Lori Balog

The same writing gets rated 30% worse the moment it's labeled "AI." That bias is doing real damage, and schools are where it hurts most.

Lori Balog — Thu, 20 Aug 2026 14:02:42 GMT

Researchers at UC Santa Barbara ran a blunt little experiment. They showed people identical pieces of writing under two labels, “Human Generated” on one and “AI Generated” on the other. Raters preferred the “human” version by more than 30%. It was the exact same text. When the team quietly swapped the labels, the penalty followed the label. Then, in blind tests, nobody could reliably tell which was which.

The bias tracks the label, not the work.

Here’s the thing: people working with AI keep turning out measurably better work. Harvard and BCG ran a field experiment with 758 consultants. The ones that were given GPT-4 produced work that was rated about 40% higher in quality on tasks the model handles well (and worse on tasks it doesn’t, so human judgment still matters). An MIT study published in Science found that professionals finished writing tasks 40% faster with ChatGPT, and quality improved.

None of that output wrote itself. A human decided what to make, judged what came back, caught the errors, and owned the result. We erase that person every time we wave something off as “AI-generated.” There’s a human behind the augmentation, and the augmented version of that human is often doing the best work of their life. If we dismiss the work because a tool touched it, we throw out their growth along with the slop.

Schools are where this bias does its worst damage.

Stanford researchers tested seven AI detectors on essays written by real students, non-native English speakers taking the TOEFL. The detectors falsely flagged 61% of those human-written essays as AI. 97% of the essays got flagged by at least one tool. The Markup documented that there has been a series of false accusations of AI usage directed at international students on work they wrote themselves. Vanderbilt shut off Turnitin’s AI detector after doing the math: even at the company’s own claimed 1% false positive rate, that’s roughly 750 wrongly flagged papers a year at a single university.

Students learn the real lesson fast: to hide it, not disclose it, and not practice using it openly. Even though AI is a tool that will sit inside every job they’re about to walk into. The honest kids get punished, and the cynical ones… adapt.

We are walking away from the upside of AI. Stanford put an AI assistant alongside 900 human tutors working with 1,800 students. Those kids became 4 percentage points more likely to master their math topics. And for the students who were working with the least experienced tutors, they gained up to 9 percentage points. It costs about $20 per tutor per year. That’s a human, augmented, now reaching kids that they couldn’t reach before.

Just a reminder, people said the same things about calculators and spell check. Turns out both were how more people got to do harder work.

I encourage you to judge the work. Ask what the human contributed and whether the thinking holds up. “Did AI touch this” is the wrong question, and every time we use it to dismiss something good, we teach talented people to either lie to us or stop trying.

Sources:

Zhu et al., “Human Bias in the Face of AI” (UC Santa Barbara, ACL Findings 2025)

Dell’Acqua et al., Harvard Business School / BCG field experiment, “Navigating the Jagged Technological Frontier”

Noy & Zhang, “Experimental evidence on the productivity effects of generative AI,” Science (2023)

Liang, Zou et al., “GPT detectors are biased against non-native English writers” (Stanford)

The Markup: AI detection tools falsely accuse international students of cheating

Vanderbilt University: Why we’re disabling Turnitin’s AI detector

Stanford SCALE: Tutor CoPilot study on AI-assisted tutoring

The Four Assumptions Your School Was Built On

Lori Balog — Tue, 18 Aug 2026 14:34:18 GMT

A mid-sized school district gets a windfall of federal recovery money in the spring of 2022. Leaders spend a large share of it on the intervention with the strongest evidence base in education: tutoring three times a week, delivered by trained adults, during the school day. The research is detailed enough that the school board approves it unanimously.

Then the plan meets the building.

Tutoring has to displace something, and everything it might displace is already spoken for. The master schedule is built out of Carnegie units that every course must log. Staffing is budgeted for whole-class instruction, so no adult is free at 10:15. The spring tests won’t capture most of what tutoring builds anyway. By October the sessions have slid to after school, and attendance slows with them. By spring, the program is a line item nobody defends at the board meeting.

The intervention didn’t fail. It lost to the architecture. And every wall it hit was put there on purpose, by capable people, solving a real problem in a different century.

Four choices, made once upon a time, running everything since

American schooling wasn’t discovered. It was engineered, over the course of 70 years, by people solving problems that no longer exist.

Horace Mann took over the Massachusetts Board of Education in 1837 and borrowed the Prussian idea of sorting children into graded classrooms by age, taught by teachers trained at a new kind of institution: the normal school. One teacher in a one-room schoolhouse couldn’t run five ability levels at once. Age-grading fixed that, and it let schools scale.

In 1892 the National Education Association convened the Committee of Ten to standardize what a high school curriculum should even contain. In 1906, the Carnegie Foundation introduced the Carnegie unit: 120 hours of seat time equals one course credit, a fix for the fact that a Massachusetts transcript meant nothing to an Ohio admissions office, and it worked fast. By 1910, nearly every American high school and college used it.

Around the same time, the language of the factory floor moved into school administration. Ellwood P. Cubberley, one of the most widely read education writers of the era, put it plainly in 1916: “our schools are, in a sense, factories, in which the raw products (children) are to be shaped and fashioned into products to meet the various demands of life.” He wasn’t being cynical. He meant it as a compliment to efficiency, and the profession largely agreed with him.

Out of that stretch of decades came four assumptions that still run your local school:

- An expert best teaches by moving a group of similarly aged kids through the same material at the same pace.
- Standardized recall, tested with paper and pencil, measures learning well enough.
- Most students will spend a career in occupations whose skills stay fairly stable, so front-loading their education will serve them for forty years.
- Credentials need to be efficient and comparable, readable by any admissions office or employer without translation.

Each one solved a real bottleneck of 1900: too few trained teachers, no assessment technology beyond a pencil, a labor market that barely moved, and a country that needed one high school diploma to mean the same thing in Boston and Cincinnati.

Each one is now the wrong answer

The bottlenecks are gone, but the assumptions stayed.

Group pacing at a single speed made sense when expertise and printed material were scarce, and there was no way to give a child individual feedback at scale. Neither limit holds anymore. Trained tutors, working with structured programs, already deliver something close to one-on-one instruction inside a normal school day, at real scale. Benjamin Bloom showed back in 1984 that individually tutored students outperformed about 98 percent of conventionally taught classmates. He called mass tutoring too costly to run. Cost was the constraint, not the evidence, and thanks to AI, the cost has moved.

Standardized recall made sense when a pencil-and-paper test was the only assessment tool anyone had. By its own scoreboard, it’s now failing. National reading and math scores for 13-year-olds fell in the years after 2020, with the deepest losses among the lowest performers, and roughly 40 percent of fourth graders now read below the basic level nationally, the highest share in decades. Meanwhile, the jobs growing fastest reward judgment and problem-solving, skills a bubble sheet was never built to catch.

Stable skill demand made sense when a worker’s occupation barely changed shape across a career. Median job tenure in the United States is now 3.9 years, the lowest on record. Nearly a third of the skills required for the average American job changed between 2021 and 2024 alone. A curriculum front-loaded for one stable career track is preparing students for a career that doesn’t exist anymore.

And the credential problem has quietly flipped. The 1906 problem was that transcripts couldn’t be compared. Now they compare perfectly and say almost nothing. A Carnegie unit measures time in a seat, not what a student can do, and the Carnegie Foundation itself said as much in 2013, admitting the unit was “never intended as a measure of, nor was it suitable for assessing, what students learned.” Employers have started acting on that. Nearly half of middle-skill job postings dropped their degree requirement between 2017 and 2019, and several states have stripped degree requirements from state government jobs entirely.

Why you can’t fix one without breaking the others

Here’s the part that makes tinkering useless: these four assumptions don’t sit next to each other. They lean on each other.

Teaching a group of thirty at one pace requires grading them all fast, on one scale. A common scale means testing standardized recall, and standardized recall only makes sense if there’s a stable body of knowledge worth testing. That stability came from a job market where occupational skills barely changed. And slow-changing skills are exactly what made a one-time credential, earned at twenty-two, still mean something at sixty-two.

A bridge built without flex can carry traffic for a long time before it snaps. That’s what this is: four load-bearing assumptions holding each other up, none of which can move alone. Try to fix the credential without touching the pacing, and you’re asking a transcript built for seat time to describe a school that no longer runs on seat time. Try to fix the testing without touching the schedule, and the tutoring story repeats itself: a good idea, unanimously approved, quietly starved by a calendar built for something else.

That’s why the fixes on offer for the last forty years (longer school years, smaller classes, tougher standards, better-aligned tests) haven’t moved the needle. They’re renovations on a structure whose load-bearing walls were never up for debate. Sputnik didn’t touch them. A Nation at Risk didn’t touch them. No Child Left Behind didn’t touch them. A system that survives that many shocks unchanged isn’t broken in the way people assume. It’s doing exactly what it was built to do in 1910.

The next question isn’t how to patch these four assumptions. It’s what replaces them, and what a school looks like once it’s designed for the economy its students will actually enter.

If this is useful, subscribe, and I’ll send the next piece when it’s up. And if you’ve ever experienced a good program losing to a master schedule, I’d like to hear that story in the comments.

Six Things a Machine Can't Do for You

Lori Balog — Thu, 13 Aug 2026 17:22:30 GMT

Three months into kindergarten, a classmate told Maya that the class goldfish grants wishes. She narrowed her eyes and asked how he knew.

Nobody taught her to do that. She’s five. But that little squint, that reflex to ask what’s behind a confident claim, turns out to be one of the most valuable things she owns. More valuable, I’d argue, than almost anything the next thirteen years of school currently plan to give her.

Access to knowledge and the impact on schools

I’ve written before about how knowledge stopped being scarce, and how AI inverted the premise schools were built on. The short version: for most of history, getting knowledge meant proximity to the few people who had it, and school existed to close that gap. Now, a teenager with a phone can get a plausible, often correct, answer in seconds, and the cost of explaining, drafting, and computing is almost nothing.

The more interesting question is what’s still scarce. And the answer is the human abilities that decide what to ask, what to trust, and what to do with an answer. Those were always valuable. Now they’re the whole game.

The six abilities that stayed scarce

Here’s my working list. Each one shares the same profile: AI doesn’t have it, schools have historically treated it as background noise, and AI makes it more valuable rather than less.

1. Discernment

The ability to tell what’s worth attending to, what’s true, and what’s mere fluency. In a world where any prompt produces a polished paragraph, discernment is the first defense against well-formed nonsense. Maya’s goldfish squint is the five-year-old version.

2. Judgment

The ability to act well under uncertainty. To weigh competing considerations and choose when no algorithm can tell you the answer because there isn’t one to look up. Handed the job of passing out scissors, Maya gives the broken pair to no one and instead reports it. Judgment, at the scale available to her.

3. Ethical reasoning

The ability to see what’s morally at stake in a situation, to consider whose interests are affected, and to act in a way you could defend to a thoughtful other. AI systems can simulate ethical reasoning, sometimes impressively. They cannot bear its weight. When the decision goes wrong, no model takes responsibility for it. A person does.

4. Embodied skill

The competence of the trained hand and the practiced body. Music, the visual arts, dance, athletics, laboratory technique, the trades. Anywhere knowing how a thing should work differs from knowing how to make it work. These skills resist automation, and they’re gaining value because of it.

5. Relational trust

The ability to be the person someone else can rely on when things are uncertain. Someone whose word means what it says, whose presence steadies a room. Trust builds between specific human beings over time, and it’s the connective tissue of every institution that has to cooperate under stress, from the operating room to the classroom. AI can mimic the surface of trustworthiness. It cannot bear the consequences of misplaced trust.

6. Adaptive expertise

This one has the deepest research pedigree. Giyoo Hatano and Kayoko Inagaki drew a distinction in 1986 between routine expertise, which executes known procedures efficiently, and adaptive expertise, which can modify, recombine, and invent procedures when the situation demands it. The distinction has aged remarkably well. Routine expertise is exactly what AI automates. The adaptive kind is exactly what it can’t. When Maya’s glue stick runs dry, she flattens the paper to rub the stub sideways; she’s inventing a procedure nobody showed her. That’s the seed of adaptive expertise.

Why AI raises their value instead of lowering it

The intuitive story says AI makes human ability matter less, but the evidence points the other way.

A generation ago, the bottleneck on human productivity was access to information and the ability to execute well-defined cognitive tasks. That bottleneck has moved. The new one is the ability to direct cognitive systems wisely: to decide what should be done, to judge whether what was produced is any good, to take responsibility for the consequences, and to be the kind of person whose judgment others will rely on.

Employers are already saying this. The World Economic Forum’s 2025 Future of Jobs report projects 170 million jobs created and 92 million displaced by 2030, a churn touching 22 percent of all jobs. And when those same employers rank the skills they most need, the top of the list reads like my six in workplace clothing: analytical thinking, resilience and flexibility, leadership and social influence, creative thinking. Manual dexterity and rote computation are projected to decline in importance, because machines now perform them at lower cost than the median worker.

This is basic economics. When one input becomes abundant, its complements become precious. Fluent answers are now abundant. The abilities that decide which answer to trust, which question to ask, and who takes responsibility for the outcome are the complements. Every drop in the price of machine cognition is a rise in the price of human judgment.

But here’s the thing- nearly every kindergartner shows up with all six of these in rough, unpracticed form. A morning in any classroom will give you the full inventory. What varies is whether the next thirteen years treat that inventory as the point of the enterprise, or as charming static around the real work of producing answers.

The answers have been automated. The inventory has not.

The discernment generation

I call this next generation of learners (the ones that will enter the AI-augmented workforce) the discernment generation. The cohort entering kindergarten now will graduate into an economy where fluent answers cost nothing and judging answers is most of the job. They’re the first students whose schooling has to be organized around evaluating machine output rather than competing with it.

Whether they can will depend on whether these six abilities are treated as part of the curriculum or left to chance.

That’s the argument I’ll be building out here, post by post: what each ability looks like in a real classroom, what the learning science says about developing it, and what a school designed around scarcity of judgment rather than scarcity of knowledge would actually look like. If that’s a question you care about, subscribe and follow along. Maya starts kindergarten in September. The clock is running.

It's Not the Phones (Well, Not Only)

Lori Balog — Tue, 11 Aug 2026 21:32:42 GMT

In 2022, average math scores for American nine-year-olds fell for the first time in the roughly fifty-year history of the assessment that tracks them. Not the biggest drop. The first one, ever. The 2024 results confirmed the losses hadn’t been recovered. And this isn’t an American story: PISA, which tests fifteen-year-olds across 81 systems, recorded a math decline between 2018 and 2022 of about 15 points across OECD countries, three times larger than any previous change in its history.

You already know the popular explanation. Jonathan Haidt’s The Anxious Generation argues that smartphones and social media rewired childhood starting around 2010, collapsing adolescent mental health and dragging academics down with it. The book struck a nerve because it’s partly right. The OECD’s own analysis of the PISA data found that students spending more than five to seven hours a day on devices for leisure scored significantly lower in math than moderate users, even after controlling for family income.

So yes, the phones are in the room. But “phones did it” is a diagnosis that asks nothing of the system. The honest version has four parts, and the phones are only one of them.

Failure one: a 1925 structure with a 2025 device in it

American schools still run on their early-twentieth-century chassis. Bell schedules, grade levels, rows of desks, lecture and seatwork, the summative test at the end of the unit. Layering a Chromebook onto that structure doesn’t change the structure. It replaces a worksheet with a screen.

We’ve known this for a decade. The OECD’s 2015 report Students, Computers and Learning found that the countries investing most heavily in classroom computers showed no real improvement in reading, math, or science, and in some cases declines. Andreas Schleicher, the OECD’s education director, put it plainly: “Technology can amplify great teaching, but great technology cannot replace poor teaching.”

It’s like adding indoor plumbing to a house built without it. The space wasn’t designed for the addition; something of the original house gets torn out to make room, and the result never works as well as a house designed with plumbing from the start.

Failure two: teachers were handed the tools and not the training

Mishra and Koehler’s TPACK framework names three kinds of knowledge a teacher has to integrate to use technology well: content, pedagogy, and technology. Two decades of research since then keeps finding the same thing. Most teachers get serious preparation in the first two and almost none in the third. Devices arrived in classrooms by the millions. The professional development that would have made them useful mostly didn’t.

If you teach, I suspect this one needs no citation. Tell me in the comments what your tech rollout actually looked like.

Failure three: the tools were distributed like everything else in American education

Kids from lower-income families live with a strange double reality when it comes to screens. They often rack up more daily phone and screen time than their wealthier peers, yet they’re also the ones most likely to carry what researchers call the homework gap: the disadvantage of a home without reliable broadband or a working device, the two things school actually requires. During the closures of 2020 and 2021, an estimated 15 to 16 million American children couldn’t fully participate in remote instruction, and the pandemic-era losses that NAEP recorded fell hardest on exactly those students. Technology handed out unevenly doesn’t close achievement gaps. It widens the ones income and geography already built.

Failure four: the attention market, which is the part Haidt got right

American adolescents now average more than 8 hours a day on screen media outside of school. The platforms capturing that time are engineered around variable-ratio reward schedules, the same unpredictable-payout design that makes slot machines profitable. That design entrenches habit and dampens exactly the executive control kids are supposed to be developing.

Notice where this failure sits, though: outside the classroom. School inherits the fragmented attention, the lost sleep, the displaced reading time, without having had any hand in producing them. This is the one failure of the four that schools didn’t commit. It’s also the only one the public conversation talks about.

The Diagnosis

Put the four together and you get a more accurate diagnosis than “technology is making children less capable.” Schools pushed powerful tools into structures designed before those tools existed, asked unprepared teachers to make it work, distributed access unevenly, and then watched an attention economy compete for their students’ minds after the bell. The decline in scores reflects the refusal to restructure more than the technology itself.

Used well, by prepared teachers, on fair terms, classroom technology produces small but reliably positive effects on learning. The tool works. The system it was dropped into doesn’t.

If the phones are the whole story, the fix is bans and screen-time rules, and the system gets to stay as it is. If the structure is the story, bans buy us quiet hallways and nothing else, because the 1925 architecture keeps failing with or without the devices. The worst reversals in the history of NAEP and PISA will not be fixed by better measurement of a structure that is itself the problem.

I want to know what you’re seeing, because this is the rare education debate where nearly every reader has firsthand evidence. If you teach: which of the four failures costs your students the most? If you’re a parent: did your kid’s school hand out devices with an actual instructional plan, or did the Chromebook just show up one August? And if you think I’m underweighting Haidt and the phones really are the main event, make that case. I’ll be in the comments.

Three Eras of American School, and Why This One Is Different

Lori Balog — Sun, 09 Aug 2026 12:37:30 GMT

Every school you have ever walked into was built on the same bet: knowledge is scarce, and school is where you go to get it. The teacher at the front, the textbook on the desk, the timed exam at the end, even the long halls and identical classrooms carry the bet in their bones. And, for nearly two centuries, it paid off.

But it just stopped paying.

American public education has reorganized itself around a new purpose roughly every two generations, and each reorganization tracked the economic and informational conditions of its time. There have been three eras so far. Each made sense in its moment, and each one did the same kind of work: it reformed the system. It changed who got taught, how they were sorted, what got measured. None of the three ever asked whether the system itself should exist. That question didn’t come up, because until now, it didn’t need to.

Here’s the argument in one sentence before the history: the first three eras were reforms of a machine that still made sense. AI is not a fourth reform. It’s the moment the machine itself stops making sense, and no amount of adjusting its dials will fix that.

The Common School Era: when books were rare

This era started around 1830. Horace Mann, in his twelve annual reports as Secretary of the Massachusetts Board of Education, argued that a free, tax-supported, non-sectarian common school was the precondition for a functioning republic. Picture the single-room schoolhouse: hard pews, a slate, a teacher barely out of school herself, and the children of farmers, shopkeepers, and recent immigrants seated together.

What those reformers actually did was take knowledge that had belonged to the clergy, the gentry, and apprenticeship networks, and deliver it, imperfectly, to ordinary children. Before the common school, learning meant proximity to someone who already had expertise: a physician, a lawyer, a master craftsman willing to take you on. Most people never got that proximity.

When books were rare and hard to reach, a school that delivered books was revolutionary.

The Cardinal Principles Era: when the factory set the terms

By the early twentieth century, industrialization and immigration had remade the country, and the school reorganized to match. In 1918 the National Education Association’s Commission on the Reorganization of Secondary Education issued Cardinal Principles of Secondary Education, which gave the American high school its aims: health, command of fundamental processes, worthy home membership, vocation, civic education, worthy use of leisure, and ethical character.

The school stopped being just a knowledge-distribution mechanism. It became the instrument by which a diverse, urbanizing population would be sorted and socialized for an industrial labor market. David Tyack called the result “the one best system,” and he showed that its architects borrowed factory and corporate metaphors as actual design language. The bell schedule, the grade levels, the rows: that’s the factory floor, translated for children.

The values that school transmitted were never fully agreed on. Catholics, Lutherans, southern Black communities, and Indigenous families distrusted the project, and the record shows they were right to. The school helped decide which students got which opportunities, and those decisions often ran along lines of race and class. We’re still arguing about schools in the terms that era set.

Still, on its own logic, it worked. When the economy needed interchangeable workers, a school that produced interchangeable graduates was efficient, and it benefited both students and the wider society.

The Standards Era: when the scoreboard took over

Around 1983, A Nation at Risk, the report of the National Commission on Excellence in Education, warned of “a rising tide of mediocrity” and recast schooling as the engine of national economic competitiveness. The decades that followed gave us Goals 2000, the No Child Left Behind Act of 2001, and the Common Core State Standards in 2010.

The through line was measurement: define what students should know, test whether they know it, hold schools responsible for the gap. You can criticize plenty about how that played out. But when the question was whether American students could keep pace with peers in Tokyo and Berlin, a school accountable to standardized measures was at least answering the right question.

Reform, reform, reform, then something else

Look at what all three eras actually share. The common school distributed scarce knowledge. The industrial school sorted people by their ability to execute. The standards school measured that execution against international benchmarks. These were three different fixes with one constant target: information was scarce and it is education’s role to disseminate it.

That’s what reform means here. You keep the premise and change the machinery around it. Nobody in 1830, 1918, or 1983 had reason to question whether school itself, as a system for moving knowledge into heads, was the right tool. It obviously was. The only live question was how to run it better.

AI makes the premise obsolete.

Why AI is not era four

The instinct AI triggers is the same instinct that produced the last three eras: convene a commission, write the report, draft the “AI-ready” standards, run the next reform cycle the way we did in 1918 and 1983. That instinct has been reliable for two centuries. It is about to fail for the first time, because it solves the wrong problem.

A reform assumes the underlying system is still the right one and only needs tuning: better inputs, better sorting, better measurement. A discontinuity means the system itself is what’s obsolete. AI creates discontinuity, and here is why. A teenager with a phone can query the largest aggregation of human knowledge ever assembled and get a plausible, well-formatted, often correct answer in seconds. The marginal cost of an explanation has fallen close to zero, and so has the marginal cost of drafting, summarizing, translating, and computing. The ‘machine’ that Tyack described was built to move knowledge from those who had it to those who didn’t. That machine has run out of a job since knowledge is no longer scarce.

If we bolt AI onto a structure built for scarcity, it will fail in a way we’ve already seen happen. The OECD found in 2015 that countries that invested most heavily in classroom computers showed no improvement in reading, math, or science and, in some cases, even a decline. Layering a Chromebook onto a 1925 instructional model just adds a more expensive worksheet. The recent NAEP and PISA reversals, the worst in the history of both assessments, reflect a system that responded the only way the Standards Era knows how: by measuring harder. But the structure is the real problem, and measuring it more precisely doesn’t address it.

So to say it plainly: the three eras reformed the system, and AI requires redesigning it.

What stays scarce

If knowledge is no longer scarce, what is? It’s the human skills that will never be replaced by AI. Discernment: the ability to tell what’s true from what’s merely fluent. Judgment under uncertainty, when no algorithm can make the call. Ethical reasoning a person can actually stand behind. Embodied skill, the trained hand in the lab, the studio, the trades. The kind of trust between specific people that institutions run on. And what Hatano and Inagaki called adaptive expertise: the ability to invent a new procedure when the known one stops working.

None of these are new; they have just treated them as extras in curriculum, rather than the point.

A school built for 1830 delivered books, and that was enough. A school built for 1918 delivered workers. A school built for 1983 delivered scores. None of those schools needed to teach anyone what to do when the answers were free. A school built for this moment has to deliver something no earlier era ever needed to name: people whose judgment can be trusted precisely because the answers came free.

That’s not a new coat of paint on the same building. It’s a different building.

This essay is adapted from Chapter 1 of my book in progress, The Discernment Generation. Subscribe to follow the argument as it unfolds, and if you’ve watched this play out in your own kids’ classrooms, I’d like to hear about it in the comments.

The Question That Sent Me Back to School

Lori Balog — Wed, 05 Aug 2026 16:16:14 GMT

For years, senior executives asked me versions of the same question:

“How are we going to teach the incoming generation to use AI well when they do not yet have the experience to know when it is wrong?”

It is the right question, and it is much bigger than just an AI training problem.

Over more than two decades in workforce and talent development, including…

Our Schools Were Optimized for a World That No Longer Exists

Lori Balog — Mon, 03 Aug 2026 20:09:39 GMT

Some institutions keep running long after the conditions that justified them have changed. They keep answering yesterday’s questions with yesterday’s tools, until one day someone notices the questions themselves have moved.

For most of human history, the person at the front of the room had something you couldn’t get anywhere else. The teacher or the mast…