not have the autonomy to determine programs that are purchased for them by school districts. Most classroom teachers have to either create their own resources or rely on a textbook driven program. Teachers need to become thoughtful consumers of the resources they are given and examine materials with the same lens they would examine history; with critical thinking, checking for bias, and noting that relationships matter. Teachers cannot be replaced with a computer program, and learning cannot effectively happen with students sitting in a silo in front of a computer. History, of all subjects, is the one that should teach us to distrust an authoritative voice confidently telling us what to think. We should extend that suspicion to the machine, too—and keep a human in the room who knows why it matters.
The danger is specific and worth naming clearly: it’s the move from tool to teacher. The moment we hand a class over to a system that fabricates confidently, flattens interpretation into false certainty, fragments the attention the subject requires, hides its biases in fluent prose, feels nothing about the suffering it describes, and projects an authority it cannot back up, we haven’t upgraded history education. We’ve quietly abolished it and replaced it with something that only looks the same from a distance.
