Det skal være vanskelig

Erik Børve Rasmussen · article · 2026-07-10 source ↗ Reader 🖥️

practical · argumentative op-ed (NRK Ytring kronikk); normative and prescriptive — A persuasive practical argument about what education should do with AI. Its engine is analogy: it imports Lisanne Bainbridge's 'Ironies of Automation' (1983) and the aviation/nuclear-control-room deskilling literature and maps it onto schooling. Bylined to Erik Børve Rasmussen, but the text self-describes as co-authored ('skriver kronikkforfatterne').

Structure

Key terms

• ønskelige vanskeligheter (desirable difficulties) Desirable Difficulty & Productive Struggle
  Robert Bjork's principle, borrowed as the article's fulcrum: the effortful resistance in learning is not a flaw to be engineered away but the very mechanism by which learning happens.
• autopilot vs. simulator
  The article's central design distinction (from Bainbridge): an autopilot does the job for you and deskills; a simulator forces you to do the job yourself, with feedback, and builds skill. The same AI model can be either, depending on how it is deployed.
• stigen (the ladder)
  The sequence of deliberately-unnecessary tasks whose only value is the capability their steps build. At work AI eats the lower rungs juniors climbed; in education the whole institution is the ladder, so automating the steps leaves no ladder standing.
• biprodukt / produkt (by-product / product)
  An inversion: the essay text is the by-product; the real product of schoolwork is the change in the writer's mind. AI lets students keep the by-product (text) and discard the product (the thinking).
• Ironies of Automation (Bainbridge) Ironies of Automation
  The organizing frame: the paradox that the better automation gets, the more crucial and difficult the human role becomes, while the very skills the human needs atrophy from disuse.
• systemansvar (system responsibility)
  The claim that whether a student meets the autopilot or the simulator cannot be left to a fifteen-year-old's willpower; it must be engineered at the system level through incentives, assessment forms that require visible thinking, and tools built as simulators.

Propositions

• explicitThe government's near-ban on AI in primary school is right but incomplete: a ban removes the autopilot without building the simulator.conf: high
• explicitAI is now better than most people at exactly what school asks students to practice — analysis, writing, and reasoning — so the question is no longer whether AI is good enough for schoolwork.conf: high
• explicitestablishedBainbridge's 'Ironies of Automation' (1983) is the governing precedent: the better the automation, the more crucial and difficult the human's role becomes, and unused skills atrophy — as Air France 447 showed 26 years after she described it.conf: high
  supports ←Human Oversight Is Not a Human in the Loop · p4 — Both cite Bainbridge's 1983 'Ironies of Automation' for the same mechanism: better automation → less practice → harder recovery during rare failures.
  supports ←Should You Still Become a Software Engineer in 2026? GitHub VP · p6 — Both invoke Bainbridge's ironies-of-automation logic: the more capable the tool, the more critical (and more atrophy-prone) the human's residual understanding.
• explicitWhat atrophies under AI is no longer manual hand-skill but thinking itself — writing, analysis, judgment — and unlike a wobbling plane, there is no alarm that warns your judgment has rusted.conf: high
  extendsPresentation: The Ironies of A^2 I^2 (Ironies of Automation and Artificial Intelligence in Incidents) · p2 — Reed (via Bainbridge) holds that manual diagnostic/intervention skills deteriorate when unused. This piece extends the same atrophy claim from physical hand-skill to thinking itself — writing, analysis, judgment — and adds the education-specific twist that in a student the skill does not merely decay, it never forms; there is no alarm for a judgment that never came to be.
• explicitAt work, delegating to AI yields a real product and skill atrophy is a side effect; in school the text was never the point — the point was the change in the writer's mind — so AI keeps the by-product and discards the product.conf: high
• explicitestablishedBjork's 'desirable difficulties' mean the effort in learning is not a flaw but the mechanism — and AI is almost by definition a machine for removing effort.conf: high
  supportsWhat we lose when AI writes for us · p9 — Franco claims the friction and discomfort of writing are not bugs but the mechanism of selfhood. This piece supplies an independent learning-science grounding via Bjork's desirable difficulties: the effort is not a flaw in learning, it is the mechanism — and AI is a machine for removing exactly that effort.
  extends ←Agentic Skill Decay · p3 — Independently arrives at Bjork's desirable-difficulty logic (effort is the learning mechanism, not a flaw) and applies it to professional coding practice rather than school assessment.
  supports ←How I Use Agents Without Stopping My Own Growth · p3 — Currie's plan-first habit is a deliberately reintroduced 'desirable difficulty' — voluntarily doing work an agent would remove, in order to preserve the learning that effort produces.
  supports ←The More Accessible Information Is, the Less Employees Remember · p2 — Supplies direct experimental evidence for Bjork's desirable-difficulties mechanism: lowering the effort of retrieval (via social connectivity, not AI specifically) measurably lowers content retention, exactly as 'the effort is the mechanism' predicts.
• explicitUnlike the calculator, which automated the calculation but not the reasoning it was part of, AI automates the reasoning itself.conf: high
• explicitAI camouflages learning failure the way automation camouflages error: polished submissions disconnect the signal a teacher monitors, so the gap stays hidden until the exam or working life, where it becomes visible — and expensive.conf: high
• explicitA controlled, unassisted exam is an unfalsifiable channel — like the directly-wired analog instruments British nuclear engineers demanded in 1977 — which is why calls for more controlled exam forms are not nostalgia.conf: high
• explicitThe deepest problem is generational: at work AI eats the first drafts and routine cases that were the ladder juniors climbed, but in education the whole institution is the ladder — every task is deliberately something the world doesn't need done, so automating the steps leaves no ladder standing.conf: high
  extendsPresentation: The Ironies of A^2 I^2 (Ironies of Automation and Artificial Intelligence in Incidents) · p4 — Reed warns that automated systems ride on the skills of former manual operators and later generations cannot be expected to have them. This piece extends that generational stock-and-flow to its foundation: education is the replenishment flow for society's judgment, and because in school the whole institution is the ladder, automating the steps depletes the flow at its source.
• explicitJudgment can only be built by doing the work: those who best evaluated AI's answers were those who had done the tasks manually for years, and greater trust in AI correlates with less critical thinking (Microsoft/Carnegie Mellon, 319 knowledge workers).conf: high
  supportsWhy cultivating agency matters more than cultivating skills in the AI era | Max Schoening (Head of Product, Notion) · p13 — Schoening argues taste/judgment is built through iterative reps with feedback, not innate talent. This piece supports the mechanism empirically (those who best judged AI had done the tasks manually for years) and sharpens the bootstrapping tension Schoening leaves open: if AI does the reps, the judgment that would evaluate AI never gets built.
• explicitThe same AI model can be an autopilot (does the job for you) or a simulator (forces you to do the job yourself, with feedback); which one a student meets is decided not by the model but by incentives — and student incentives point to the autopilot: the grade, tonight, with the least pain.conf: high
  refinesWhat we lose when AI writes for us · p1 — Franco frames AI as monolithically corrosive to cognitive struggle ('AI removing cognitive struggle is bad'). This piece refines that: the same model is an autopilot when it removes struggle but a simulator when it imposes it, so whether AI is corrosive is a design-and-incentive choice, not an intrinsic property.
• explicitdevelopedBecause fifteen-year-olds cannot be expected to impose on themselves the difficulty the technology offers to remove, ensuring the simulator over the autopilot is a system responsibility — visible-thinking assessment, simulator-built tools, shielded rooms for unassisted work as training, and measurement through the same unassisted channels.conf: high
  extends ←Teaching Coding When AI Can Write the Code · p10 — Concretizes the 'visible-thinking assessment is a system responsibility' prescription into three specific mechanisms: public studio work, AI-as-assessor conversation, and live performance.
• explicitPersonal one-to-one tutoring is among the most effective known educational interventions but never scaled; AI is the first realistic possibility of delivering it — the upside available only if AI is built as a simulator.conf: high
• implicitcorroboratedThe process, not the product, is the locus of learning — therefore ease is pedagogically corrosive whenever the goal is skill formation: removing the difficulty removes the learning.conf: high
  supports ←Teaching Coding When AI Can Write the Code · p10 — Independently arrives at the same 'process, not product, is the locus of learning' claim and supplies three concrete coding-education implementations of it.
• explicitcorroboratedBainbridge's final irony applies directly: the most successful automated systems require the largest investments in training humans; education is that training, so automating it tears away what was supposed to make automation safe.conf: high
  supports ←Human Oversight Is Not a Human in the Loop · p4 — Same Bainbridge lineage: the essay's institutional-competence-maintenance point (p15) matches this doc's claim that the most successful automated systems require the largest investment in training humans.

Arguments

Narratives

• Air France 447 crashed into the Atlantic in 2009; the pilots had thousands of flight hours, almost all on autopilot, and when the automation disengaged they lacked the hand-skill to fly the aircraft manually.
  → Concrete, high-stakes proof that skills atrophy under automation — the empirical anchor for the whole deskilling frame, made sharper by the claim that Bainbridge had described the accident 26 years before it happened.
• The author ran a course on AI for experienced professionals in spring 2026; those who best managed to evaluate the AI's answers were the ones who had done the tasks manually for years.
  → Firsthand evidence for the claim that the capacity to judge AI is built only by having done the work — links the aviation analogy to knowledge work.
• In 1977, engineers at British nuclear plants proposed that the most important process values be shown on directly-wired, analog instruments — a channel that cannot be falsified.
  → Analogy that defines what a school exam is: an unfalsifiable measurement channel, justifying the call for controlled, unassisted exams.

Questions addressed

• ✅ Should AI be banned from school?
  No. A ban removes the autopilot but does not build the simulator; the answer is to redesign AI tools and assessment so they force visible, unassisted thinking.
• ✅ Why is AI in school categorically different from AI at work?
  At work the output is a real product and skill atrophy is a side effect; in school the output was never the point, so AI discards the actual product — the learning — and keeps only the by-product.
• ✅ How can school capture AI's genuine upside (scalable one-to-one tutoring) without its downside (deskilling)?
  Build AI as a simulator rather than an autopilot and make it a system responsibility via incentives, visible-thinking assessment, and shielded unassisted work.
• ⭕ Which difficulties are 'desirable' (they build the mind) and which are mere friction or busywork?
  Raised implicitly by leaning on Bjork's 'desirable difficulties' but never operationalized — effort is treated as monolithic.

Gaps

• incompleteThe simulator prescription is not shown to be incentive-compatible at scale. The author concedes student incentives point to the autopilot and calls the fix a 'system responsibility,' but never specifies how a system enforces simulator-use or prevents a 'simulator' (a tool that questions your draft, adapts to your mistakes) from being gamed back into a friendlier autopilot.
  Read next: Mechanism-design and assessment-security literature; evaluations of AI tutoring systems (e.g. Khanmigo) that resist gaming; Bjork on how to structure desirable difficulties students will not circumvent.
  closed byTeaching Coding When AI Can Write the Code — Offers a concrete, structural (not self-discipline-dependent) answer to how a 'simulator' can be made gaming-resistant at the point of use: public studio critique makes AI-only work visible to the room, an AI-as-assessor conversational agent follows up until it gets independent evidence of understanding rather than accepting a first answer, and live on-stage coding performance makes real-time paste-and-pray structurally impossible. This is a partial closure only — Freeman's own evidence is uncontrolled, and he concedes the practices work 'especially well' in creative coding's public/performative culture, not shown to generalize to standard, large-scale CS instruction, which is exactly the 'at scale' qualifier the original gap raised.
• incomplete'Desirable difficulty' is asserted but not operationalized. Effort is treated as monolithic, so the argument cannot distinguish difficulties that build the mind (spacing, retrieval, generation) from mere friction, toil, or busywork — not all struggle is generative, and some ease is legitimately good pedagogy.
  Read next: Bjork & Bjork on which specific difficulties are desirable; Ericsson on deliberate practice vs. rote effort; cognitive-load theory on when friction helps vs. overwhelms.
  closed byAgentic Skill Decay — That gap asks for 'desirable difficulty' to be operationalized — distinguishing effort that builds the mind from mere friction — outside the school-assessment context it was raised in. This article names four specific interventions (hypothesize before prompting, ask why, read the diff and predict what could still fail, occasionally work by hand) as the professional-coding analogue, giving Bjork's principle a concrete, practitioner-facing form.
• incompleteThe empirical base is thin for a strong, society-wide generational claim: one 319-worker Microsoft/CMU study (a correlation between AI trust and critical thinking, read as causal) plus the author's own course anecdote.
  Read next: Longitudinal studies of learning outcomes under AI-assisted vs. traditional instruction; the Dreyfus skill-acquisition model; replication of the Microsoft/CMU finding and its direction of causation.
• incompleteThe argument is pinned to Norwegian policy specifics (the government's ban, the Malthe-Sørenssen committee) while advancing a domain-general claim about how minds learn; it does not test whether its prescriptions travel beyond the Norwegian school/university system, nor engage equity between schools that can and cannot build 'simulators.'
  Read next: Comparative ed-tech policy; UNESCO 2023 'Technology in education: A tool on whose terms?'; equity-of-access critiques of school AI infrastructure.

Closes gaps elsewhere

• Addy Osmani’s Post · g1 — Difficulty/effortful reps ARE the mechanism that earns the judgment and taste Osmani names as the durable moat, so removing the reps (agents delivering the first 70-80%) removes the source of the taste — supplying the mechanism the bootstrapping gap points at.
• Putting educators at the center of AI learning · g3 — That extraction flags that the Google.org piece never engages the risks of AI in education — over-reliance, deskilling of students, and AI-mediated learning replacing human reasoning practice. This article is precisely that missing engagement: it argues in detail that AI removes the difficulty that constitutes learning and that in school the deskilling is not decay of an existing skill but the non-formation of one. (It closes the deskilling / reasoning-practice dimension of that gap, not its equity-gap dimension.)

Related documents

• Why cultivating agency matters more than cultivating skills in the AI era | Max Schoening (Head of Product, Notion)
• AI coding creates two kinds of debt. You're only measuring one
  Cognitive Surrender Desirable Difficulty & Productive Struggle Learning Culture & Slack Mental Models & Understanding
• The best response to AI slop, infinite advice, and online noise is from Robin Williams
  Cognitive Surrender Taste & Judgment Writing as Thinking
• Brain Food: Answers Without Effort
  Learning Culture & Slack Mental Models & Understanding Writing as Thinking
• Teaching Coding When AI Can Write the Code
  Cognitive Surrender Desirable Difficulty & Productive Struggle Learning Culture & Slack
• The More Accessible Information Is, the Less Employees Remember
  Cognitive Surrender Desirable Difficulty & Productive Struggle Learning Culture & Slack
• Research: Why Some Junior Employees Work Well with AI—and Others Don’t
  AI Labor Disruption Cognitive Surrender Learning Culture & Slack
• Agentic Skill Decay
  Desirable Difficulty & Productive Struggle Learning Culture & Slack
• AI Is Undermining Leaders’ Judgment. Here’s What to Do About It.
  Cognitive Surrender Ironies of Automation
• Cognitive Surrender
  Cognitive Surrender Mental Models & Understanding