On an early January morning, in the quiet of his Morningside bedroom, Columbia business student Vincent Lo hovers over his MacBook, debating CAiSEY—his Strategy class’s new generative A.I. chatbot.

It’s pretty intuitive to use, he explains.

“First, I’ll usually spend about 40 minutes reading over the case itself, and then will visit the CAiSEY site to start the debate. It will start off posing a high level question.” For example, Should Netflix invest more in original content than third-party licensing? Should Rivian buy, partner, or build its battery supply? “And after each of my answers, it will challenge my thinking by taking the opposite stance.”

Debating—the easy exchange of ideas— isn’t something Lo considers himself particularly good at.

“I’ve had a tendency not to speak up,” he says, thinking back on past classes and jobs. “A tendency to need more time to process my thoughts. Managers have given feedback that I need to be more assertive.”

It was what worried him most at the start of the semester when his professor and developer of CAiSEY, Dan Wang, first introduced it to his class, explaining that each of their conversations would be recorded, graded, and critiqued.

“Just doing something like this,” Lo says, “where you have to talk to an LLM just felt really strange to me. I wasn’t sure if it would be good enough to facilitate a healthy debate like a peer or professor.”

Now, after weeks of its use, he isn’t so reluctant.

In the twenty minutes it takes to complete his assignment, the questions from the artificial tenor come probing and tough. Before, Lo might have paused between replies to collect his ideas, first thinking through the arguments from the lengthy case study. Now, to best mimic the real conversations of corporate finance to which he’d soon be a part, he floats his cursor over the “start talking” button right away, leaving himself little respite.

With CAiSEY, it seems, he was finally finding his voice.

Even as late as last year, classes like Professor Wang’s seemed to many students radical. By then, articles like New York Magazine’s,“Everyone is Cheating Their Way Through College,” still dominated headlines, and Lo and his peers continued to buy into the college’s view on A.I.’s ethical concerns and practical limitations, even while outsourcing much of their brainstorming and outlining—and in some cases, much more— to the likes of Claude and ChatGPT.

But, as Professor Sasha Sidorkin, Chief A.I. Officer for California State University, suggests in his study on the subject, that didn’t last long. Just a year after the rollouts of many Large Language Models (LLMs), rhetoric was discernibly shifting. Nationwide, hundreds of academic institutions that once had expelled or failed students for A.I. use, began partnering in 2023 with major corporations of LLMs and crafting their own A.I. initiatives. In January, Purdue became the first university to impose A.I. graduation requirements. Public school systems like New York City reversed their initial bans, and despite the hardy efforts of a few politicians to stop it, today 300 universities offer degrees in Artificial Intelligence, double the rate from 2022.

“Campus and school district leaders are under enormous pressure from a public that demands A.I. education,” Professor Sidorkin explains. “They think A.I. is the future.”

Not so clear, however, is what effects this new technology might have on his students or if it’s worth it. And yet, as tensions continue to mount, it seems many students like Lo are caught in the confusion, left to gamble—or not—on their own education.

By December, 2022, Professor Dan Wang had already decided: his teaching needed radical change.

For a month, Wang, like many Americans, had followed the news closely around the then-debut Artificial Intelligence, ChatGPT— for a month had watched, not without certain worry and awe, reports of its meteoric rise and human-like capabilities in reasoning and writing. He was impressed, but his concern ultimately lay with the fast-approaching spring semester.

For years Wang has taught a graduate section of business at Columbia University called Technology Strategy, a classic Socratic seminar with discussions and written responses to business hypotheticals. It was the variety of humanities whose early death much of the news then seemed to anticipate.

Preparing for the first day, he soon realized, “Oh, students are just going to ask ChatGPT the question and submit that as their homework. It’s centrally trivial.”

Where before an assignment might have taken 40 minutes to complete and required students to draft complex written arguments, with LLM’s, “it makes it possible to do in less than a second.”

Professor Wang knew he had to innovate. That’s when he decided to try something avant-garde. The next semester, he encouraged students to use it, something that was by 2023, he admits, “already radical.”

But on the first assignment, only a fourth of students accepted— half, if he accounted for no-tells. No small number but hardly the near-ubiquity he expected.

Explanations varied. Some students, he believed, were slow to adopt the new technology, others wary of its mistakes and frequent hallucinations. And then there were the ethicists, students leery of the paradox between Wang’s encouragement and the university’s own censure, a problem that continues to vex many graduates today. Could this technology really better education? Wang wasted little time finding out.

In the summer of 2023, collaborating with a former MBA student, Wang developed the “very janky” proto-type for CAiSEY. It was a tool he modeled on the “Oxbridge System,” a tutorship program pioneered by the sibling universities that emphasize intensive student guidance and supervision. There students work one-on-one with professors, discussing, debating, and thinking through the week’s lessons. It’s a system that’s intimate and personalized. In other words, the antithesis to most big American-styled universities that prioritize lecture halls and broad, general assessments.

“Most universities optimize for one thing, and that’s scale,” Wang says. “The whole point of a lecture is to scale the dissemination of knowledge—written assignments, so that you can scale critical thinking. The trade-off is depth, passivity. When you write an essay, you’re basically writing to nobody. You might get feedback from your instructors, but very likely, they don’t have much time to read your essay.”

With CAiSEY, his thinking went, he wouldn’t need to. It was personalizable and scalable. It was a fix to the long-standing problem of crowded American universities that had cropped up as far back as the 50s after the wave of students from the GI Bill and Higher Education Act. Wang could at last peer behind students’ responses, trace their arguments’ evolution, ask for feedback, and see their thinking, all in short order. And what he found surprised him.

Students were spending four times as long on assignments than Wang expected – not because they were struggling, but because they were thriving. Some who otherwise had difficulty writing out their ideas, like English Language Learners, could clearly and more easily express key concepts; and depending on who dug in on their point or gave ground, Wang could better understand their personalities.

Nevertheless, it was still a prototype. It wouldn’t be until a sabbatical in 2024, when Wang finally partnered with another MBA student, Jill Cohen, and an API voice-set on Github—that warm tenor from Lo’s laptop—that CAiSEY was born. The next semester, in spring 2025, Wang quickly embedded it into his class. And the semester after that, Lo was holding forth to it in the solitude of his bedroom. To date, 1,400 voice conversations have been recorded, according to Wang, with close to 5,000 student uses, among them Lo.

By historical accounts, Professor Wang’s optimism—and his A.I. chatbot—are hardly unusual, says Johnathan Zimmerman, Professor Emeritus of Education and History at the University of Pennsylvania, who currently serves on its A.I. committee.

“Colleges have always been looking for a shiny technological object that’s going to revolutionize instruction,” he says.

In the 1960s, during the Cold War, there was television, a tool that so beguiled techno-optimists that at one point, Zimmerman recalls, on campuses like Penn State, nearly a quarter of all classes were taught via closed-circuit televisions. Even earlier, in the 19th century, there were motion pictures. In an epigraph to his 1986 book, Teachers and Machines, Stanford Professor of Education Larry Cuban gets a kick at quoting a hopelessly naive Thomas Edison on their presumed power.

“I believe that the motion picture is destined to revolutionize our educational system and that in a few years it will supplant largely, if not entirely, the use of textbooks.”

Of course, none of those technologies have completely transformed education. Students still sit in classrooms, take instruction from a teacher, and do homework. In some cases, like the oft-cited calculator, those technologies are eventually embraced, adopted to deepen the subject, consumed and digested—albeit slowly— in the great churn of standard education. In others, like Edison’s motion pictures, they are waylaid, relegated to niche classes like film study or else, set aside as mere diversions.

In each case, though, the three-act play is always the same, says Cuban. A new technology comes up, is heralded as revolutionary for education, and then, just as quickly, discarded or, as is more often the case, adapted for the classroom.

“This is a familiar pattern—a lot of policy talk,” he adds. “But when researchers go into the classroom, you don’t see that many changes. You see a great deal of stability in how teachers teach,” he says.

That stability, explains Cuban, derives from public schools’ model of age-based classrooms which, at least as far back as their introduction in 1840, has stood as a semi-permeant bulwark against the radicalizing forces of society, including tech. Because students are grouped together, one classroom, one teacher, that teacher inherently wields tremendous educational discretion. When a new technology arises—as it inevitably does—some ban it, ignore it, or adopt it—following Everett Roger’s so-called Theory of Innovation. But whatever they do, it is their decision.

“A lot of people do not understand the influence structure has on teachers once they close their classroom doors,” Cuban says.

But then, there are the technological failures that aren’t so benign. Technologies that, despite all their initial promise and pomp, actually cost schools not only much valued time and money, but a whole generation of students learning. For example, one need look no further than the laptop and cellphone scourge of the last thirty years.

In 1996, when then President Bill Clinton first pledged two billion dollars in grants for the Technology Literacy Challenge Fund, the one-computer-one-child idea seemed promising.

For years, schools had been told they were failing, argues Alex Molnar in his book School Commercialism: From Democratic Ideal to Market Commodity. It was a narrative drummed up by the explosive 1983 federal report about America’s declining educational achievements, A Nation at Risk. That failure, writes Molnar, “was premised on the assumption that the ‘monopoly’ held by public schools was the reason for the alleged failure,” and that, “corporations had expertise that could improve the schools.”

“Thus, A Nation at Risk helped create and maintain the pressure on schools to cooperate with—and be seen cooperating with—corporations,” Molnar writes.

Enter: big corporations like Microsoft and IBM and their new technology, the computer.

Here was a technology, EdTech developers argued and President Clinton believed, that would “deepen creativity, expand knowledge, and better prepare students for an ever changing world,” writes neuroscientist Jared Cooney Horvath in his book The Digital Delusion. It was new, powerful, and expansive. A student using the device could stretch their imaginations with learning programs on CD-ROMS, compose essays on Microsoft Word, and by the early 2000s, surf Google. Teachers would save time grading, could effortlessly evaluate student progress, and diversify their lessons. It was the future of work, learning, and achievement.

In a press release as late as 2016, Microsoft touted its Office 365 platforms as flexibility for students to “learn anything and anywhere,” and “fostering strong collaboration.”

“EdTech positioned itself as the savior of learning,” Hovarth writes.

And schools bought in. Soon, computer labs sprung up, laptop programs rolled out, and legislators passed screen-time standards. The gap between the number of students and computers shrank. Today, writes Horvath, 88% of public-school districts now operate one-to-one device programs. In a 2019 Gallup Poll, nearly 60% of students report using digital learning tools in school. In most classrooms, screens are ubiquitous, and though, as Professor Cuban points out, education did not fundamentally change, it did, at the very least, certainly alter.

But the benefits to student learning were less clear. Beginning in the early 2000s, IQ scores began to decline, reversing a steady upward trend since the 1930s, in what is now referred to as the “Reverse Flynn Effect.” Average scores across reading, math, and science that first stalled in the early aughts, fell steadily after 2013, according to National Assessment for Educational Progress, NAEP, 2024 data. In a Survey of Adult Skills released only last year by the OECD, one-third of Americans now struggle with basic reading and math, up ten percentage points since 2004. That same study also reports that the share of 18 year olds that report thinking difficulty, concentrating, and learning new things has markedly risen since the 1990s.

And, as Horvath points out in his book, many of these trends can be directly tied back to the screens. Using data from the Program for International Student Assessment, PISA, Horvath shows that the daily time spent on screens in school is negatively correlated with standardized test scores. For example, students who spent just two hours of screen time in class saw an average score decline of 20 points.

One reason, argues Nicholas Carr, in his Pulitzer Prize winning book, The Shallows: What the Internet Is Doing to Our Brain, could be screen's and the internet's fragmenting nature, which breaks up information, hyperlinks and de-contextualizes it, encouraging shallow reading and multi-tasking, which in turn overloads cognition and short-term memory and dampens the parts of the brain responsible for problem-solving and critical thinking. And all this, of course, before the laptop, tablet, smartphone, and apps, which only double-downed on the internet's ills, further dividing students' attention, reasoning, and understanding.

But many experts would not reckon with this until it was too late, and by then, EdTech was already big business. A rush of EdTech companies were flooding schools promising ever greater attention, ease and flexibility. Classdojo, Kahoot, NewsELA, Accelerate 3000—the litany was endless.

Teachers then had to sift through company products to determine what to use, says Professor Boninger.

By 2024, school districts were spending nearly 30 billion dollars annually on EdTech. Standards or legislation, when they did pass, were slow to arrive or fully supportive.

Then came the pandemic, just another large-scale version of Dr. Zimmerman’s MCOOs, which would precipitate its own massive decline in reading and math scores and further stoke the educational crisis. Following a desperate push to get laptops into the hands of its students, already strapped school districts like Auburn, Maine and Hoboken spent millions on one-to-one laptop programs, only to discard them years later as expensive frivolities. All this to say nothing of smartphones, which have recently seen their own backlash, after a decade of increased rates of teen suicides, school bullying, and overdose deaths.

"We're in an incredibly polarized moment in this country," Dr. Zimmerman said. "But the only thing red and blue agree on right now: letting the kids bring their phones to school was a shitty idea."

As Dr. Horvath testified to members of congress earlier this year, “We have academic research stretching back to 1962 that shows the exact same story for 60 years: when tech enters education, learning goes down.”

But to some proponents of digital technology, like John Watson, Director of The Community Advancing Digital Learning, all that failure of digital tech was foreseeable.

“You’re layering technology into a system that’s really hard to change. It’s kind of like healthcare in that regard.”

In order for technology to work in education, Watson says, it will require a complete shift of Professor Cuban's stable school model; it will require innovative instructional approaches, classroom redesign, and, of course, changes to policy. Fortunately, some of those changes are already here, he says, citing among others, the revisions to educational savings accounts and the private school voucher law in Texas, both of which financially incentivize school choice by siphoning taxpayer funds from public schools.

And yet, for all GenAI’s troubling similarities, many educational institutions seem poised for its embrace. And quickly.

According to a study by the Center on Reinventing Public Education, districts offering teacher development programs around A.I. rose to a high of 86% last year; teachers’ unions like UFT, United Federation of Teachers, have already partnered with major corporations like Anthropic and OpenAI; and some A.I. companies, like MagicSchool, are some of the fastest growing technology products for schools, according to its CEO Adeel Kahn.

Indeed, if Professor Sidorkin’s analysis is correct, the transformation from debut to educational adoption took only three years for major LLMs like ChatGPT, a trend of increasing swiftness some scholars have termed “The Great Compression.” For context, the computer took sixteen years to saturate the market, the internet, only seven, and the smartphone, swifter still in just five. A.I. may very well be nothing new, as Professor Cuban believes, but the speed of its adoption, very much is.

One reason for this may be what Professor Sidorkin calls A.I.'s "self-penetrating quality." Whether teachers wanted it or not, students were already using LLMs in their work within months of ChatGPT’s initial rollout in 2022. A study by College Board reports that between January and May 2025 alone the percentage of high school students who reported using GenAI tools for schoolwork increased from 79% to 84%.

Another reason for such swift adoption: the teachers themselves.

Haunted by the specter of smartphones and laptops, and in an effort to get out ahead, many educators and their institutions have paradoxically begun welcoming in the very same type of technology that ruined education before, all but accelerating the process. One of those organizations: The United Federation of Teachers is an off-shoot of one of the nation’s biggest teachers unions, NYSUT.

Since their 23 million dollar partnership with Microsoft, OpenAI and Anthropic, the organization has regularly hosted A.I. teacher academy workshops, where hundreds of teachers from across the country—often after work and mostly unpaid— gather in the heart of their downtown office to learn, utilize, and adopt the biggest LLMs for their class. The partnership, explains UFT Vice President Mary Vaccaro, is mutual. Trainers and engineers from the platforms guide teachers through best use cases, while teachers provide feedback to the liaisons, which they use to adjust and improve their applications.

To some, the optics may be brow-raising, but it hasn’t swayed teachers, who return in droves for tricks on lesson differentiation and engagement. Most, however, see a more pressing need: the very real need to head-off a problem that seems to have already arrived.

“We don’t want what happened with social media,” Vaccaro says. “We saw what happened, but we just didn’t jump in.”

But quick adoption is not without its own set of problems. As with any new educational program, its effect on students will always be— for a while at least—unknown, leaving many districts to operate in what Susanna Loeb, Educational Researcher at Stanford University, calls a “data vacuum.”

As Director of Stanford’s SCALE Initiative, Loeb has spent the last two years studying A.I. in education and is mostly heartened by its potential use, but she warns, that doesn’t mean there aren’t risks, namely what researchers call A.I.’s potential for cognitive off-loading.

“This is the idea that instead of using it to develop curiosity and to really dive deep into topics, you can instead use it to substitute for thinking. And that kind of use will really get in the way of student development.”

It’s a dangerous drawback that “we have to address before allowing too much of this to go into classes,” she adds.

And yet, in the early days of its release, many students were using it for just that. In a report released in September 2025, University of Southern California researchers found that the majority of students in their survey used A.I. mainly to shortcut assignments, get quick answers, and complete tasks more easily. Indeed, early studies like MIT’s Your Brain on ChatGPT seem only to confirm Loeb’s major concern. Though small and without peer-review, the study nonetheless shows deleterious effects on critical thinking and neural connectivity, particularly among younger participants. That’s something educators—always the proverbial canaries in the coal-mines— have been sounding the alarm on for years, anecdotally, at least.

Still, Loeb acknowledges, the potential for A.I.’s benefits are clear. “It is a transformational technology. It makes communication easier. You can differentiate more easily and more quickly. You can schedule things more responsibly.” A.I. may finally reach the students that need it most.

But, she fears, it also may not. Since, for any technology to really work in the classroom, she says, it needs uptake, students wanting to use it, and that only seems to happen, based on her early research, if an adult is next to them.

“It’s so easy and inexpensive to say ‘oh look, we have this individualized tutoring for students and they can just engage those tools. But you’re not going to get the students who really need the support the most, volunteering to take this up. You need an adult to engage them.” Something it appears the CAiSEY model hadn't foreseen.

Guided by this early research, many educators have begun developing their own policies based on their own set of ethical standards. Only last year in March the UFT released their guardrails with priorities like advancing “equity and fairness” and “democracy.”

Organizations like A.I. for Education, a group dedicated to training a million teachers a year in A.I. use, said in a statement that, “A.I. should help students think more deeply, write more clearly, and reflect more thoughtfully... not bypass the learning process.”

How they plan to ensure that, though, is still unclear. The problem, says Professor Sidorkin, is that right now, no clear definition of “ethical” even exists.

“When you try to explain what that means, none of them actually can. There’s no theory there.”

And it may take years before one emerges. To date, no ethical standards have been adopted by the NYC Department of Education, nor have any national A.I. standards been released.

But what happens when they are?

Many teachers fear that their enforcement will inevitably fall to them, for which many have already expressed a grave unpreparedness. In a conversation at Stanford’s AI+Education Summit last week, Randi Michel, Senior Advisor for Technology for Governor of California, Gavin Newsom clearly agreed. Teachers, she said, will have to do more to deal with “mental health impacts, cyber bullying, cognitive decline, deep fake pornography, non-consensual intimate imagery, and child sexual abuse material.”

Nevertheless, states like Governor Newsom’s continue to adopt new curriculum around the technology, despite what Michel in that same interview acknowledged was a clear absence of evidence for its cognitive benefits.

The state’s adoption is not without its precedence, though, says Mathematics Professor David Klein. A then outspoken critic of the curriculum reform movements of the 90s, Klein can recall when another disruptive technology, the calculator, was a centerpiece for California’s innovative views on curriculum change.

‘The framework argued that calculators should be available to all students at all times in all grades starting in kindergarten, and that there’s too much emphasis on algebra. It was a disaster.”

The problem, Klein argued then, was how the calculator undermined mathematical thinking.

“It distracted from reasoning and simple examples to illustrate big ideas. It’s like studying the trees and not knowing the forest.”

Klein can see a rough analogy with A.I.’s effect on writing and reading today. As the calculator had the potential to replace conceptual thinking of math, so too does A.I. with writing, leading to a generation of students who can write only with the aid of A.I., but can’t read. Or, put another way, can’t see the forest through the trees.

Quick curriculum adoption with any new technology at the center, to Klein, has the real potential to harm students.

After those quick curriculum changes in the 90s, Klein and others have pointed out, “test scores plummeted.” At his own university, California State, Northridge, “there were remedial courses that went down to the elementary level.”

“By the time they got through all the remedial math, they didn’t have time to catch up.”

Years later, California would eventually revise those standards, seeking input from the math department at Stanford, of which Klein was a part, only to revise it several years after that with the introduction of Common Core.

Meanwhile, Klein says, “It was just a lot of wasted lives.”

But no matter how fast educational institutions move on A.I., they may never be fast enough. Its growth, studies now suggest, is exponential, and its capabilities, still more impressive.

In a report published last month, Epoch AI, the nonprofit research institute, found that, “Global A.I. computing capacity is doubling every 7 months.” Stanford’s 2025 AI-Index reported continued significant improvement on A.I. benchmarks. And, as of 2025, CEOs of major A.I. companies, like Perplexity’s Aravind Srinivas, have switched from annual reports to quarterly just to keep pace. In a now-viral blog post, akin to a Paul Revere-esque broadside, Co-founder and CEO of Otherside AI, Matt Schumer, warned of the dangers of rapid advancement.

"The most recent A.I. models make decisions that feel like judgment. They show something that looked like taste: an intuitive sense of what the right call was, not just the technically correct one. A year ago that would have been unthinkable."

For most educational institutions who currently regard A.I. as a tool, this acceleration, explains Pause AI’s CEO, Maxime Fournes, poses a risk.

“Whenever we observe progression on A.I., it doesn’t reach human capacity then plateau. It overshoots,” like in games like chess or shogi, he says. “And at this point, you’re not talking about a tool anymore. You’re talking about a system that just outsmarts you.”

For years Fournes worked as an engineer in A.I. research, but it wouldn’t be until 2022, with the release of ChatGPT 3 and its staggering improvement from its second iteration, that he finally quit to helm the nonprofit, Pause, intent on slowing down A.I. adoption.

As he explains, “We’re not talking about any new technology like electricity or the internet. We are talking about something that automates the entirety of human cognition, and the scale of the implications are quite literally civilizational.”

But pausing exponentially growing technology, he adds, will require exponential foresight, of which people—and educators—are uniquely incapable. Most people plan for today, not the future.

“It seems like most people are just now debating whether or not it’s okay to have students use ChatGPT,” he says, “but they still cannot extrapolate to what will exist in two years or the impact it will have on universities, which will most likely disappear.”

Though, Fournes points out, this may not be entirely education’s fault. LLM’s updates, improvements and gains are getting harder to detect. Where a corporation like OpenAI might have, two years ago, released a major leap in innovation to their ChaptGPT model, now they do so monthly, which he likens to the proverbial frog in boiling water. Companies, too, have tightened their rhetoric, he claims, and distanced themselves from doomsday prophesying and forecasting the potential destruction to many college majors, like Microsoft's 2024 report on the 40 Jobs Most Likely Facing A.I. Disruption. (Number 20 on that list? Business Teacher.)

Though the most significant impact may very well be A.I. companies' lobbying efforts to cozy up to schools, which have totaled nearly 20 million dollars in 2024 from Microsoft, Anthropic and OpenAI alone, with some estimates as high as 120 million in 2025. And efforts appear to be paying off.

In April 2025, President Trump signed the executive order, "Advancing Artificial Intelligence Education for American Youth", which, though not legally binding, creates a coalition of organizations and corporations dedicated to instilling A.I. in education. As of this year, more than 60 organizations have signed on. Eight months later, Trump would sign another executive order, this one banning all statewide A.I. regulations, the legal merits of which are still contested.

The business side, too, appears equally lucrative. According to company data from AIPRM, the A.I. in education market is expected to grow to 89 billion dollars by 2032. Almost one-fifth of OpenAI’s Enterprise users were in education, and education remains one of the fastest growing markets for A.I. products.

“A lot of money’s been invested, a lot of money spent,” says Dr. Boninger. But, “Nothing’s profitable yet. People are going to lose a lot if we don’t all adopt. You may classify that as cynical, but it’s just money working the way it works.”

To educators like Professor Klein, that cynicism is harder to shake.

“The role of technology in education is to make a profit for the company,” he says. “Education is a side effect, if it happens at all.”

Of course, profit alone does not necessarily negate the educational benefits of these and other technologies. As many would argue today, just because corporations like Microsoft and Apple got extraordinarily powerful from the computer, for example, doesn't mean they weren't a boon to education and learning. If corporations can shake up a slow and lumbering bureaucracy and concurrently deliver educational progress, the thinking goes, why not let them?

But as Professor Boninger and others have argued, these come with their own set of risks, most notably companies’ veiled and unexplained decision making, a black box into which the public cannot see. Companies can and often have, she explains, manipulate algorithms, restrict information, and perpetuate ideologies—all without any input from the public, a process that, when incorporated into a public institution like schools, becomes fundamentally anti-democratic and a security threat. As more students come to rely on the information LLMs provide, the easier it becomes for corporations to privatize, gatekeep and control access, argues Jathan Sadowski in his book, The Mechanic and the Luddite.

This access, or lack thereof, worries researcher Loeb at Stanford, who believes the incorporation of A.I. into schools may only perpetuate the already wide academic divide and income-education disparities. He points to places like the Bronx, for example, which have yet to close the first digital divide widened with the introduction of laptops, wifi, and internet access in schools. As more well-off schools and universities receive better training on A.I. and begin using it more sophisticatedly, Loeb explains, other, poorer schools may not, thereby undermining their education and deepening the rich-get-richer narrative, formally known as The Mathew Effect.

The most important thing to remember though, says Professor Boninger, is that many of those who work in A.I. are not educators. They are foremost technology businesses, run by engineers, mathematicians, and entrepreneurs, whose incentives aren't necessarily educators'.

“Everybody thinks they know how to teach because they’ve been in a classroom,” Professor Boninger says. But the truth is that teaching is difficult and requires a tremendous wealth of experience, pedagogy, and familiarity with instruction. Without it, these companies are once again experimenting with a generation of students whose education hangs in the balance.

Today, Professor Zimmerman’s TVclassrooms of the 60s no longer exist. They, along with so many of education’s other failed attempts—Edison’s motion pictures, Keller’s Personal System of Instruction —remain forgotten, relegated to a time of earlier optimism and hopeless naivete. With each one, their advocates were hopeful, giddy even, that finally a new technology had come along to fix a system that failed to deliver on its promise.

But, as Professor Zimmerman pointed out, they all had one Achilles Heel. Students found them “deadly boring.” Those TV courses—or MOOCs as they were known…they may have been efficient, but they didn’t work. They “discounted the human element.”

For now, that human element is still integral to the classroom, says Professor Wang. He loves being a professor and uses CAiSEY as a way to simply deepen discussions, give confidence to the quiet, and democratize ideas.

“There are technologies that engage different forms of communication, and we’re only kind of learning what that is. In so many ways, the human experience is more important for generating compelling ideas and content.”

Though, he admits later, “I can’t rule out that we might develop an A.I. that’s a facsimile of an instructor.”

Back in his bedroom, Lo isn’t worried.

“I’m a huge fan of the concept,” he says. “Obviously there are some small bugs and it's not a perfect solution, but it does the job.”

He acknowledges that writing is still important and not every class would benefit from an A.I. model, but, “I can imagine a ton of other classes that are much more discussion based, perfect for a tool like CAiSEY.”

Hundreds of other students like him, in their quiet bedrooms, debating their little black screens.

Home

Another Brave New World: How A.I. Infiltrated Schools