As is typical for edtech hype, the initial burst of enthusiasm for artificial intelligence in education focused on student-facing applications. Products like IXL, Zearn, and Khan Academy’s chatbot Khanmigo could take on the heavy lifting and personalize instruction for every kid! Who needs tutors, or even teachers, when kids can learn from machines?

Thankfully, the real-life limits of AI instruction surfaced quickly, given how hard it is for non-humanoids to motivate children and teens to pay attention and persist through hard work for any length of time (for example, see “The 5 Percent Problem,” features, Fall 2024). The apps are still popular, but it’s not clear that AI will crowd out live human instruction anytime soon.

If AI can’t replace teachers, maybe it can help them get better at their jobs. Multiple companies are pairing AI with inexpensive, ubiquitous video technology to provide feedback to educators through asynchronous, offsite observation. It’s an appealing idea, especially given the promise and popularity of instructional coaching, as well as the challenge of scaling it effectively (see “Taking Teacher Coaching To Scale,” research, Fall 2018).

While these efforts seem tailor-made for teachers looking to improve, there are clear applications across the spectrum of effectiveness. Like bodycams worn by police, video recordings and attendant AI tools could open a window into every classroom, exposing poor performers to scrutiny and helping to keep bad behavior in check.

Apps for observations

Video-based observations are not new. The underlying, pre-AI idea is for teachers to record themselves providing instruction, choose some of their best samples, and upload those clips to a platform where an instructional coach or principal can watch and provide feedback. Indeed, this model was an important innovation of the Measures of Effective Teaching (MET) project launched in 2009 by the Bill & Melinda Gates Foundation (see “Lights, Camera, Action!” What Next, Spring 2011).

Edthena is one company that has built out a coaching-via-video-feedback service. Its founder, Adam Geller, started as a science teacher in St. Louis before moving on to the national strategy team at Teach For America. At the time, the organization was looking for a way to provide more frequent feedback to its corps members, given growing evidence that the best professional learning comes from educators regularly reviewing, discussing, and critiquing instructional practice together. It’s hard for instructional coaches or principals to visit every teacher’s classroom with much frequency, but recorded lessons allow anyone to observe and deliver feedback anytime from anywhere. That gave Geller an idea, which he later turned into Edthena.

For more than a decade, Geller claims, his platform has narrowed the “feedback gap” dramatically. Research studies find that video coaching via Edthena can improve teacher retention, competence, and confidence. Still, it is a large investment in staff resources. After all, coaches or administrators must find time to watch the videos and offer feedback, and there are only so many hours in the day.

Enter AI. Edthena is now offering an “AI Coach” chatbot that offers teachers specific prompts as they privately watch recordings of their lessons. The chatbot is designed to help teachers view their practice relative to common professional goals and to develop action plans to improve.

To be sure, an AI coach is no replacement for human coaching. An analogy might be the growing number of mental health chatbots on the market, many of them based on cognitive behavioral therapy (CBT), which can help patients reflect on their own thoughts and feelings and help them see things in a more constructive way. In the same way, Edthena’s AI Coach is helping teachers engage in “deep reflection about the classroom teaching,” Geller says. And because the AI tool is responding to teachers’ own self-evaluations, and not the lessons themselves, it’s relatively straightforward to train.

Gathering data for self-improvement

If Edthena is about “deep reflection,” then TeachFX is about hardcore data. The app captures audio recordings from the classroom and uses voice recognition AI to differentiate between teacher and student speech during lessons. Teachers receive visualizations of class time spent on teacher talk, student talk, group talk, and wait time to assess student engagement, as well as more sophisticated analyses of verbal exchanges during class. It’s like a Fitbit for instruction.

TeachFX founder Jamie Poskin, a former high school teacher, got the idea while interviewing a school principal as a Stanford University graduate student. They discussed the challenge of providing feedback to teachers, especially new ones. Recording lessons was intriguing, they agreed, but when could principals find the time to watch the videos? The principal wondered, what if AI could be trained to look for the indicators of good practice—the teacher “moves” that are universally applicable regardless of grade level or subject matter?

The first version of TeachFX focused on a single metric: teacher talk versus student talk, based on voluminous research evidence that the more kids talk during direct instruction, the more they tend to learn. And though classrooms can be cacophonous (especially elementary ones), the technology could readily distinguish between teacher and student voices. Not only were such analyses doable, according to internal company data, but also just turning on the TeachFX app helped teachers more than double the amount of student talk during class. According to the company, almost 80 percent of teachers in a typical implementation use the tool on a recurring basis.

Over time, as the technology has improved, the platform added more metrics aligned with evidence-based best practices. For example: What proportion of a teacher’s questions are open-ended? How long is she waiting for students to answer? A study by Dorottya Demszky and colleagues published in 2023 found that teachers receiving feedback from TeachFX increased their use of “focusing questions,” which prompt students to reflect on and explain their thinking, by 20 percent.

A role for AI in evaluation?

It’s one thing to use AI to provide constructive, no-stakes feedback to teachers about their instructional practice. But what about incorporating it into formal performance evaluations?

Nobody I talked to liked that idea.

Thomas Kane of the Harvard Graduate School of Education, who ran the MET project, said, “AI could make it easier for teachers to get more frequent feedback, without the taint of a supervisory relationship.” But introduce that “supervisory relationship,” and you lose teachers’ willingness to give these technologies a try.

Indeed, neither company founder I spoke with was eager to see their tech used for teacher evaluations. As TeachFX’s Poskin told me, “You want teachers to learn and grow.” The more often teachers upload recordings to the platform, the better. Yet formal evaluations usually only happen every few years. They are the antithesis of constructive feedback.

That said, leaders of both companies welcome teachers’ deciding to use their recordings, or the data and “reflection logs” derived from them, in coaching sessions or formal evaluations. In all cases, the key is leaving those decisions to teachers and letting them keep control of the process and data.

To me, these apps sound like great tools for conscientious teachers eager to improve—as Geller and Poskin no doubt were. But it strikes me that teacher motivation to use them as intended must be an issue, just as it is for students. Teachers are crazy-busy, and apps like these are, ultimately, extra work.

To their credit, some districts provide incentives, such as counting the time teachers spend using the apps against professional learning requirements or allowing recordings to stand in for weekly classroom walkthroughs. Those are steps in the right direction—but we shouldn’t expect uptake to be universal. To me, it seems likely that the worst teachers, who arguably would have the most to gain, are the least likely to engage with these sorts of technologies.

From bodycams to classroom cams

I don’t think it would be crazy, then, for someone to develop a version of this idea that is less about helping well-meaning teachers get better, and more about holding the small number of ineffective teachers accountable. Our schools have long faced the “street-level bureaucrat” problem, coined by political scientist Michael Lipsky in 1969. The idea is that some government services depend so much on the judgment and discretion of people on the ground that it’s hard to evaluate their work or hold them accountable. Teaching is one of those fields; policing is another.

In the world of law enforcement, dash cams and bodycams have changed the equation by providing a clear record of police officers’ interactions with the public, for good or ill. No doubt this has spurred all manner of questions and challenges, such as when to release footage, how to interpret it, and what is admissible in court. Bodycam mandates have garnered some support along with serious concerns about privacy and reliability. But there’s little doubt that police brutality and misconduct face greater scrutiny now than in the past.

So why not bring the same line of thinking into public schools? Put cameras and microphones in every classroom. Turn them on and keep them on. Send the recordings to the cloud and let machine learning do its thing (with strict privacy and security protocols in place, of course). If AI already can differentiate between good and bad questions, surely it can tell principals or department chairs if a teacher starts instruction late and ends it early, or shows movies every Friday, or allows kids to roam the hallways, or makes no effort to stop them from cheating on tests. If such technology could stop the most egregious forms of bad teaching, it might provide a significant boost to student achievement.

Alas, given education politics, that will probably remain just one wonk’s dream. In the meantime, let’s use AI to help as many motivated teachers as possible go from good to great.

Michael J. Petrilli is president of the Thomas B. Fordham Institute, visiting fellow at Stanford University’s Hoover Institution, and an executive editor of Education Next.

When it comes to digital technology, educators and school systems haven’t historically been fleet of foot. But artificial intelligence is partially bucking the trend. Many teachers are embracing it, even as school systems follow form and are moving slowly, or barely at all.

Among the myriad ways school systems can respond, there are two obvious poor choices. On one end of the spectrum, they could turn entirely away from AI—which districts like New York City, Los Angeles, and Seattle initially moved to do. On the other, they could rush to use AI for its own sake rather than for a clear educational purpose. There’s plenty of pressure to put AI in the classroom—both from vendors hawking AI products and superintendents wanting to show bold leadership. It would be all too easy for districts to jump on the AI trend and repeat the mistakes of the past. Remember fads like open classrooms in the 1970s and whole language in the 80s?

AI isn’t like CD-ROMs—it’s a rapidly evolving, transformational technology. School systems should act quickly but strategically to find a sensible, educationally sound path. The best policies will integrate AI with intentionality and help students and schools make progress over the long haul.

What’s the best way forward? Don’t focus on AI. Focus on the problems that matter—and see where AI can help.

Initially Adrift

District responses to AI have been all over the map, and many districts have lurched from one approach to another. Several big-city districts banned ChatGPT almost immediately after it was launched in November 2022. But months later, most had rolled back their bans and instead started to encourage the use of AI.

For example, Walla Walla Public Schools in Washington State initially banned ChatGPT. Then, the district repealed the policy and trained its teachers in how to use AI tools.

“[I was] a little bit red-faced, a little bit embarrassed that we had blocked [ChatGPT] in the spring,” Keith Ross, the district’s director of technology and information services, told a local-news outlet. “[It] really shed light that we need to not wait on this and get moving and find out how to supply the tool to the students.”

Recent surveys of teachers and administrators reveal similar contradictions. In an EdWeek Research Center survey conducted in late 2023, about one in five teachers said their district lacked clear policies regarding AI products, and the same share reported that students are not allowed to use it. That same survey also found that more than half of teachers believe that AI usage in school will grow next year.

A survey of district technology leaders by edtech company eSpark in November 2023 found that only 4 percent of districts had a formal, documented policy governing the use of AI. Thirty-nine percent of respondents said their districts were working on one, but 58 percent said their districts had yet to start developing such a policy. Meanwhile, 87 percent of district technology leaders reported they participated in a webinar or presentation about AI in schools in the past 6 months. Some 52 percent said their teachers were independently incorporating AI into their practice, but only 9 percent said they were doing something systematic with AI.

It’s no wonder why. The AI product landscape is teeming with new options for teachers to try, and few have been thoroughly evaluated by their districts. The barriers to entry to creating an AI education startup are extremely low right now—even if the sustainability and impact of such efforts are open questions. According to Reach Capital, a venture capital firm specializing in education companies, there were at least 280 education tools that “incorporate generative AI as a core engine of their product” as of September 2023. More are emerging every month, and many offer “freemium” access so that teachers can try them for free.

Along with ChatGPT, free AI tools for teachers like MagicSchool and Ethiqly have become integral to the daily work of Rachel Morey, who teaches English Language Arts at Walnut Creek Middle School in the suburbs of Erie, PA. She has used these programs to “brainstorm lesson plans, write tests, create worksheets, adapt texts to meet the needs of diverse learners,” she said, as well as to support students in writing essays and delivering feedback. One of the biggest appeals of AI, she said, is how it helps her save time.

Tools and Guidance Emerge

How can districts close the policy and practice gap? An important first step is safeguarding sensitive student and teacher data and ensuring that clear guidelines are in place regarding plagiarism and academic work. These are separate issues from how schools actually use AI and draw on sophisticated technological and legal expertise. Right now, rather than focusing on detailed specifics—which is almost impossible given how quickly AI is evolving—districts need to level-up and focus on key principles to help educators, students, and administrators use AI-powered products responsibly.

These are complex questions, but districts do not need to figure it all out on their own. In October 2023, the Consortium for School Networking, a professional association for school technology administrators, and the Council of the Great City Schools jointly published a “K–12 Generative AI Readiness Checklist.” The detailed questionnaire covers AI readiness from a half-dozen views, including leadership, data, operational, and legal readiness, and was developed in partnership with Amazon Web Services.

That same month, TeachAI published its “AI Guidance for Schools Toolkit.” The initiative was created by more than 60 individuals, governments, and organizations, including Code.org, ETS, the International Society for Technology in Education, Khan Academy, and the World Economic Forum. Its three-part framework for implementing AI in schools, which starts with guidance and policy to address the risks to learning that AI poses, notes that “the first step should be ensuring that AI use complies with existing security and privacy policies, providing guidance to students and staff on topics such as the opportunities and risks of AI, and clarifying responsible and prohibited uses of AI tools, especially uses that require human review and those related to academic integrity.”

States have gotten in the game as well. The North Carolina Department of Public Instruction, for example, released guidance that prods districts to “review current EdTech providers deploying generative AI to vet their safety, privacy, reliability, and efficacy, to determine if they are appropriate to be used for your school, and which users they will be open to based on their Terms of Service and school or district policies.” Ohio published a five-part AI Toolkit for school districts, which it created with the aiEDU nonprofit organization.

Principles to Design a Path to Progress

Despite the slow pace of district-level policies, it’s also reasonable to worry that districts may move too quickly and rush to use AI without intention, just to say they are doing something with it. According to Scott Muri, superintendent of Ector County Independent School District in Texas, “What’s missing from [several of the frameworks and conversations] around AI is the vision. What are we trying to do or achieve? Where are we going?”

As education thought leader Tom Vander Ark said, “Schools need to shift the primary question from ‘how do we do integrate AI into our school’ to ‘what does great learning look like and how can we use AI to support that? And what kind of work can students do with smart tools?’”

The Readiness Checklist framework thankfully starts there, as the first question asks, “Does the use of Generative Artificial Intelligence (Gen AI) align to achieving your district’s mission, vision, goals, and values?” This isn’t a rhetorical question. The answer may be no.

The risks here are great. Far too often, districts base edtech questions on a search for technology for its own sake. School systems should not frame their efforts as an “AI initiative” unless the focus is how to prepare students for a world with AI or to make sure that schools know how to safeguard against its downsides. Instead, leaders should follow a tried-and-true design thinking process to successfully innovate and put AI to its best use.

That means starting with the problem the district needs to solve and the goal it seeks to achieve. Leaders should ask, is what they’ve identified a priority? Some problems relate to serving mainstream students in core subjects, while others arise because of gaps at the margins, such as not offering a particular elective. Both areas are worthy of innovation. But schools shouldn’t embrace a classroom technology unless it’s saving teachers time, extending their reach, or deepening their understanding of their students.

With the problem or goal identified, school systems then need to be specific about what success would look like. How would they know if they had made progress? What’s the measure they would use?

From there, the focus should be identifying the student and teacher experiences needed to make progress toward the goal. And only then should schools consider the physical and virtual setup to deliver those experiences. In other words, the “stuff”—the content, curriculum, analog and digital technologies, including those powered by AI—should come at the end of the process, not the beginning.

By considering a potential role for AI within this greater context, schools can avoid succumbing to a short-lived fad without sitting on their hands and watching the world pass them by. In these early years of our AI-powered futures, the goal should be measured investments that will stand the test of time.

Michael B. Horn is an executive editor of Education Next, co-founder of and a distinguished fellow at the Clayton Christensen Institute for Disruptive Innovation, and author of From Reopen to Reinvent.

This article appeared in the Summer 2024 issue of Education Next. Suggested citation format:

Horn, M.B. (2024). AI Is Officially Here, There, Everywhere, and Nowhere: Districts playing catch-up can still adopt sound policies for artificial intelligence. Education Next, 24(3), 80-83.

In the fall of 1945, when my father was not quite eight years old, his teacher told my grandmother that he was failing 2nd grade. My father doesn’t remember her reasons, or maybe my grandmother never told him, but the teacher felt he wasn’t ready for 2nd-grade work.

“If he’s not succeeding in 2nd grade,” my grandmother suggested, “why not try him in 3rd?” And she found a tutor, a retired teacher from a different school.

For seven weeks, my father met for an hour a day with the tutor, who gave him homework after each session. The tutor’s charge was to make sure my father mastered the curriculum, not just for 2nd grade but for enough of 3rd grade that he could slip into a 3rd-grade classroom in January 1946, a year early, without needing further help.

But the tutor overdid it. Not only did my father encounter nothing in 3rd grade she hadn’t taught him, but he coasted through 4th and 5th grade as well.

Around 1960, while shopping at Filene’s Basement in downtown Boston, my grandmother ran into an old neighbor—a mom who’d moved away when my grandmother was seeking a tutor to help her son escape from 2nd grade. After bragging about her own family, the neighbor asked if my father was all right.

“He’s fine!” said my grandmother triumphantly. “He’s at Oxford, on a Rhodes Scholarship.”

Stories like this give the impression that tutors can work miracles. For centuries after Aristotle tutored Alexander the Great, certain fortunate individuals—including Albert Einstein, Felix Mendelssohn, Agatha Christie, and practically every British monarch before Charles III—were educated partly or entirely by private tutors and family members. While no scholar regrets the spread of mass schooling, many suspect that the instruction students receive from a teacher in a large classroom can never match the personalized instruction that comes from a tutor focused only on their individual needs.

In a 1984 essay, Benjamin Bloom, an educational psychologist at the University of Chicago, asserted that tutoring offered “the best learning conditions we can devise.” Tutors, Bloom claimed, could raise student achievement by two full standard deviations—or, in statistical parlance, two “sigmas.” In Bloom’s view, this extraordinary effect proved that most students were capable of much greater learning than they typically achieved, but most of their potential went untapped because it was impractical to assign an individual tutor to every student. The major challenge facing education, Bloom argued, was to devise more economical interventions that could approach the benefits of tutoring.

Bloom’s article, “The 2 Sigma Problem,” quickly became a classic. Within two years of its publication, other scholars were citing it weekly—50 times a year—and it has only grown in influence over the decades. In the past 10 years, the article has been cited more than 2,000 times (see Figure 1).

The influence of Bloom’s two-sigma essay reached well beyond the scholarly literature. As the computing and telecommunication revolutions advanced, visionaries repeatedly highlighted the potential of technology to answer Bloom’s challenge. Starting in the 1980s, researchers and technologists developed and eventually brought to market “cognitive computer tutors,” which Albert Corbett at Carnegie Mellon University claimed in 2001 were “solving the two sigma problem.” In the 2010s, improvements in two-way video conferencing let students see human tutors at off hours and remote locations, bringing the dream of universal access closer—though there were still simply not enough tutors to go around.

Then, in late 2022, startling improvements in artificial intelligence offered students a way to converse with software in flexible, informal language, without requiring a human tutor on the other end of a phone or video connection. Sal Khan, founder of Khan Academy, highlighted this promise in a May 2023 TedX talk, “The Two Sigma Solution,” which promoted the launch of his AI-driven Khanmigo tutoring software.

Enthusiasm for tutoring has burgeoned since the Covid-19 pandemic. More than two years after schools reopened, average reading scores are still 0.1 standard deviations lower, and math scores are 0.2 standard deviations lower, on average, than they would be if schools had never closed. The persistence of pandemic learning loss can make it look like an insurmountable problem, yet the losses are just a fraction of the two-sigma effect that Bloom claimed tutoring could produce. Could just a little bit of tutoring catch kids up, or even help them get ahead?

Are Two-Sigma Effects Realistic?

But how realistic is it to expect any kind of tutoring—human or AI—to improve student achievement by two standard deviations?

Two sigmas is an enormous effect size. As Bloom explained, a two-sigma improvement would take a student from the 50th to the 98th percentile of the achievement distribution. If a tutor could raise, say, SAT scores by that amount, they could turn an average student into a potential Rhodes Scholar.

Two sigmas is more than twice the average test score gap between children who are poor enough to get free school lunches and children who pay full price. If tutors could raise poor children’s test scores by that much, they could not only close the achievement gap but reverse it—taking poor children from lagging far behind their better-off peers to jumping far ahead.

Two sigmas also represents an enormous amount of learning, especially for older students. It represents more than a year’s learning in early elementary school—and something like five years’ learning in middle and high school.

It all sounds great, but if it also sounds a little farfetched to you, you’re not alone. In 2020, Matthew Kraft at Brown University suggested that Bloom’s claim “helped to anchor education researchers’ expectations for unrealistically large effect sizes.” Kraft’s review found that most educational interventions produce effects of 0.1 standard deviations or less. Tutoring can be much more effective than that but rarely approaches two standard deviations.

A 1982 meta-analysis by Peter Cohen, James Kulik, and Chen-Lin Kulik—published two years before Bloom’s essay but cited only half as often—reported that the average effect of tutoring was about 0.33 standard deviations, or 13 percentile points. Among 65 tutoring studies reviewed by the authors, only one (a randomized 1972 dissertation study that tutored 32 students) reported a two-sigma effect. More recently, a 2020 meta-analysis of randomized studies by Andre Nickow, Philip Oreopoulos, and Vincent Quan found that the average effect of tutoring was 0.37 standard deviations, or 14 percentile points—“impressive,” as the authors wrote, but far from two sigmas. Among 96 tutoring studies the authors reviewed, none produced a two-sigma effect.

So where did Bloom get the idea that the characteristic benefit of tutoring was two standard deviations? Was there anything behind Bloom’s two-sigma claim in 1984? Why are we still repeating it 40 years later?

What evidence did Bloom have?

Bloom’s Figure 1—reproduced in Khan’s TEDx talk, among many other places—ostensibly showed the distribution of post-test scores for students who received tutoring, comparing them to students who received conventional whole-group instruction and to students who received a version of what Bloom called “mastery learning,” which combined whole-group instruction with individualized feedback. But the graph was only illustrative—hand-drawn in a smooth, stylized fashion to show what a two-sigma effect might look like. It wasn’t fit to actual data.

Later in the essay, Bloom’s Table 1 compared the effects of different educational interventions. Tutoring appeared at the top of the list, with an effect of 2.00 standard deviations. Below tutoring, the table listed reinforcement learning (1.20 standard deviations), mastery learning (1.00 standard deviation) and a variety of other effects that seem startlingly large by modern standards.

Where did Bloom get these large, curiously round estimates? He claimed that he had adapted them from a paper summarizing early meta-analyses published a month earlier by Herb Walberg, a professor at the University of Illinois at Chicago. But Walberg’s and Bloom’s tables do not entirely agree (see Table 1). Although several of Bloom’s estimates lined up with Walberg’s, at least when rounded, most of the effects in Bloom’s table did not appear in Walberg’s, and most of the effects in Walberg’s table did not appear in Bloom’s. And the two professors definitely did not agree on the effect of tutoring.

Walberg didn’t put tutoring at the top of his list, and he estimated tutoring’s effect to be 0.40 standard deviations—close to the average effects reported in meta-analyses. Bloom did repeat Walberg’s estimate of 0.40 standard deviations, but he described it somewhat narrowly as the effect of “peer and cross-age remedial tutoring.” Walberg’s estimate wasn’t so circumscribed; he described it simply as the effect of tutoring.

Bloom relied on two students

Why did Bloom relabel Walberg’s tutoring effect of 0.40, and where did Bloom get his own estimate of 2.00? It seems Bloom was placing his faith in the dissertation studies of two of his PhD students, Joanne Anania and Arthur J. Burke. Both Anania and Burke reported two-sigma effects when comparing tutoring to whole-group classroom instruction—and substantial effects, though not as large, from mastery learning.

Because Anania and Burke provided essentially all the empirical evidence that backed Bloom’s claim of two-sigma tutoring, it’s a little shocking that Bloom didn’t credit them as coauthors. Bloom did cite his students’ dissertations, but if Burke and Anania had been coauthors on an instant classic like “The 2 Sigma Problem,” they might have gotten jobs that provided the resources to conduct further research on tutoring and mastery learning. Instead, Anania published a journal version of her dissertation research, which has been cited just 77 times to date. She taught at three universities in the Chicago area, where she specialized in reading, children’s literature, and adult literacy. Her 2012 obituary doesn’t mention her work on tutoring. Burke never published his dissertation research—or anything else on tutoring. Years later, he published half a dozen reports for the Northwest Regional Laboratory on suspension, expulsion, and graduation—not tutoring.

Bloom also did little work on tutoring after 1984. His next and last major project was an edited book titled Developing Talent in Young People. Published in 1985, the book relied on interviews with accomplished adults to reconstruct how they had developed their talents for music, sculpture, athletics, mathematics, or science. Bloom, who wrote only the introduction, summarized his two-sigma claim in a single paragraph that did not mention Anania or Burke. Bloom retired in 1991 and died in 1999.

It’s a little odd, isn’t it? If these three individuals—two of them just starting their research careers—really discovered a way to raise students’ test scores by two standard deviations, why didn’t they do more with it? Why didn’t they conduct more research? Why didn’t they start a tutoring company?

The two-sigma effect wasn’t just from tutoring

Did Anania and Burke really find two-sigma effects of tutoring? I must admit I was feeling skeptical when I printed out their dissertations. Few 40-year-old education findings hold up well, and student work, half of it unpublished, whose effects have never been replicated, seemed especially unpromising.

To my surprise, though, I found a lot to like in Anania’s and Burke’s dissertations. Both students ran small but nicely designed experiments to test the effect of a thoughtful educational intervention. They randomly assigned 4th, 5th, and 8th graders to receive whole-class instruction, mastery learning, or tutoring. The 4th and 5th graders learned probability; the 8th graders learned cartography. On a post-test given at the end of the three-week experiment, the tutored group really did outscore the whole-class group by two standard deviations on average.

But the tests that students took were very specific. And the tutoring intervention involved a lot more than just tutoring.

Students took a narrow test. Burke and Anania chose the topics of probability and cartography for a specific reason—because those topics were unfamiliar to participating students. There is nothing wrong with choosing an unfamiliar topic; experiments in the science of learning commonly do so. But it’s easier to produce a large effect when students are starting from zero. Cohen, Kulik, and Kulik’s 1982 meta-analysis reported that tutoring effects averaged 0.84 standard deviations when measured on narrow tests developed by the study authors, versus just 0.27 standard deviations when measured on broader standardized tests. In 2020, Matt Kraft reported that effects of educational interventions generally—not just tutoring—are about twice as large when they are evaluated based on narrow as opposed to broad tests.

While Anania’s and Burke’s intervention did achieve two-sigma effects on tests of the material covered in their three-week experiment, it is doubtful that they could achieve similar effects on a broad test like the SAT, which measures years of accumulated skills and knowledge, or on the state math and reading tests that so many parents and teachers have worried about since the pandemic.

Certainly not in three weeks.

Tutored students received extra testing and feedback. Burke’s and Anania’s two-sigma intervention did involve tutoring, but it also had other features. Perhaps the most important was that tutored students received extra testing and feedback. At the end of each unit, all students took a quiz, but any tutored student who scored below 80 percent (in Anania’s study) or 90 percent (in Burke’s) received feedback and correction on concepts that they had missed. Then the tutored students took a second quiz with new questions—a quiz that students in the whole-class condition never received. If the tutored students still scored below 80 or 90 percent, they got more feedback and another quiz.

Bloom acknowledged that his students’ experiments included extra quizzes and feedback, but he asserted that “the need for corrective work under tutoring is very small.” That assertion was incorrect. Clearly the tutored students benefited substantially from feedback and retesting (see Figure 2). For example, in week one of Anania’s experiment, tutored students scored 11 percentage points higher on the retest than they did on the initial test. In week two, tutored students scored 20 percentage points higher on the retest than on the initial test, and in week three, they scored 30 percentage points higher on the retest than on the initial test.

These boosts to performance, and their benefits for longer-term learning, are examples of the testing effect—an effect that, though widely appreciated in cognitive psychology today, was less appreciated in the 1980s. Students learn from testing and retesting, especially if they receive corrective feedback that focuses on processes and concepts instead of simply being told whether they are right or wrong. Burke’s and Anania’s tutors were trained on how to provide effective feedback. Indeed, Burke wrote, “perhaps the most important part of the tutors’ training was learning to manage feedback and correction effectively.” The feedback and retesting also provided tutored students with more instructional time than the students receiving whole-class instruction—about an hour more per week, according to Burke.

How much of the two-sigma effect did the extra testing and feedback explain? About half. You can tell because, in addition to the tutored and whole-class groups, there was a third group of students who engaged in “mastery learning,” which did not include tutoring but did include feedback and testing after whole-class instruction. On a post-test given at the end of the three-week experiment, the mastery-learning students scored about 1.1 standard deviations higher than the students who received whole-class instruction. That’s just a bit larger than the effects of 0.73 to 0.96 standard deviations reported by meta-analyses that have estimated the effects of testing and feedback on narrow tests.

If feedback and retesting accounted for 1.1 of Bloom’s two sigmas, that leaves 0.9 sigmas that we can chalk up to tutoring. That’s not too far from the 0.84 sigmas that the Cohen, Kulik, and Kulik meta-analysis reports for tutoring’s effect on narrow tests.

Tutors received extra training. Extra testing and feedback might have been the most important extra in Anania’s and Burke’s tutoring intervention, but it wasn’t the only extra.

Anania’s and Burke’s tutors also received training, coaching, and practice that other instructors in their experiments did not receive. Burke mentioned training tutors to provide effective feedback, but tutors were also trained “to develop skill in providing instructional cues . . . to summarize frequently, to take a step-by-step approach, and to provide sufficient examples for each new concept. . . . To encourage each student’s active participation, tutors were trained to ask leading questions, to elicit additional responses from the students, and to ask students for alternative examples or answers”—all examples of active, inquiry-based learning and retrieval practice. Finally, “tutors were urged to be appropriately generous with praise and encouragement whenever a student made progress. The purpose of this training was to help the tutor make learning a rewarding experience for each student.”

Although previous tutoring studies had not found larger effects if tutors were trained, the training these tutors received may have been exceptional. Anania and Burke could have isolated the effect of training if they had offered it to some of the instructors in the whole-class or mastery-learning group. Unfortunately, they didn’t do that, so we can’t tell how much of their tutoring effect was due to tutor training.

Tutoring was comprehensive. Many public and private programs offer tutoring as a supplement to classroom instruction. Students attend class with everyone else and then follow up with a tutor afterwards. But the tutoring in Burke’s and Anania’s experiments wasn’t like that. Tutoring didn’t supplement classroom instruction; tutoring replaced classroom instruction. Tutored students received all instruction from their tutors; they didn’t attend class at all. That’s important because, according to Cohen, Kulik, and Kulik’s meta-analysis, tutoring is about 50 percent more effective when it replaces rather than substitutes for classroom instruction.

It’s great, of course, that Burke’s and Anania’s students received the most effective form of tutoring. But it also means that it wasn’t the kind of tutoring that students commonly receive in an after-school or pull-out program.

All That Glitters

My father may have had a two-sigma tutor in 1945. His tutor couldn’t foresee Anania’s and Burke’s experiments, 40 years in the future, but her approach had several components in common with theirs. She met with her student frequently. She was goal-oriented, striving to ensure that my father mastered the 2nd- and 3rd-grade curricula rather than just putting in time. She didn’t yoke herself to the pace of classroom instruction but moved ahead as quickly as she thought my father could handle. And she checked his comprehension regularly—not with quizzes but with short homework assignments, which she checked and corrected to explain his mistakes.

But not all tutoring is like that, and some of what passes for tutoring today is much worse than what my father received in 1945.

In the fall of 2020, I learned that my 5th grader’s math scores had declined during the pandemic. I knew that they hadn’t been learning much math, but the fact that their skills had gone backward was a bit of a shock.

To prepare them for what would come next, I told them the story about my father’s 2nd-grade tutor.

“Grandpa got tutored every day for seven weeks?” they asked me. “That seems excessive.”

“You think so?” I asked.

“Yeah—it’s 47 hours!”

“Come again?” I asked.

They reached for a calculator.

Once a week I drove them to a for-profit tutoring center at a nearby strip mall. It was a great time to be in the tutoring business, but this center wasn’t doing great things with the opportunity. My child sat with four other children, filling out worksheets while a lone tutor sat nearby—available for questions, but mostly doing her own college homework and exchanging text messages with her friends. One day my child told me that they had spent the whole hour just multiplying different numbers by eight. They received no homework. From a cognitive-science perspective, I was pretty sure that practicing a single micro-skill for an hour once a week was not optimal. The whole system seemed designed not to catch kids up, but to keep parents coming back and paying for sessions.

Unfortunately, overpriced and perfunctory tutoring is common. In an evaluation of private tutoring services purchased for disadvantaged students by four large school districts in 2008–2012, Carolyn Heinrich and her colleagues found that, even though districts paid $1,100 to $2,000 per eligible student (40 percent more in current dollars), students got only half an hour each week with a tutor, on average. Because districts were paying per student instead of per tutor, most tutors worked with several children at once, providing little individualized instruction, even for children with special needs or limited English. Students met with tutors outside of regular school hours, and student engagement and attendance were patchy.

Only one district—Chicago—saw positive impacts of tutoring, and those impacts averaged just 0.06 standard deviations, or 2 percentile points.

My grandmother would never have stood for that.

After these results were published, some of Chicago’s most disadvantaged high schools started working with a new provider, Saga Education. Compared to the tutoring services that Heinrich and her colleagues evaluated, Saga’s approach was much more structured and intense. Tutors were trained for 100 hours before starting the school year. They worked with just two students at a time. Tutoring was scheduled like a regular class, so that students met with their tutor for 45 minutes a day, and the way the tutor handled that time was highly regimented. Each tutoring session began with warmup problems, continued with tutoring tailored to each student’s needs, and ended with a short quiz.

The cost of Saga tutoring—$3,500 to $4,300 per student per year—was higher than the programs that Heinrich and her colleagues had evaluated, but the results were much better. According to a 2021 evaluation by Jonathan Guryan and his colleagues, Saga tutoring raised math scores by 0.16 to 0.37 standard deviations. The effect was “sizable,” the authors concluded—it wasn’t two sigmas, but it doubled or even tripled students’ annual gains in math.

Is Two-Sigma Tutoring Real?

The idea that tutoring consistently raises achievement by two standard deviations is exaggerated and oversimplified. The benefits of tutoring depend on how much individualized instruction and feedback students get, how much they practice the tutored skills, and on the type of test used to measure tutoring’s effects. Tutoring effects, as estimated by rigorous evaluations, have ranged from two full standard deviations down to zero or worse. About one-third of a standard deviation seems to be the typical effect of an intense, well-designed program evaluated against broad tests.

The two-sigma effects obtained in the 1980s by Anania and Burke were real and remarkable, but they were obtained on a narrow, specialized test, and they weren’t obtained by tutoring alone. Instead, Anania and Burke mixed a potent cocktail of interventions that included tutoring; training and coaching in effective instructional practices; extra time; and frequent testing, feedback, and retesting.

In short, Bloom’s two-sigma claim had some basis in fact, but it also contained elements of fiction.

Like some science fiction, though, Bloom’s claim has inspired a great deal of real progress in research and technology. Modern cognitive tutoring software, such as ASSISTments or MATHia, was inspired in part by Bloom’s challenge, although what tutoring software exploits even more is the feedback and retesting required for mastery learning. Video tutoring makes human tutors more accessible, and new chatbots have the potential to make AI tutoring almost as personal, engaging, and responsive. Chatbots are also far more available and less expensive than human tutors. Khanmigo, for example, costs $9 a month, or $99 per year.

My own experience suggests that the large language models that undergird AI tutoring, by themselves, quickly get lost when trying to teach common math concepts like the Pythagorean theorem. But combining chatbots’ natural language capabilities with a reliable formal knowledge base—such as a cognitive tutor, a math engine, or an open-source textbook—offers substantial promise.

There is also the question of how well students will engage with a chatbot. Since chatbots aren’t human, it is easy to imagine that students won’t take them seriously—that they won’t feel as accountable to them as my father felt to his tutor and his mother. Yet students do engage and even open up to chatbots, perhaps because they know they won’t be judged. The most popular chatbots among young people are ones that simulate psychotherapy. How different is tutoring, really?

It seems rash, though, to promise two-sigma effects from AI when human tutoring has rarely produced such large effects, and no evidence on the effects of chatbot tutoring has yet been published. Over-promising can lead to disappointment, and reaching for impossible goals can breed questionable educational practices. There are already both human and AI services that will do students’ homework for them, as well as more well-intentioned but still “overly helpful” tutors who help students complete assignments without fully understanding what they’re doing. Such tutors may raise students’ grades in the short term, but in the long run they cheat students of the benefits of learning for themselves.

In the early going, it would be sensible simply to aim for effects that approximate the benefits of well-designed human tutoring. Producing benefits of one-third of a standard deviation would be a huge triumph if it could be done at low cost, on a large scale, and on a broad test—all without requiring an army of human tutors, some of whom may not be that invested in the job. Effects of one-third of a standard deviation probably won’t be achieved just by setting chatbots loose in the classroom but might be within reach if we skillfully integrate the new chatbots with resources and strategies from the science of learning. Once effects of one-third of a standard deviation have been produced and verified, we should be able to improve on them through continuous, incremental A/B testing—slowly turning science fiction into science fact.

Paul von Hippel is a professor and associate dean for research in the LBJ School of Public Affairs at the University of Texas, Austin.

This article appeared in the Spring 2024 issue of Education Next. Suggested citation format:

von Hippel, P.T. (2024). Two-Sigma Tutoring: Separating Science Fiction from Science Fact. Education Next, 24(2), 22-31.

In a little more than a year, freely available artificial intelligence technology has evolved from generating half-right passages of slightly awkward text to creating artistic original images, generating error-free computer code, and even passing an MBA exam at the University of Pennsylvania’s Wharton School. If a user-friendly computer assistant like ChatGPT can already do all of that, AI seems poised to upend traditional work practices and hiring patterns—even when it comes to knowledge-economy jobs.

There are signs that high school and college students around the world are anxious about AI and this uncertain future. While educators fret about plagiarism, cheating, and how to use AI to improve instruction, students are wrestling with more fundamental questions about what they are learning and why. They are looking at the fast-changing world and wondering if their coursework is properly preparing them for the workplaces of tomorrow.

“We’ve seen an increase in… nervousness around students. All of us have moments in school when we’re like, ‘When are we ever going to use this?’,” said Keeanna Warren, CEO of the Purdue Polytechnic High School network of university-affiliated charter schools. “And so now with ChatGPT, students are asking themselves about that with everything. ‘You’re teaching me to write an essay? When am I ever going to use this? You’re teaching me to make a presentation? We’re never going to use this.’ And then they take it to the next level, ‘What am I going to do?’”

At a recent panel discussion at Harvard University about students’ perspectives on generative AI, law student Yusuf Mahmood said he has serious concerns about how well the school is preparing future lawyers to use AI tools for work, especially at big firms. Musicology graduate student Siriana Lundgren said that introductory-level research courses need to change to keep pace with AI’s rapid rise.

High school students share these same worries. Sam Cheng, a junior at Design Tech High School in Redwood City, Calif., said in an interview that AI is just one more technology causing schools to be out of step with what students need. “This problem has been around for a really long time of students feeling like school isn’t preparing me for the real world,” Cheng said. AI, in his view, only adds to that pervasive problem.

Attitudes and Impacts

Despite common assumptions to the contrary, students don’t appear to like or use AI more than parents and teachers. A 2023 survey by the Walton Family Foundation found that 61 percent of parents and 58 percent of K–12 teachers report favorable views of ChatGPT compared to 54 percent of students aged 12–17. Teachers are more likely to use ChatGPT than students, at 63 percent compared to 42 percent.

Recent graduates report feeling threatened and worried by the rise of AI, according to the 2023 edition of the Cengage Group’s annual “Employability Report.” Among 1,000 graduates who had finished a degree or non-degree program in the past month, roughly 46 percent said they felt threatened by AI, and 52 percent said it made them question their preparedness for the workforce.

Meanwhile, workers are voicing the same worries. A 2023 Gallup survey found that 22 percent of U.S. workers are concerned that technology will make their jobs obsolete. That’s a rise of 7 percentage points since the 2021 survey, in a measure that had changed little since Gallup started tracking it in 2017. It’s striking that the increase is due almost entirely to a rise in anxiety among college-educated workers, which suggests that those trained for the knowledge economy don’t feel all that secure.

Although many believe that AI will likely be most powerful when it complements humans, not replaces them, students aren’t wrong to ponder whether schools are preparing them for the futures that will still exist when they leave school. According to the McKinsey Global Institute, because of generative AI “almost 12 million occupational changes will need to take place between now and 2030, with over 80 percent of those jobs falling into four occupations: customer service, food service, production or manufacturing, and office support.”

While students worry about tomorrow, their teachers are applying AI in the classroom today (even if that’s just to check for cheating or plagiarism). But schools have not yet grappled with the broader issue of whether or how curriculum should change.

“I’m concerned that my schools aren’t embracing AI or teaching us how to use it,” said Jared Peterson, a senior at Allen High School in Allen, Tx. “For example, all eight of my teachers have warned us that we can’t use AI for any of our schoolwork. Only one of the eight has encouraged us to experiment with AI, but not on schoolwork. … [And] one of my teachers did a presentation on both the benefits and the dark side of AI technology, but focused more on the dark side of AI.”

The Question of Curriculum

Some forward-looking educators, however, believe that the opportunities associated with AI—and the influence those opportunities should have on curriculum—are more important topics of conversation.

According to Martin West, academic dean at the Harvard Graduate School of Education and editor-in-chief of Education Next, when generative AI burst onto the scene, faculty at his school identified three areas to tackle. First, define a policy on appropriate use to address concerns about cheating. Second, redesign assignments to give students experience using AI in productive ways. And third, consider how course learning goals need to change. Harvard faculty believe that students should be prepared for a professional world where AI use is not just permitted, but expected.

This last priority is ultimately the most important, West said, but it is also the most daunting, which shouldn’t be much of a surprise. After all, rethinking standards and curriculum was fraught even before AI. Just witness how hard it is to alter the K–12 history curriculum or the heated debates in California around what is taught in math.

Yet this is precisely the conversation that needs to take place, said Paul J. LeBlanc, president of Southern New Hampshire University, one of the largest universities in the country. In a recent essay written with Forward College founder and CEO Boris Walbaum, LeBlanc noted that machine learning will be used for many procedural jobs, such as basic accounting and administration, but workers whose work surpasses results from AI will prosper. “Therefore, universities must drastically raise the cognitive bar for students. Less accumulating knowledge and more metacognition: that is, the fundamentals of interacting with knowledge. Learning will move from worrying about what one knows to how one knows it.” In Inside Higher Ed, LeBlanc wrote that the threat posed by AI to high-paying, seemingly secure knowledge economy jobs calls for a paradigm shift and wholesale changes across institutions. “Curricula across a wide range of fields are being rendered out of date at this very moment; we just don’t know in what ways yet.”

Not everyone agrees. If the widespread availability of knowledge means that learning knowledge is no longer important, wouldn’t the emergence of tools like Google and Wikipedia more than two decades ago already have caused that shift? After all, as technology writer Ben Thompson has observed, “It’s important to keep in mind that ChatGPT is a large language model, not a knowledge repository. It has no knowledge of right or wrong, or truth or untruth; it is simply predicting the next word.”

Or, consider this perspective from computer science professor Charles Lee Isbell Jr., an interactive AI expert who is University of Wisconsin–Madison provost and vice chancellor for academic affairs. In response to those who point out that occasionally ChatGPT “hallucinates” and starts “making things up,” he noted: “It’s always making things up. It just so happens that the things that it makes up sound reasonable most of the time.”

And as E.D. Hirsch has previously argued, Google clearly didn’t end of the importance of mastering knowledge. “The Internet has placed a wealth of information at our fingertips,” Hirsch wrote. “But to be able to use that information—to absorb it, to add to our knowledge—we must already possess a storehouse of knowledge.” The ability to think critically relies on having factual knowledge in a given domain.

This observation points to a middle ground of how curriculum may need to change.

New Roles for Knowledge and Experience

One of the biggest changes in work will be knowing how to work alongside generative AI tools, according to Ryan Craig, an author and managing director of Achieve Partners, an education-focused investment firm. To do that, workers will have to know which prompts and questions will generate the right outcomes. And knowing what to ask will require subject-matter expertise. Or, as Craig said in a recent blog post:

If your job’s in claims management, you need to have some understanding of how the insurance industry works and its lexicon. If you’re a digital marketer, you need to know industry-standard platforms, tools, and metrics. Underscoring all this is an ability to understand the subject matter. As specialized LLMs [large language models] evolve for every industry and job function (and likely for each industry-function pairing), experience and pattern recognition will become even more important.

But subject-matter expertise won’t be enough. Individuals will also need to learn how and when to ask the right questions. And that requires a sense of why they are asking the questions they are and what problems they are trying to solve.

Craig’s argument boils down to this: given the rapidly changing nature of work, traditional academic learning from static content is unlikely to make the grade. Instead, academic learning needs to become much more tightly integrated with real work experience given the unpredictable interdependence between the two right now. He noted that a 2023 IBM report on AI predicted, “AI won’t replace people, but people who use AI will replace people who don’t.” Craig concludes: “As a result, keeping students penned in classrooms will impede career launch. While digital transformation has already put a premium on learning-by-doing, AI will make work experience mandatory for every learning journey.”

What does that mean in the classroom? According to Cheng, the California high-school junior, it’s less about teaching “how to use AI” and more about how to take the information and skills that they’ve learned and use them with AI to think critically, creatively, and consciously. “Then even when we’re out of school, even when new technology comes around, we’ll have a toolkit for how to interact with it,” he said.

Beckett Miller, a senior at Design Tech, concurred. “It’s important to have enough knowledge about things and how things work,” he said. But then he argued it’s important to learn how to use tools like AI with conscientious intention. He cited the example of using ChatGPT to help him iterate far faster on an essay he was writing, which ultimately helped him deepen and clarify his thinking, as well as improve the communication of his ideas.

To create opportunities that are connected more tightly with the workplace, high schools could source projects from actual employers as part of the curriculum through companies like Riipen, which pairs college students and curriculum-related internships and jobs. Schools also could allow students to take part in curated internships and externships as part of the regular school year, like the Summit Public Schools charter network in California and Washington State has done with its expeditionary learning blocks. Schools could also turn to organizations like the CAPS Network, which organizes onsite, work-based learning experiences for high school students, to integrate career and technical education for all students. Or schools could offer apprenticeships, akin to what Coweta County in Georgia is doing for sophomores through the Georgia Consortium for Advanced Technical Training. And when schools want to teach about AI itself, they could use up-to-date online courses from places like Coursera rather than seek to reinvent the wheel.

AI is more than a homework helper or fast-track to cheating. It is a transformative tool, and students know it. These sorts of innovations could start to address the concerns of students like Peterson, who worries that high schools “are more focused on cheating and stopping AI usage than on how they can use AI to make education better.”

Michael B. Horn is an executive editor of Education Next, co-founder of and a distinguished fellow at the Clayton Christensen Institute for Disruptive Innovation, and author of From Reopen to Reinvent.

This article appeared in the Spring 2024 issue of Education Next. Suggested citation format:

Horn, M.B. (2024). Artificial Intelligence, Real Anxiety: How should educators use AI to prepare students for the future? Education Next, 24(2), 72-75.

It’s understandable. The education world is awash in articles trying to figure out what artificial intelligence is going to mean for schools and students (see “AI in Education,” features, Fall 2023). But before we get too focused on the latest technological breakthrough, let’s not pretend that we have figured out how to cope with the previous one. Over the last decade, smartphones have become commonplace. Today, 95 percent of American teenagers have a supercomputer in their pocket.

Jonathan Haidt, Jean Twenge, and others have brought necessary attention to the likelihood that smartphones and social media are partly to blame for the teenage mental health epidemic gripping our nation. It’s not a watertight case, because it’s nearly impossible to prove a causal relationship with a phenomenon as ubiquitous as this one.

What scholars can say is that the sudden rise in teenage anxiety and depression, suicidal ideation, and suicide all happened at the same time that teenagers’ adoption of smartphones passed the 50 percent mark—around 2012 or 2013. They can also show that the children most likely to engage in heavy use of smartphones and social media—girls, especially liberal girls—also experienced the greatest increase in mental health challenges. And they can point to other countries that show similar patterns.

My purpose here is not to evaluate this evidence, though I generally agree with Haidt that we should adopt the precautionary principle and assume that phones and social media are likely doing real damage to our kids. Then we should act accordingly.

My immediate question, however, is whether phones and social media might also be behind the plateauing and decline of student achievement that we’ve seen in America, also starting around 2013, long before pandemic-era shutdowns sent test scores over a cliff.

I don’t believe this was the only cause of our achievement woes in the 2010s. As I’ve argued before, I believe the Great Recession was also to blame, both because of its impact on families’ home circumstances, and because of the sudden and significant budget cuts that followed in 2013 and 2014, especially in high-poverty schools. Kirabo Jackson has been particularly persuasive that these spending cuts had a measurable negative impact on achievement (see “The Costs of Cutting School Spending,” research, Fall 2020). Another potential factor was a shift away from school accountability; in 2012 the Obama administration softened the consequences for low test scores targeted by the No Child Left Behind Act. Then in 2015, and Congress replaced it with the Every Student Succeeds Act.

But I do think we need to take the smartphone hypothesis seriously. Especially because, unlike the Great Recession or the pandemic, these trends are not receding in the rearview mirror. Indeed, adolescent phone use continues to rise. If it is one reason that students aren’t learning as much as they did in the pre-smartphone era, that’s a problem we need to grapple with.

So what’s the evidence? First and foremost, as mentioned above, the timing lines up (see Figures 1 and 2). We see smartphone ownership really taking off among adolescents in middle and high school around 2013. That’s also when median achievement on the 8th-grade math test in the National Assessment on Educational Progress (NAEP) peaked. It’s fallen modestly ever since. For our lowest-performing students—those at the 10th and 25th percentiles—the declines were more dramatic.

Another piece of evidence comes from Catholic schools, which serve as a plausible control group for the smartphone hypothesis (see Figure 3). Catholic-school students also take NAEP math and reading tests. But they are not directly impacted by changes in education policy such as the shifts in federal school-accountability rules or cuts in public-school spending. So if Catholic schoolkids also saw achievement declines around 2013, which in fact happened, especially in reading, that could be an indication that something outside education policy is to blame.

But there is also some conflicting evidence. The drops in achievement in the 2010s tended to be for our lowest-achieving students, who are disproportionately poor, Black, Hispanic, and male. And yet, as we know from the studies that Haidt and others point to, phone and social media use was most concentrated among middle-class girls (at least initially). So that doesn’t match up.

Before I conclude with the obligatory call for more research, it’s worth pondering what mechanisms could link smartphone and social media use to lower student achievement. Most obvious are problems around attention, as students’ brains adapt to the rush from “likes,” YouTube videos, TikToks, and other platforms, and then struggle to listen to (much less read) slower-moving and less-vivid presentations, such as the ones they are likely to encounter in class and homework. (Our poor teachers!) Or it could be phones’ impact on mental health; it’s hard to learn when you’re anxious or depressed.

There’s also the issue of sleep (see Figure 4). This is cited in the mental health literature, too, as we know that kids sleep less today than before phones and social media entered the scene, and we also know that there’s a relationship between less sleep and poor mental health.

But so too is there a relationship between less sleep and less student learning. After all, sleep is when the brain works much of its magic, forming connections and cementing ideas in long-term memory. Plus, it’s hard to learn when you’re tired, and it’s really hard to learn when you stay home from school because you have been up much of the night. So there is an angle here that also connects with our chronic absenteeism crisis.

What to make of all of this? If we return to the precautionary principle, the least we can do is try to encourage parents to curb their tweens’ and teens’ phone and social media use. Educators can do their part by setting and enforcing classroom rules that phones be turned off or at least stowed away, unless there’s a compelling instructional reason to use them—though that is admittedly an uphill battle (see “Take Away Their Cellphones,” features, Fall 2022). Abolition is likely impossible, though some legislative proposals to make it harder for kids to access social media apps until they are 16 might help. But schools could certainly encourage parents to limit screen time to a reasonable number of hours per day, be much tougher about earlier bedtimes, and require kids to dock their phones outside their bedroom during sleeping hours. There’s a strong foundation of research to back up any effort to protect and promote students’ sleep, which may help ease some uncomfortable conversations (see “Rise and Shine,” research, Summer 2019).

Indeed, more sleep might be the killer app that could make a huge difference—both for students’ academic achievement and mental health. It’s a good reminder that as we contemplate the future impact of AI on schools and society, what likely matters most aren’t the machines we use but the attention we give to our children’s timeless human needs.

Michael J. Petrilli is president of the Thomas B. Fordham Institute, visiting fellow at Stanford University’s Hoover Institution, and an executive editor of Education Next.

This article appeared in the Winter 2024 issue of Education Next. Suggested citation format:

Petrilli, M.J. (2024). Anxiety, Depression, Less Sleep… and Poor Academic Performance? A decade of smartphone dominance and negative NAEP trends. Education Next, 24(1), 76-79.

In Neal Stephenson’s 1995 science fiction novel, The Diamond Age, readers meet Nell, a young girl who comes into possession of a highly advanced book, The Young Lady’s Illustrated Primer. The book is not the usual static collection of texts and images but a deeply immersive tool that can converse with the reader, answer questions, and personalize its content, all in service of educating and motivating a young girl to be a strong, independent individual.

Such a device, even after the introduction of the Internet and tablet computers, has remained in the realm of science fiction—until now. Artificial intelligence, or AI, took a giant leap forward with the introduction in November 2022 of ChatGPT, an AI technology capable of producing remarkably creative responses and sophisticated analysis through human-like dialogue. It has triggered a wave of innovation, some of which suggests we might be on the brink of an era of interactive, super-intelligent tools not unlike the book Stephenson dreamed up for Nell.

Sundar Pichai, Google’s CEO, calls artificial intelligence “more profound than fire or electricity or anything we have done in the past.” Reid Hoffman, the founder of LinkedIn and current partner at Greylock Partners, says, “The power to make positive change in the world is about to get the biggest boost it’s ever had.” And Bill Gates has said that “this new wave of AI is as fundamental as the creation of the microprocessor, the personal computer, the Internet, and the mobile phone.”

Over the last year, developers have released a dizzying array of AI tools that can generate text, images, music, and video with no need for complicated coding but simply in response to instructions given in natural language. These technologies are rapidly improving, and developers are introducing capabilities that would have been considered science fiction just a few years ago. AI is also raising pressing ethical questions around bias, appropriate use, and plagiarism.

In the realm of education, this technology will influence how students learn, how teachers work, and ultimately how we structure our education system. Some educators and leaders look forward to these changes with great enthusiasm. Sal Kahn, founder of Khan Academy, went so far as to say in a TED talk that AI has the potential to effect “probably the biggest positive transformation that education has ever seen.” But others warn that AI will enable the spread of misinformation, facilitate cheating in school and college, kill whatever vestiges of individual privacy remain, and cause massive job loss. The challenge is to harness the positive potential while avoiding or mitigating the harm.

What Is Generative AI?

Artificial intelligence is a branch of computer science that focuses on creating software capable of mimicking behaviors and processes we would consider “intelligent” if exhibited by humans, including reasoning, learning, problem-solving, and exercising creativity. AI systems can be applied to an extensive range of tasks, including language translation, image recognition, navigating autonomous vehicles, detecting and treating cancer, and, in the case of generative AI, producing content and knowledge rather than simply searching for and retrieving it.

“Foundation models” in generative AI are systems trained on a large dataset to learn a broad base of knowledge that can then be adapted to a range of different, more specific purposes. This learning method is self-supervised, meaning the model learns by finding patterns and relationships in the data it is trained on.

Large Language Models (LLMs) are foundation models that have been trained on a vast amount of text data. For example, the training data for OpenAI’s GPT model consisted of web content, books, Wikipedia articles, news articles, social media posts, code snippets, and more. OpenAI’s GPT-3 models underwent training on a staggering 300 billion “tokens” or word pieces, using more than 175 billion parameters to shape the model’s behavior—nearly 100 times more data than the company’s GPT-2 model had.

By doing this analysis across billions of sentences, LLM models develop a statistical understanding of language: how words and phrases are usually combined, what topics are typically discussed together, and what tone or style is appropriate in different contexts. That allows it to generate human-like text and perform a wide range of tasks, such as writing articles, answering questions, or analyzing unstructured data.

LLMs include OpenAI’s GPT-4, Google’s PaLM, and Meta’s LLaMA. These LLMs serve as “foundations” for AI applications. ChatGPT is built on GPT-3.5 and GPT-4, while Bard uses Google’s Pathways Language Model 2 (PaLM 2) as its foundation.

Some of the best-known applications are:

ChatGPT 3.5. The free version of ChatGPT released by OpenAI in November 2022. It was trained on data only up to 2021, and while it is very fast, it is prone to inaccuracies.

ChatGPT 4.0. The newest version of ChatGPT, which is more powerful and accurate than ChatGPT 3.5 but also slower, and it requires a paid account. It also has extended capabilities through plug-ins that give it the ability to interface with content from websites, perform more sophisticated mathematical functions, and access other services. A new Code Interpreter feature gives ChatGPT the ability to analyze data, create charts, solve math problems, edit files, and even develop hypotheses to explain data trends.

Microsoft Bing Chat. An iteration of Microsoft’s Bing search engine that is enhanced with OpenAI’s ChatGPT technology. It can browse websites and offers source citations with its results.

Google Bard. Google’s AI generates text, translates languages, writes different kinds of creative content, and writes and debugs code in more than 20 different programming languages. The tone and style of Bard’s replies can be finetuned to be simple, long, short, professional, or casual. Bard also leverages Google Lens to analyze images uploaded with prompts.

Anthropic Claude 2. A chatbot that can generate text, summarize content, and perform other tasks, Claude 2 can analyze texts of roughly 75,000 words—about the length of The Great Gatsby—and generate responses of more than 3,000 words. The model was built using a set of principles that serve as a sort of “constitution” for AI systems, with the aim of making them more helpful, honest, and harmless.

These AI systems have been improving at a remarkable pace, including in how well they perform on assessments of human knowledge. OpenAI’s GPT-3.5, which was released in March 2022, only managed to score in the 10th percentile on the bar exam, but GPT-4.0, introduced a year later, made a significant leap, scoring in the 90th percentile. What makes these feats especially impressive is that OpenAI did not specifically train the system to take these exams; the AI was able to come up with the correct answers on its own. Similarly, Google’s medical AI model substantially improved its performance on a U.S. Medical Licensing Examination practice test, with its accuracy rate jumping to 85 percent in March 2021 from 33 percent in December 2020.

These two examples prompt one to ask: if AI continues to improve so rapidly, what will these systems be able to achieve in the next few years? What’s more, new studies challenge the assumption that AI-generated responses are stale or sterile. In the case of Google’s AI model, physicians preferred the AI’s long-form answers to those written by their fellow doctors, and nonmedical study participants rated the AI answers as more helpful. Another study found that participants preferred a medical chatbot’s responses over those of a physician and rated them significantly higher, not just for quality but also for empathy. What will happen when “empathetic” AI is used in education?

Other studies have looked at the reasoning capabilities of these models. Microsoft researchers suggest that newer systems “exhibit more general intelligence than previous AI models” and are coming “strikingly close to human-level performance.” While some observers question those conclusions, the AI systems display an increasing ability to generate coherent and contextually appropriate responses, make connections between different pieces of information, and engage in reasoning processes such as inference, deduction, and analogy.

Despite their prodigious capabilities, these systems are not without flaws. At times, they churn out information that might sound convincing but is irrelevant, illogical, or entirely false—an anomaly known as “hallucination.” The execution of certain mathematical operations presents another area of difficulty for AI. And while these systems can generate well-crafted and realistic text, understanding why the model made specific decisions or predictions can be challenging.

The Importance of Well-Designed Prompts

Using generative AI systems such as ChatGPT, Bard, and Claude 2 is relatively simple. One has only to type in a request or a task (called a prompt), and the AI generates a response. Properly constructed prompts are essential for getting useful results from generative AI tools. You can ask generative AI to analyze text, find patterns in data, compare opposing arguments, and summarize an article in different ways (see sidebar for examples of AI prompts).

One challenge is that, after using search engines for years, people have been preconditioned to phrase questions in a certain way. A search engine is something like a helpful librarian who takes a specific question and points you to the most relevant sources for possible answers. The search engine (or librarian) doesn’t create anything new but efficiently retrieves what’s already there.

Generative AI is more akin to a competent intern. You give a generative AI tool instructions through prompts, as you would to an intern, asking it to complete a task and produce a product. The AI interprets your instructions, thinks about the best way to carry them out, and produces something original or performs a task to fulfill your directive. The results aren’t pre-made or stored somewhere—they’re produced on the fly, based on the information the intern (generative AI) has been trained on. The output often depends on the precision and clarity of the instructions (prompts) you provide. A vague or poorly defined prompt might lead the AI to produce less relevant results. The more context and direction you give it, the better the result will be. What’s more, the capabilities of these AI systems are being enhanced through the introduction of versatile plug-ins that equip them to browse websites, analyze data files, or access other services. Think of this as giving your intern access to a group of experts to help accomplish your tasks.

One strategy in using a generative AI tool is first to tell it what kind of expert or persona you want it to “be.” Ask it to be an expert management consultant, a skilled teacher, a writing tutor, or a copy editor, and then give it a task.

Prompts can also be constructed to get these AI systems to perform complex and multi-step operations. For example, let’s say a teacher wants to create an adaptive tutoring program—for any subject, any grade, in any language—that customizes the examples for students based on their interests. She wants each lesson to culminate in a short-response or multiple-choice quiz. If the student answers the questions correctly, the AI tutor should move on to the next lesson. If the student responds incorrectly, the AI should explain the concept again, but using simpler language.

Previously, designing this kind of interactive system would have required a relatively sophisticated and expensive software program. With ChatGPT, however, just giving those instructions in a prompt delivers a serviceable tutoring system. It isn’t perfect, but remember that it was built virtually for free, with just a few lines of English language as a command. And nothing in the education market today has the capability to generate almost limitless examples to connect the lesson concept to students’ interests.

Chained prompts can also help focus AI systems. For example, an educator can prompt a generative AI system first to read a practice guide from the What Works Clearinghouse and summarize its recommendations. Then, in a follow-up prompt, the teacher can ask the AI to develop a set of classroom activities based on what it just read. By curating the source material and using the right prompts, the educator can anchor the generated responses in evidence and high-quality research.

However, much like fledgling interns learning the ropes in a new environment, AI does commit occasional errors. Such fallibility, while inevitable, underlines the critical importance of maintaining rigorous oversight of AI’s output. Monitoring not only acts as a crucial checkpoint for accuracy but also becomes a vital source of real-time feedback for the system. It’s through this iterative refinement process that an AI system, over time, can significantly minimize its error rate and increase its efficacy.

Uses of AI in Education

In May 2023, the U.S. Department of Education released a report titled Artificial Intelligence and the Future of Teaching and Learning: Insights and Recommendations. The department had conducted listening sessions in 2022 with more than 700 people, including educators and parents, to gauge their views on AI. The report noted that “constituents believe that action is required now in order to get ahead of the expected increase of AI in education technology—and they want to roll up their sleeves and start working together.” People expressed anxiety about “future potential risks” with AI but also felt that “AI may enable achieving educational priorities in better ways, at scale, and with lower costs.”

AI could serve—or is already serving—in several teaching-and-learning roles:

Instructional assistants. AI’s ability to conduct human-like conversations opens up possibilities for adaptive tutoring or instructional assistants that can help explain difficult concepts to students. AI-based feedback systems can offer constructive critiques on student writing, which can help students fine-tune their writing skills. Some research also suggests certain kinds of prompts can help children generate more fruitful questions about learning. AI models might also support customized learning for students with disabilities and provide translation for English language learners.

Teaching assistants. AI might tackle some of the administrative tasks that keep teachers from investing more time with their peers or students. Early uses include automated routine tasks such as drafting lesson plans, creating differentiated materials, designing worksheets, developing quizzes, and exploring ways of explaining complicated academic materials. AI can also provide educators with recommendations to meet student needs and help teachers reflect, plan, and improve their practice.

Parent assistants. Parents can use AI to generate letters requesting individualized education plan (IEP) services or to ask that a child be evaluated for gifted and talented programs. For parents choosing a school for their child, AI could serve as an administrative assistant, mapping out school options within driving distance of home, generating application timelines, compiling contact information, and the like. Generative AI can even create bedtime stories with evolving plots tailored to a child’s interests.

Administrator assistants. Using generative AI, school administrators can draft various communications, including materials for parents, newsletters, and other community-engagement documents. AI systems can also help with the difficult tasks of organizing class or bus schedules, and they can analyze complex data to identify patterns or needs. ChatGPT can perform sophisticated sentiment analysis that could be useful for measuring school-climate and other survey data.

Though the potential is great, most teachers have yet to use these tools. A Morning Consult and EdChoice poll found that while 60 percent say they’ve heard about ChatGPT, only 14 percent have used it in their free time, and just 13 percent have used it at school. It’s likely that most teachers and students will engage with generative AI not through the platforms themselves but rather through AI capabilities embedded in software. Instructional providers such as Khan Academy, Varsity Tutors, and DuoLingo are experimenting with GPT-4-powered tutors that are trained on datasets specific to these organizations to provide individualized learning support that has additional guardrails to help protect students and enhance the experience for teachers.

Google’s Project Tailwind is experimenting with an AI notebook that can analyze student notes and then develop study questions or provide tutoring support through a chat interface. These features could soon be available on Google Classroom, potentially reaching over half of all U.S. classrooms. Brisk Teaching is one of the first companies to build a portfolio of AI services designed specifically for teachers—differentiating content, drafting lesson plans, providing student feedback, and serving as an AI assistant to streamline workflow among different apps and tools.

Providers of curriculum and instruction materials might also include AI assistants for instant help and tutoring tailored to the companies’ products. One example is the edX Xpert, a ChatGPT-based learning assistant on the edX platform. It offers immediate, customized academic and customer support for online learners worldwide.

Regardless of the ways AI is used in classrooms, the fundamental task of policymakers and education leaders is to ensure that the technology is serving sound instructional practice. As Vicki Phillips, CEO of the National Center on Education and the Economy, wrote, “We should not only think about how technology can assist teachers and learners in improving what they’re doing now, but what it means for ensuring that new ways of teaching and learning flourish alongside the applications of AI.”

Challenges and Risks

Along with these potential benefits come some difficult challenges and risks the education community must navigate:

Student cheating. Students might use AI to solve homework problems or take quizzes. AI-generated essays threaten to undermine learning as well as the college-entrance process. Aside from the ethical issues involved in such cheating, students who use AI to do their work for them may not be learning the content and skills they need.

Bias in AI algorithms. AI systems learn from the data they are trained on. If this data contains biases, those biases can be learned and perpetuated by the AI system. For example, if the data include student-performance information that’s biased toward one ethnicity, gender, or socioeconomic segment, the AI system could learn to favor students from that group. Less cited but still important are potential biases around political ideology and possibly even pedagogical philosophy that may generate responses not aligned to a community’s values.

Privacy concerns. When students or educators interact with generative-AI tools, their conversations and personal information might be stored and analyzed, posing a risk to their privacy. With public AI systems, educators should refrain from inputting or exposing sensitive details about themselves, their colleagues, or their students, including but not limited to private communications, personally identifiable information, health records, academic performance, emotional well-being, and financial information.

Decreased social connection. There is a risk that more time spent using AI systems will come at the cost of less student interaction with both educators and classmates. Children may also begin turning to these conversational AI systems in place of their friends. As a result, AI could intensify and worsen the public health crisis of loneliness, isolation, and lack of connection identified by the U.S. Surgeon General.

Overreliance on technology. Both teachers and students face the risk of becoming overly reliant on AI-driven technology. For students, this could stifle learning, especially the development of critical thinking. This challenge extends to educators as well. While AI can expedite lesson-plan generation, speed does not equate to quality. Teachers may be tempted to accept the initial AI-generated content rather than devote time to reviewing and refining it for optimal educational value.

Equity issues. Not all students have equal access to computer devices and the Internet. That imbalance could accelerate a widening of the achievement gap between students from different socioeconomic backgrounds.

Many of these risks are not new or unique to AI. Schools banned calculators and cellphones when these devices were first introduced, largely over concerns related to cheating. Privacy concerns around educational technology have led lawmakers to introduce hundreds of bills in state legislatures, and there are growing tensions between new technologies and existing federal privacy laws. The concerns over bias are understandable, but similar scrutiny is also warranted for existing content and materials that rarely, if ever, undergo review for racial or political bias.

In light of these challenges, the Department of Education has stressed the importance of keeping “humans in the loop” when using AI, particularly when the output might be used to inform a decision. As the department encouraged in its 2023 report, teachers, learners, and others need to retain their agency. AI cannot “replace a teacher, a guardian, or an education leader as the custodian of their students’ learning,” the report stressed.

Policy Challenges with AI

Policymakers are grappling with several questions related to AI as they seek to strike a balance between supporting innovation and protecting the public interest (see sidebar). The speed of innovation in AI is outpacing many policymakers’ understanding, let alone their ability to develop a consensus on the best ways to minimize the potential harms from AI while maximizing the benefits. The Department of Education’s 2023 report describes the risks and opportunities posed by AI, but its recommendations amount to guidance at best. The White House released a Blueprint for an AI Bill of Rights, but it, too, is more an aspirational statement than a governing document. Congress is drafting legislation related to AI, which will help generate needed debate, but the path to the president’s desk for signature is murky at best.

It is up to policymakers to establish clearer rules of the road and create a framework that provides consumer protections, builds public trust in AI systems, and establishes the regulatory certainty companies need for their product road maps. Considering the potential for AI to affect our economy, national security, and broader society, there is no time to waste.

Why AI Is Different

It is wise to be skeptical of new technologies that claim to revolutionize learning. In the past, prognosticators have promised that television, the computer, and the Internet, in turn, would transform education. Unfortunately, the heralded revolutions fell short of expectations. 

There are some early signs, though, that this technological wave might be different in the benefits it brings to students, teachers, and parents. Previous technologies democratized access to content and resources, but AI is democratizing a kind of machine intelligence that can be used to perform a myriad of tasks. Moreover, these capabilities are open and affordable—nearly anyone with an Internet connection and a phone now has access to an intelligent assistant. 

Generative AI models keep getting more powerful and are improving rapidly. The capabilities of these systems months or years from now will far exceed their current capacity. Their capabilities are also expanding through integration with other expert systems. Take math, for example. GPT-3.5 had some difficulties with certain basic mathematical concepts, but GPT-4 made significant improvement. Now, the incorporation of the Wolfram plug-in has nearly erased the remaining limitations. 

It’s reasonable to anticipate that these systems will become more potent, more accessible, and more affordable in the years ahead. The question, then, is how to use these emerging capabilities responsibly to improve teaching and learning. 

The paradox of AI may lie in its potential to enhance the human, interpersonal element in education. Aaron Levie, CEO of Box, a Cloud-based content-management company, believes that AI will ultimately help us attend more quickly to those important tasks “that only a human can do.” Frederick Hess, director of education policy studies at the American Enterprise Institute, similarly asserts that “successful schools are inevitably the product of the relationships between adults and students. When technology ignores that, it’s bound to disappoint. But when it’s designed to offer more coaching, free up time for meaningful teacher-student interaction, or offer students more personalized feedback, technology can make a significant, positive difference.” 

Technology does not revolutionize education; humans do. It is humans who create the systems and institutions that educate children, and it is the leaders of those systems who decide which tools to use and how to use them. Until those institutions modernize to accommodate the new possibilities of these technologies, we should expect incremental improvements at best. As Joel Rose, CEO of New Classrooms Innovation Partners, noted, “The most urgent need is for new and existing organizations to redesign the student experience in ways that take full advantage of AI’s capabilities.”

While past technologies have not lived up to hyped expectations, AI is not merely a continuation of the past; it is a leap into a new era of machine intelligence that we are only beginning to grasp. While the immediate implementation of these systems is imperfect, the swift pace of improvement holds promising prospects. The responsibility rests with human intervention—with educators, policymakers, and parents to incorporate this technology thoughtfully in a manner that optimally benefits teachers and learners. Our collective ambition should not focus solely or primarily on averting potential risks but rather on articulating a vision of the role AI should play in teaching and learning—a game plan that leverages the best of these technologies while preserving the best of human relationships.

John Bailey is a strategic adviser to entrepreneurs, policymakers, investors, and philanthropists and is a nonresident senior fellow at the American Enterprise Institute.

This article appeared in the Fall 2023 issue of Education Next. Suggested citation format:

Bailey, J. (2023). AI in Education: The leap into a new era of machine intelligence carries risks and challenges, but also plenty of promise. Education Next, 23(4), 28-35.

For more, please see “The Top 20 Education Next Articles of 2023.”

Just as there’s no one-size-fits-all way to educate students, perhaps there’s no one-size-fits-all-way to get them to school either.

That’s the argument behind HopSkipDrive, a startup that is seeking to complement and redefine the traditional bus model of taking students to school. Instead of kids making their way to the pre-determined route of a 72-passenger school bus, HopSkipDrive creates customized pick-ups in cars and vans based on individual students’ needs. It’s a rideshare-like model with similarities to and major differences from Uber and Lyft.

The company provides an app that connects families and students with highly vetted drivers, all of whom are also experienced caregivers who either currently have a child at home or have cared for children in the past. They use their personal, pre-approved cars to drive students to and from school or after-school activities, while families and the contracting agency—such as a school, welfare office, or group home—can track the car’s movement in real time. Rides are scheduled well in advance but can be changed with a few hours’ notice.

When three parents launched the app in 2015, “it had nothing to do with school transportation,” said Chief Executive Officer Joanna McFarland, a cofounder and mother of two. “It really had to do with creating options for full-time working moms and families like mine, who were really struggling to just make the logistics of running a family work.”

But safely transporting kids turned out to be a widely shared challenge. Today, Los Angeles-based HopSkipDrive contracts with schools, districts, counties, and nonprofit partners in eight states and Washington, D.C., with a focus on meeting the transportation needs of students who fall outside traditional school-bus norms.

A Widespread Problem

The initial idea for HopSkipDrive was hatched at a child’s birthday party, and the original intended market was parents like McFarland and her cofounders, who wanted to get their kids to after-school activities safely without interrupting their own workdays. But company leaders soon realized there was a broader need for more flexible student transportation options, fueled by phenomena such as open enrollment within districts, students attending out-of-neighborhood “choice” schools, the less-predictable needs of students experiencing homelessness, and varied start times within districts. Within about two years, HopSkipDrive started to work with counties and school boards that were struggling to meet transportation needs because of a growing shortage of bus drivers.

After the abrupt closure and eventual reopening of schools due to Covid-19, those shortages grew more severe. Many districts initially furloughed drivers during school-building closures, and many of those drivers did not return to the job after schools reopened. A 2021 survey by the National Association for Pupil Transportation found that every region of the country was experiencing driver shortages, and at least two-thirds of districts had altered service as a result. Some 51 percent of respondents described shortages as “desperate.”

With fewer school-bus drivers, districts have consolidated routes and lack back-up options when a driver calls out sick. Some districts in Georgia cut bus service for students attending charter schools or schools of choice. Service gaps have caused many students to arrive late or miss school if their parents and neighbors are unable to carpool on the fly. Some state and district leaders have gotten creative in their responses to the problem: Massachusetts called on National Guard troops to drive students, and Washington, D.C., launched a free school-bus driver training program for city cab drivers.

HopSkipDrive, which initially laid off much of its staff and launched an adult ride option during the pandemic, proved another ready solution—and one that had been designed by parents with children and family needs in mind.

“Caregivers on Wheels”

HopSkipDrive rides are performed by what the company calls “CareDrivers”—adults who are vetted both for their experience with children and driving abilities. CareDrivers must have five years of caregiving experience as parents, guardians, nannies, teachers or paraprofessionals, or in other childcare roles. They submit to a 15-point certification process, which includes a fingerprint-based background check, motor vehicle history search, and ongoing monitoring for new driving infractions. The driver must use a vehicle that is less than 10 years old and seats four to seven passengers. New drivers go through a live orientation with a member of the HopSkipDrive team.

The company facilitated more than 400,000 rides for children and families in 2021, with a group of more than 4,000 active CareDrivers serving about 350 school districts. Over 2 million children have been driven safely over the company’s brief history. These rides totaled more than 7 million miles in 2021, and 99.7% concluded without a safety-related issue. In all, HopSkipDrive rides have traveled more than 20 million miles to date, with no critical safety incidents.

The company keeps close watch on drivers during rides. Using telematics—the long-distance transmission of digital data—HopSkipDrive tracks the whereabouts of drivers and children. It also checks for unsafe driving behaviors by monitoring phone usage, speeding, acceleration, hard braking, and hard turning. This technology provides visibility without installing in-vehicle cameras, which pose privacy concerns. In addition, the company also provides districts and parents with backup drivers in case of a last-minute cancellation, as well as real-time location updates.

Other innovators in this space use similar driver vetting and real-time location tracking via apps, but with different focuses. For example, Zum contracts with schools and districts to modernize routes and fleets, enhance efficiency, and provide carbon-neutral transportation in cities such as Los Angeles, San Francisco, and Seattle. TaxiMom offers a subscription-based transportation service for families in the Dallas-Fort Worth and Houston markets.

Part of the Transportation Puzzle

HopSkipDrive is not looking to replace the school bus. In a densely populated area, filling a 72-passenger bus is still the most efficient way to get students to school. Facilitated ridesharing is instead one part of a more comprehensive transportation solution for school districts, two-thirds of which own their fleet of buses.

But the company does look like a disruptive innovation relative to the school bus. It is bringing a more affordable and customizable technology-based solution that isn’t as good as the status quo, judging by historical measures of performance, to non-consumers for whom the alternative is nothing at all. In this case, HopSkipDrive is using technology to reach students who previously had no district-provided ride or who were over-served by a large school bus in a lightly populated area.

For example, one of HopSkipDrive’s larger partnerships is with the Aurora Public Schools in Colorado. The district uses HopSkipDrive to provide school transportation for about 70 students who are experiencing homelessness, are in foster care, or have special needs or disabilities that make riding a school bus challenging.

“These are kids who don’t fit neatly on a school bus route or kids who are highly mobile and moving frequently,” McFarland said. “If you are a child in the foster system who moves placements at 10:30 at night, you can’t reroute a school bus to get that child to school. And yet HopSkipDrive can get them to school the next day with no change in schooling. And we know that every time a child moves—and the average child in foster care might move three to five times a year—that might mean switching schools. Credits don’t transfer. It can take several weeks to change schools. What one district teaches in fourth grade, another district teaches in fifth grade… We have federal mandates to fix that, but transportation is one of the biggest barriers. And a service like HopSkipDrive really helps districts provide transportation for those hardest-to-serve students in a really flexible way.”

At Rocky Mountain Prep, a charter school in Colorado, the arrangement works differently. There, HopSkipDrive helps specific students who are struggling to get to school, buying time for the school and families to develop longer-term solutions to the attendance challenge. In one instance, Rocky Mountain Prep provided HopSkipDrive to a parent who had recently given birth to help get her older child to school for a few weeks.

HopSkipDrive may also lead to better optimization of school transportation in the longer run, said McFarland.

“Some of our more innovative districts are really starting to think in a much more holistic way, and we are helping them really think about: ‘What does your fleet size need to look like? How many buses do you need? How many drivers do you need?’” she said. “Because we’re operating in this world of shortages, but when you think about it as a utilization problem, maybe the number of bus drivers that we have is actually sufficient. Maybe we can … save money and use that money to increase bus driver salaries and retain drivers.”

Driving into the Future

The question for any transportation innovation now is whether it can scale. Regulation will play a role.

Many states have antiquated regulations around who can drive students to schools, according to McFarland. For example, Massachusetts requires a driver to have a certification that is close to a commercial driver’s license—a rule designed for safety back when a school bus was the only transportation a district would have provided for students to get to school. Similarly, many districts require their own unique background checks for potential drivers, a roadblock for vetted CareDrivers who want to drive for multiple districts.

States are updating their laws, however, to make room for new transportation solutions. A 2021 Maryland law allows districts to transport students in vehicles other than official school buses, and new regulations in Georgia permit the use of “alternative vehicles” for students who receive special-education services, are in foster care, or are experiencing homelessness.

The ultimate solution seems to be to update transportation regulations to focus on outcomes—namely safety and reliability—instead of inputs. That means setting a baseline for safety consistent with different vehicle types and using technology to enforce safety and create accountability.

Looking beyond once-in, once-out traditional school busing will be especially important in states where educational options are proliferating, through charter schools, education savings accounts and micro-grant programs, microschools, and other flexible and innovative schooling arrangements. Today, somewhere between 1.1 and 2.1 million students are enrolled in microschools or learning pods, for example.

But these options aren’t real choices if students don’t have a way to get there.

“We see over and over again, parents would choose a particular school for their kids, but they don’t enroll in that school because it’s across town or it requires two hours on a public bus or they just can’t get their kids there,” McFarland said.

Some states aren’t just modernizing their transportation regulatory framework, they are seeking to spur innovation. The Arizona Transportation Modernization Grants Program, for example, awarded $19 million in its first year to 24 schools and nonprofit education agencies with novel transportation solutions. That included the Black Mothers Forum, a nonprofit that operates microschools in Tempe and South Phoenix, which developed a community carpool app and provided transportation for all students. And it included Tolleson Union High School District, which contracted with HopSkipDrive to create a hub-and-spoke model to provide transportation to out-of-district students that want to attend one of its specialized programs.

The models are different, but the goals are the same: Reduce the time in transit for students and get them to their educational options on-time, safely, and ready to learn.

“You might have a bus that is full in the morning but in the afternoon is empty because kids are doing different activities,” McFarland said. “So, when you start to think about that, you can open up so many opportunities both inside and outside the classroom. Just thinking a little bit more holistically and a little bit more creatively.”

Michael Horn is an executive editor of Education Next, co-founder of and a distinguished fellow at the Clayton Christensen Institute for Disruptive Innovation, and author of From Reopen to Reinvent.

This article appeared in the Summer 2023 issue of Education Next. Suggested citation format:

Horn, M.B. (2023). Beyond the Big Yellow Bus: Can transportation apps reinvent how students get to school? Education Next, 23(3), 76-79.

For all the speculation about ChatGPT’s potential to upend K–12 writing instruction, there has been little investigation into the underlying assumption that the AI chatbot can produce writing that makes the grade.

We put OpenAI’s ChatGPT to the test by asking it to write essays in response to real school curriculum prompts. We then submitted those essays for evaluation. The results show that ChatGPT produces responses that meet or exceed standards across grade levels. This has big implications for schools, which should move with urgency to adjust their practices and learning models to keep pace with the shifting technological landscape.

Background

When it burst onto the scene in November 2022, ChatGPT’s clear and thorough written responses to user-generated prompts sparked widespread discussion. What it might mean for K–12 education was one area of speculation. Some worried about the potential for plagiarism, with students dishonestly passing off computer-generated work as their own creative product. Some viewed that threat as particularly formidable, pointing to three attributes that make ChatGPT different from past tools. First, it generates responses on-demand, meaning that students can receive a complete essay tailored to their prompt in a matter of seconds. Second, it is not repetitive. It tends to answer multiple submissions of the same prompt with responses that are distinct in their arguments and phrasing. And third, its output is untraceable, as it is not stored in any publicly accessible place on the Internet.

Education decision makers are already moving to respond to this new technology. In January, the New York City Department of Education instituted a ban on ChatGPT by blocking access to it on all its devices and networks. Los Angeles, Oakland, Seattle, and Baltimore school districts have imposed similar prohibitions. As leaders in other districts, schools, and classrooms grapple with if, when, and how to make changes in response to this technology, they need a read on how well ChatGPT, in its present form, can deliver on the threat it is purported to pose.

To help answer this question, we took three essay prompts per grade level from EngageNY’s curriculum for grades 4 through 12, which are the grades in which students produce long-form essays. For each grade level, the three essay prompts covered the three main types of writing —persuasive, expository, and narrative—that students do. The tasks ranged from creating a choose-your-own-adventure story about an animal and its defense mechanisms to selecting a central idea common to Robert Browning’s poem “My Last Duchess,” William Shakespeare’s Hamlet, and an excerpt from Virginia Woolf’s essay “A Room of One’s Own” and explaining how the texts work together to build an understanding of that idea. We then asked ChatGPT to produce an essay response in the voice of a student from the respective grade level. With the essays in hand, we commissioned a K–12 grading service to assess ChatGPT’s writing. The human graders evaluated each essay using rubrics from the Tennessee Department of Education that were tailored to the grade level and writing task. The graders assessed the essays across four categories of criteria—focus and organization, idea development, language, and conventions—and produced a numerical grade.

There is a caveat to our approach in answering this question. The way in which we asked ChatGPT to write its essays would likely differ from the way in which a discerning student using ChatGPT to plagiarize would use it. We provided little reworking of the essay prompts—just enough so that ChatGPT could understand them. Nor did we edit the chatbot’s output. In contrast, students would likely evaluate the AI’s writing and resubmit prompts that coached it toward a better product or edit its work to improve it where they saw fit. A student seeking to have ChatGPT write them an essay could likely take this cyborg approach to their assignment—presuming they did the work with more than just an hour of lead time—and earn higher grades than in our approach of using the chatbot on its own.

ChatGPT Clears the Bar

ChatGPT earned passing marks on each essay type at each grade level. It performed especially well in response to the writing tasks of the lower grades. Figure 1 depicts its average essay scores across the range of grades. Even at the upper grade levels, ChatGPT is a solid B or C student. Although its performance on high school prompts would not land it a spot at the valedictorian’s podium, it would still get it to the graduation stage, which has important implications for schools.

 

Figure 1

Figure 2 compares ChatGPT’s abilities across the four criteria categories: organization, idea development, language, and conventions.

 

Figure 2

ChatGPT’s strength in language and conventions show that it is a clear writer, capable of crafting fluent, grammatically sound prose. The chatbot either met or exceeded standards in both these categories for all 27 essays submitted.

The AI has the most room for improvement in its development of ideas. The graders’ written feedback reveals that it sometimes fails to support its claims with reasons or evidence and, in a few instances, makes assertions that are flat out false. It struggles the most to develop its ideas in response to literature. All five of the instances in which it earned a D+ —its lowest grade—were those in which the chatbot was asked to demonstrate its understanding of long-form prose, such as A Midsummer Night’s Dream or The Joy Luck Club. This calls into question if and how closely the chatbot has “read” these materials.

Now what?

These findings suggest that ChatGPT is already powerful enough that educators must change the status quo of writing instruction. Schools should evolve their practices, pedagogy, and policies to address the underlying forces that compel students to use technology like ChatGPT in counterproductive ways without resorting to blocks and bans that limit the use of this potentially powerful learning tool.

Use ChatGPT to Help Students Go Beyond its Capabilities

Some of the early actions taken in response to ChatGPT have focused on catching students misusing it or preventing them from accessing it altogether, such as the districts banning or blocking the AI. Elsewhere, teachers are now requiring students to do their writing with pen and paper in an effort to thwart copy-and-pasting from the chatbot. Developers are also creating new plagiarism detection software designed to identify AI-generated writing.

Each of these options comes with its own set of tradeoffs, but one drawback common to all of them is the cat-and-mouse dynamic they establish between schools and students. Each sends the message that students cannot be trusted with technology.

There’s another approach, however. By inviting ChatGPT into the classroom instead of locking it out, schools can push students toward independent thinking in a way that doesn’t signal mistrust. There are indications that, despite some of the high-profile bans, many teachers are thinking along these lines. According to a survey of more than 2,000 teachers commissioned by the Walton Family Foundation, 40 percent of teachers report already using ChatGPT at least once a week. Thirty-eight percent of teachers report allowing students to use ChatGPT, whereas only 10 percent say they have caught the students using it without their permission. And 72 percent say that “ChatGPT is just another example of why we can’t keep doing things the old way for schools in the modern world.”

One way to leverage ChatGPT is by intentionally using it in class. For example, teachers can have ChatGPT generate writing samples in response to different essay questions, which they can then use with their class to dissect the quality of the writing. Weighing the strengths and weaknesses of writing examples is a commonly used teaching tool. It can force students to think deeply about what constitutes good prose. Doing this with ChatGPT’s output has multiple potential benefits. Because ChatGPT can generate writing on-demand, teachers can produce samples tailored to the needs of their lesson without sinking valuable time into crafting the examples themselves.  In addition, by using ChatGPT in this way, teachers can show students where the chatbot falls short and how they are expected to go above and beyond its capabilities. Doing so disincentivizes inappropriate use of ChatGPT—not by threatening punishment or imposing bans but by communicating high expectations.

Teachers can also use the essays that ChatGPT produces to help check that their own prompts are asking enough of students. If not, they may want to consider revising their assignments. This approach encourages students to think outside the bot and, in doing so, helps them build writing skills that cannot already be automated away.

Teachers can also allow students to bring the AI in as a writing aid for certain assignments. Students could use it to conduct research, refine their prose, and test their ideas to see if they make sense to ChatGPT. Some have characterized ChatGPT in this role as a calculator for writing. Like the calculator, the technology’s efficiencies and enhancements could be leveraged to push student work to levels of complexity and quality that would be otherwise outside the realm of possibility.

Flip the Classroom to Support Students Throughout Their Writing Process

In a traditional classroom, students sit in whole-class instruction during school hours and do most of their writing assignments at home. Under this model, students who do not master the skills taught in class have little recourse when it comes time to apply them on a graded take-home assignment. That, along with time management, can lead some students to resort to academic dishonesty. By flipping the classroom—which often entails students learning the content at home online and then spending class time practicing the skills taught in the digital lessons—teachers can support students in turning that confusion into understanding.

In a flipped English Language Arts classroom, a teacher might send students home to watch a video on how to organize their ideas in writing. The following day’s class could start with an activity analyzing the flow of a sample essay. Then, in the next part of class, students take time to work on their own outlines for an upcoming paper as the teacher moves around the room to help address misconceptions and to provide support to those who need it. Under such a model, students receive more intentional writing instruction. If they feel lost, they can turn to a teacher for guidance instead of looking to a chatbot for the answer.

Realign Incentives Toward Learning

More broadly, in today’s zero-sum education system, some students will likely feel tempted to turn to ChatGPT as a way of getting a leg up on their competition—their classmates. This stems from the traditional time-based grading system, which relies on one-shot assessments to award students term grades that are used to rank and group them. These marks are unchanging and follow students around for years, no matter how much learning they demonstrate after the fact. By placing such steep and long-lasting consequences on grades and such little emphasis on actual understanding, schools are communicating clearly that they value scores earned over skills learned. Given the priorities of this system, we shouldn’t be shocked that some students are willing to sacrifice a learning opportunity for a chance at a better score.

Today’s seat-time based school system, in which students advance from concept to concept after an allotted amount of time, regardless of whether they demonstrate understanding of the topic, is responsible for this traditional, one-shot assessment model. Contrast this with a mastery-based model, in which students advance only when they show they have a concept down pat. This means students are allowed multiple attempts to demonstrate their mastery through assessment. In doing so, mastery-based learning reduces the do-or-die stakes that can drive students to dishonesty. This may help more students gain the confidence to put their own thoughts on the page, even if that means risking failure on that attempt.

Technological advancement will continue to grow AI’s effectiveness not only as a writer but also as a writing coach for students. Even between the time we wrote this article and its release,  OpenAI released GPT-4, which does better on standardized tests than the previous version. Khan Academy announced it will use GPT-4 to power “Khanmigo, an AI-powered assistant that functions as both a virtual tutor for students and a classroom assistant for teachers.” It is unclear how or in what instances the work of a student guided by an AI tool can be inputted in the gradebook to spit out a letter that conforms to old-fashioned principles of grading. But in a classroom where students are evaluated on their ultimate mastery of a concept, they are free to practice in the ways that best build that mastery. More conventional schools should also rapidly adjust their pedagogical practices, including grading systems, to make full use of this new technology.

Conclusion

With ChatGPT already capable of producing prose that earns passing marks across grade levels, schools must adjust to make sure that students will learn how to write effectively and think critically. Some of the earliest actions taken by districts have missed the mark by trying to stuff the ChatGPT genie back in the bottle. Technologists predict that numerous professions – including those requiring advanced skills – will leverage ChatGPT in their day-to-day work in the near future. Lawyers will look to the chatbot instead of their more junior colleagues to create summaries of case notes and relevant laws. Journalists will use it to generate checklists of points to cover for articles on given topics. In these professional use cases, it will be critical for workers to accurately evaluate ChatGPT’s output and put it to effective and ethical use. The best place and time for a worker of tomorrow to learn how to do that is in a classroom today.

As a result, blocking ChatGPT is not only futile, but also counterproductive for students who will be forced to use this tool and others like it in a working world where they are ubiquitous. By rethinking classroom practices and restructuring learning models, schools can give students the tools, guidance, and incentives to grow their writing skills in the age of artificial intelligence.

Michael Horn is an executive editor of Education Next, co-founder of and a distinguished fellow at the Clayton Christensen Institute for Disruptive Innovation, and author of From Reopen to Reinvent. Daniel Curtis is a Master in Public Policy candidate at the Harvard Kennedy School of Government.

For more, please see “The Top 20 Education Next Articles of 2023.”

One of the hottest developments in education technology is schools banning technology.

After successive years of remote or hybrid learning, you might imagine tech-weary educators would be going after laptops and Zoom. But they are focused on cellphones, driven by three major concerns: students’ mental health, ability to stay engaged and learn during class, and struggles to focus for long stretches of time without task switching.

There’s an irony here. These bans are proliferating even as there are more useful, engaging, and instructionally sound mobile-learning applications than ever before. That suggests that cellphone bans, while useful in many school settings, shouldn’t be universal. We risk barring teachers, schools, and districts from productively using these apps to drive learning gains.

Where the Phones Aren’t

Some bans are blanket ones at the country or state level. In 2018, France passed a law that prohibited students under 15 from using phones, tablets, and smart watches in schools. The Australian state of Victoria bans phones in primary and secondary schools.

Some schools in the United States have taken similarly dramatic actions. Public schools and districts in Missouri, Pennsylvania, Maine, and New York State have instituted bans, often citing the devices’ ability to distract students from learning. And the Buxton School, a boarding school in Western Massachusetts, instituted a total ban on smartphones on campus after one of its students live streamed two others engaged in a fight. Students now are allowed “dumb” phones, but the constant alerts and capabilities of the smartphones are gone.

Other educators have counseled more moderate approaches to the same effect. Doug Lemov, author of Teach Like a Champion, wrote recently in Education Next that restricting cellphone use doesn’t “mean banning phones, it just means setting rules. These can take different forms, like setting up cellphone lockers at the main entrance, requiring students to use cellphone-collection baskets at the classroom door, or limiting use to cellphone-approved zones in the school building” (see “Take Away Their Cellphones” feature, Summer 2022).

One common method requires that students check their phones when they enter the school building. At several middle and high schools in and around Springfield, Massachusetts, phones are stored in a magnetic pouch that only educators can open until the end of the day. These metal pouches—like the one developed by Yondr, a San Francisco-based company founded in 2014—are commonly used at concerts and comedy shows to eliminate the distraction of mobile phones and allow people to engage fully in the experience before them. That same sales pitch has made pouches popular at many schools.

Although publications like the Boston Globe have editorialized in favor of these bans, not every school system is on board. Tragedies like mass school shootings in Uvalde, Texas, and Parkland, Florida, have given many parents pause about banning phones. The New York City Department of Education, for example, ended a ban on cellphones in schools in 2015, citing parents’ wishes to reach their children during the school day.

According to the federal education department, more than three quarters of public schools prohibited the non-academic use of cellphones during school hours in 2019–20. The phrasing suggests that in that number are schools that are outright banning phones, as well as those who have restricted phones but are consciously leveraging them for academic reasons.

Worries about Mental Health and Focus

Momentum to moderate cellphone usage stems from concerns about students’ mental health. American teenagers are experiencing a significant mental health crisis. According to the Centers for Disease Control, in 2021 44 percent of U.S. high-school students reported “persistent feelings of sadness or hopelessness” in the past year—up from 26 percent in 2009. Visits to emergency rooms for mental health emergencies and attempted suicides are up as well.

What is causing the crisis is disputed. Many pediatric groups and researchers—most prominently psychologist Jean Twenge, the author of the book iGen—have cited correlations in the rise of social media and smartphone use with teenagers’ increasing depression and anxiety to suggest that excessive smartphone and social media use is damaging a generation. But the reality appears more nuanced. One set of studies published in Child Development, for example, used a randomized design to tease apart the emotional impact of receiving fewer likes on social media. It found that although all students suffered a relatively immediate disappointment, it was only teenagers who were more vulnerable to social rejection that suffered a more enduring and significant negative impact.

Still, as Atlantic writer Derek Thompson observed, Twenge’s point may be misunderstood. “Social media isn’t like rat poison, which is toxic to almost everyone,” he wrote. “It’s more like alcohol: a mildly addictive substance that can enhance social situations but can also lead to dependency and depression among a minority of users.”

The other challenge with social media that Twenge cites isn’t the social media itself, but that it replaces sleep and in-person social interactions to such a high degree. Although some have argued there are silver linings to this—such as declines in binge drinking and sexual activity among teens—the impact on adolescents’ loneliness may be contributing to their decline in mental health. And if isolation is the true driver, of course pandemic-related lockdowns and school closures likely contributed to and accelerated some of these trends.

During class, student cellphones present two pressing challenges for teachers: disruption when students use their phones for non-academic purposes during class, and teenagers’ struggles to maintain the deep focus that rigorous academics demand. Sustained attention is unlike many students’ more typical mode of frequent task switching, where they toggle between different apps, which frequent smartphone alerts encourage.

As Lemov wrote, “This is no small thing. … The more rigorous the task, the more it requires what experts call selective or directed attention. To learn well, you must be able to maintain self-discipline about where you direct your attention.”

A lack of practice in focusing could damage students’ abilities to learn and do difficult work, in other words. And some studies have suggested that cellphone bans lead to better learning. One study of high schools in the United Kingdom, for example, showed that schools that banned mobile phones had improved test scores on a year-end test.

It’s All About the Learning Model

Yet while these concerns have led to more cellphone bans, there also has been an explosion in useful learning applications for mobile devices. Think of Duolingo for learning language, or ABC Mouse for learning elementary school subjects, or Quizlet for checking understanding. The ability to learn nearly anything from a phone is better than it’s ever been for all ages of learners.

With the active learning methodologies at the heart of these apps, the learning opportunities on mobile devices are in many ways superior to many of the more passive, video- and text-based ones built for laptops and personal computers. Cellphones may distract from traditional lectures or whole-class instruction. But they also command and can hold individual students’ attention—a precious resource that fuels learning, even if that learning doesn’t look like what we’ve seen before. Phones also may get in the way of students mastering required academic standards, while also connecting students to the information about which they are most curious.

How to explain the paradoxes?

In many learning models, there simply isn’t a productive place for smartphones. But is that the fault of the phone or the model?

Take a case-study classroom, for example. In it, all students are expected to participate in a group discussion to work through a specific situation with a joint set of case facts. If students are instead paying attention to their own devices, the conversation suffers and student learning slows as well.

Contrast that with a foreign-language class where all students work on personalized language modules on Duolingo, for example. They then put their phones away to participate in small-group conversations. (Even before smartphones, a version of this called “language lab” put individual students at headphone stations to work independently with the education technology of the day before rejoining group conversations.) The phone is central to the design of the learning experience. Of course, there’s a risk that students will work on tasks outside of the one assigned. But schools and teachers can use technology to block access to other apps or build on the social dynamics of the classroom to incentivize students to stay on task.

This phenomenon has been true with Internet-connected laptops as well. A 2016 study about a set of West Point classrooms showed that allowing computers when there wasn’t a key purpose for them diminished learning (see “Should Professors Ban Laptops?” research, Fall 2016). On the other hand, a blended-learning model like New Classrooms’ Teach to One relies on laptops to personalize math instruction for middle school students. Research has found students make outsized gains on math tests after successive years of participating in Teach to One classrooms.

One last argument for maintaining cellphones is that schools must teach students to use them responsibly. But many educators’ retort is that they are simply helping show students that there is a time and place for such devices—and school isn’t it.

In that respect, cellphone bans are following the larger trend of banning many things in schools—from books to speakers to certain kinds of speech or topics of debate. Cellphones may make for another easy bogeyman, but blanket bans are ill-informed and regressive. Though we might not see a big reversal in phone bans anytime soon, we should. Educators on the ground should choose for themselves when and whether to allow their students to carry cellphones to class, so they can leverage learning apps to help students make progress.

Michael Horn is an executive editor of Education Next, co-founder of and a distinguished fellow at the Clayton Christensen Institute for Disruptive Innovation, and author of From Reopen to Reinvent.

This article appeared in the Winter 2023 issue of Education Next. Suggested citation format:

Horn, M.B. (2023). Ban the Cellphone Ban: Blanket policies ignore the potential of app-powered learning. Education Next, 23(1), 76-77.

After successive school years disrupted by shutdowns, isolation, and mass experiments in remote teaching, educators returned to school in Fall 2021 to find that our classrooms and students had changed.

In the first days of the return, perhaps, we didn’t see the full scope of the changes. Yes, most of us knew that there would be yawning academic gaps. Most of us understood then what the data have since clearly borne out: despite often heroic efforts by teachers to deliver remote instruction, the pandemic had caused a massive setback in learning and academic progress. The costs had been levied most heavily on those who could least afford it, and it would take months, if not years, to make up the lost time.

But at least we were all together again, even if we were all wearing masks. We were on the road back to regular life.

As the days passed, though, a troubling reality emerged.

The students who came back to us had spent long periods away from peers, activities, and social interactions. For many young people—and their teachers—the weeks and months of isolation had been difficult emotionally and psychologically. Some had lost loved ones. Many more had endured months in a house or apartment with nearly everything they valued—sports or drama or music, not to mention moments of sitting informally among friends and laughing—having suddenly evaporated from their lives. Even students who had escaped the worst of the pandemic were out of practice when it came to the expectations, courtesies, and give-and-take of everyday life. Perhaps as a result of this, their social skills had declined.

Our students looked the same—or at least we presumed they did behind the masks—but some seemed troubled and distant. Some struggled to concentrate and follow directions. They were easily frustrated and quick to give up. Many students simply didn’t know how to get along. The media was suddenly full of stories of discipline problems, chronic disruptions due to student distractibility, lack of interest, and misbehavior in the classroom, and historic levels of student absences. In schools where no one had ever had to think about how to deal with a fight, they burst into the open like brush fires after a drought. It didn’t help that many schools were short-staffed, with leaders struggling just to get classes covered and buses on the road.

The first post-pandemic year may well have been harder than the radically disrupted 18 months of rolling lockdowns and remote learning that preceded it. The jarring disruptions related to Covid-19 aren’t the whole story, however. What has happened to our students isn’t just the impact of a protracted, once-in-a-generation adverse event, but the combined effects of several large-scale, ground-shifting trends that predate the pandemic and have reshaped the fabric of young people’s lives. As we look forward, their combined effects should cause us to think beyond short-term recovery and to reconsider how we design schools and schooling.

An Internet Epidemic

The pandemic occurred amid a broader epidemic. Long before Covid-19, the psychologist Jean Twenge had found spiraling levels of depression, anxiety, and isolation among teens. “I had been studying mental health and social behavior for decades and I had never seen anything like it,” Twenge wrote in her 2017 book iGen.

This historic downturn in the well-being of young people coincided almost exactly with the dramatic rise of the smartphone and social media. More specifically, it coincided with the moment when they both became universal and being disconnected or an infrequent user was no longer viable.

As a parent, I experienced this firsthand. Even before the pandemic, I was desperately trying to manage my own children’s device usage, wary of how the time they spent on their phones was increasing while the time they spent reading and doing, well, almost everything else was decreasing. We wanted to limit social media as much as possible. But when friends plan where to meet up via Instagram messenger or some other platform, and when the key information for every soccer game—where, when, which uniform—is communicated via group chat, there is no choice but to join.

Research by Twenge and others found that teenagers’ media use roughly doubled between 2006 and 2016 across gender, race, and class. In competition against the smartphone, the book, the idea of reading, lost significant ground. By 2016, just 16 percent of 12th-grade students read a book or magazine daily. As recently as 1995, 41 percent did. Meanwhile, social media was on the rise. By 2016, about three-quarters of teenagers reported using social media almost every day (see Figure 1).

Those trends have only accelerated. A 2019 study by Common Sense Media reported that 84 percent of American teenagers own a smartphone. Parents are raising a generation that is both more connected and more disconnected than any before.

“The smartphone brought about a planetary rewiring of human interaction. As smartphones became common, they transformed peer relationships, family relationships and the texture of daily life for everyone—even those who don’t own a phone or don’t have an Instagram account,” Twenge and co-author Jonathan Haidt wrote in the New York Times in 2021. “It’s harder to strike up a casual conversation in the cafeteria or after class when everyone is staring down at a phone. It’s harder to have a deep conversation when each party is interrupted randomly by buzzing, vibrating notifications.” They quote the psychologist Sherry Turkle who notes that we are now “forever elsewhere.”

The average 12th grader in 2016, Twenge pointed out in iGen, went out with friends less often than the average 8th grader 10 years before. American teenagers were also less likely to date, drive a car, or have a job. “The roller rink, the basketball court, the town pool, the local necking spot—they’ve all been replaced by virtual spaces accessed through apps and the web,” Twenge wrote in The Atlantic. These virtual meetups are universally associated with less happiness for young people. “Those who spend an above-average amount of time with their friends in person are 20 percent less likely to say they’re unhappy than those who hang out for a below-average amount of time,” she wrote.

And that was long before Tik Tok and the latest round of social media platforms carefully designed to ensure obsession and the lingering anxiety that you really ought to be checking your phone; before the optimization of apps like Snapchat, with posts designed to disappear as soon as they are seen and therefore undiscoverable to an adult coming to a young person’s room to see what is amiss.

Pandemic Effects

Then in March 2020, virtually everything that might have competed with smartphones suddenly disappeared. A recent Common Sense Media study found that children’s daily entertainment usage of screens grew by 17 percent between 2019 and 2021—more than it had grown during the four years prior (see Figure 2). Overall, daily entertainment screen use in 2021 increased to 5.5 hours among tweens ages 8 to 12 and to more than 8.5 hours among teens ages 13 to 18, on average. These trends were even more pronounced for students from low-income families, whose parents were most likely to have to work in person and have fewer resources to spend on alternatives to screens.

At the levels of use that are now common, smartphones are catastrophic to the well-being of young people. As Twenge wrote, “The more time teenagers spend looking at screens, the more likely they are to report symptoms of depression. . . It’s not an exaggeration to describe [this generation] as being on the brink of the worst mental-health crisis in decades.”

Indeed, the data also show spikes in teenagers’ mental-health problems during the pandemic, when just 47 percent of students reported feeling connected to the adults and peers in their schools. Some 44 percent of high-school students reported feeling sad or persistently hopeless in 2021, according to the Centers for Disease Control. School factors had a significant effect on this data. Students who said they felt “connected to adults and peers” at school were almost 60 percent less likely to report persistent feelings of sadness or hopelessness than those who did not—some 35 percent of connected students felt that way, compared with 55 percent who did not feel connected to school. The socioemotional distress students are experiencing is as much a product of the cellphone epidemic as it is a product of the Covid-19 pandemic.

In addition, all of that time on screens—even without social media—degrades attention and concentration skills, making it harder to focus fully on any task and to maintain that focus. This is not a small thing. Attention is central to every learning task and the quality of attention paid by learners shapes the outcome of learning endeavors. The more rigorous the task, the more it requires what experts call selective or directed attention. To learn well, you must be able to maintain self-discipline about where you direct your attention.

“Directed attention is the ability to inhibit distractions and sustain attention and to shift attention appropriately,” Michael Manos, clinical director of the Center for Attention and Learning at Cleveland Clinic, recently told the Wall Street Journal. “If kids’ brains become accustomed to constant changes, the brain finds it difficult to adapt to a nondigital activity where things don’t move quite as fast.”

The Trouble with Task Switching

The problem with cellphones is that young people using them switch tasks every few seconds. Better put, young people practice switching tasks every few seconds, so they become more accustomed to states of half-attention, where they are ever more expectant of a new stimulus every few seconds. When students encounter a sentence or an idea that requires slow, focused analysis, their minds are already glancing around for something new and more entertaining.

Though all of us are at risk of this type of restlessness, young people are especially susceptible. The region of the brain that exerts impulse control and self-discipline, the prefrontal cortex, isn’t fully developed until age 25. Any time young people are on a screen, they are in an environment where they are habituated to states of low attention and constant task switching. In 2017, a study found that undergraduates, who are more cerebrally mature than K–12 students and therefore have stronger impulse control, “switched to a new task on average every 19 seconds when they were online.”

In addition, the brain rewires itself constantly based on to how it functions. This idea is known as neuroplasticity. The more time young people spend in constant half-attentive task switching, the harder it becomes for them to maintain the capacity for sustained periods of intense concentration. A brain habituated to being bombarded by constant stimuli rewires accordingly, losing impulse control. The mere presence of our phones socializes us to fracture our own attention. After a time, the distractedness is within us.

“If you want kids to pay attention, they need to practice paying attention,” is how Dr. John S. Hutton, a pediatrician and director of the Reading and Literacy Discovery Center at Cincinnati Children’s Hospital Medical Center, put it in a recent Wall Street Journal article.

There is, in other words, a clear post-pandemic imperative for schools. The first step in responding to the dual crisis of learning and well-being is to set and enforce cell-phone restrictions. An institution with the dual purpose of fostering students’ learning and well-being cannot ignore an intruder that actively erodes a young mind’s ability to focus and sustain attention and also magnifies anxiety, loneliness, and depression. Cellphones must be turned off and put away when students walk through school doors. Period.

But cellphone restrictions are only part of the equation. Schools themselves will also require rewiring.

How do we do that? The answer isn’t simple. My colleagues at Uncommon Schools Denarius Frazier, Hilary Lewis, and Darryl Williams, and I have written a book describing actions we think schools should consider. Here’s a road map of some of the things we think will be necessary.

Rewiring Classrooms for Connectedness

Extracurricular activities and social and emotional learning programs can be significant factors in shaping students’ experiences. But we should also recognize that the classroom is the single most important space when it comes to shaping students’ sense of connectedness to school. Out of a typical school day, at least five or six hours will be spent in classrooms—the overwhelming majority of students’ time. If classroom practices do little to instill a sense of belonging, students will feel a weak connection to the primary purpose of school.

But just as important, building classrooms to maximize belonging cannot come at the expense of academic achievement. We are in the midst of a learning crisis of historic proportions too. Students’ lack of progress in science, math, and reading, their reduced knowledge of history, their lessened exposure to the arts—these will have lifelong costs. Teaching needs to be better, not diluted. Classrooms need to maximize belonging and learning. It can’t be one or the other.

Happily, we think this is eminently possible. I’m thinking of a math class taught by my co-author Denarius Frazier, the principal of Uncommon Collegiate Charter High School in Brooklyn, N.Y. During a discussion about trigonometry, two dozen students engaged vigorously and energetically with one another. That is, until the beautiful moment when a student named Vanessa, who had been speaking authoritatively about her solution to the problem, suddenly realized that she had confused reciprocal and inverse functions—and that her solution is dead wrong.

Vanessa paused and glanced at her notes. “Um, I’d like to change my answer,” she said playfully, without a trace of self-consciousness. Then she laughed, and her classmates laughed with her. The moment was beautiful because it was lit by the warm glow of belonging. And that was not accidental.

Consider the image below: Vanessa is speaking as her classmates listen and offer affirming gazes. Their eyes are turned to Vanessa to show encouragement and support. Their expressions communicate psychological safety, reassurance, and belonging. In fact, it’s hard to put into words just how much their glances are communicating—and each one is a little different—but these wordless expressions are as critical to shaping the moment as Vanessa’s own character and humility. This interaction fosters and protects a space in which her bravery, humor, and openness can emerge. A space where she feels important.

How someone acts in a group setting is shaped as much by the audience and the social norms that the speaker perceives as it is by internal factors. And here those perceptions are not accidental. Frazier has socialized his students to “track”—or actively look at—the speaker and to endeavor to keep their body language and nonverbal cues positive. In Teach Like a Champion 3.0, I call that technique Habits of Attention. It is a small but critical aspect of how classrooms can maximize belonging and achievement.

Students also validated each other in other ways throughout the lesson. When a young woman named Folusho joined the discussion, she started by saying, “Ok, I agree with Vanessa…” So often, after a student speaks in class, no one other than the teacher responds or communicates that the statement was important. But when a peer’s comment begins, “I agree with…” it says implicitly that what my classmate just said is important. Such validation makes it more likely that students feel supported and successful, and that the speaker will contribute to the discussion again.

Again, this is not a coincidence. Frazier has taught his students to use phrases like that and weave their comments together, so their ideas are connected and those who have contributed feel the importance of their contributions. That technique is called Habits of Discussion. Along with Habits of Attention, it helps connect and validate students as they learn.

In addition, as Folusho was talking, her classmates began snapping their fingers. In Frazier’s classroom, that means “I agree” or “I support you.” It was a powerful dose of positive feedback at the precise moment when she, like almost anyone speaking aloud to a group of people, was most likely to momentarily wonder, “Am I making sense at all? Do I sound stupid?” Folusho suddenly got a supportive response—the snapping told her, “You’re doing great! You’re family. Let’s go!”

Again, that was deliberately woven into the fabric of the classroom. The technique, called Props, establishes procedures for students to recognize when their classmates are doing well and send affirming signals without disrupting class.

All three techniques show how a teacher like Frazier can intentionally establish a culture that reinforces both academic endeavor and a much stronger sense of belonging. And though it looks organic, there’s nothing natural about it. It’s a deliberate rewiring of social norms to maximize positive outcomes. Some skeptics label these sorts of techniques coercive or controlling, but it’s hard to watch Frazier and his students and hold on to those suspicions. Engineering the classroom to ensure positive peer-to-peer norms is about honoring young people and creating an environment that not only maximizes their learning but also their belonging—the pervading senses that school is for me and I am successful here. It’s a rewiring of the classroom that requires hard work and doggedness on the part of the teacher. But it is nothing less than students deserve.

Rewiring Schools for Belonging

Rewiring a school for belonging involves rethinking many of the things we do, such as extracurricular activities. Nashville Classical Charter School provides an example of how schools might do this. In 2021, school leaders were reconsidering how its programs could intentionally build a sense of connectedness and belonging among students. Head of School Charlie Friedman and his colleagues decided to dramatically expand after-school sports programs, to allow students to explore their identities, build relationships with trusted adults, and perform in front of a crowd.

Nashville Classical extended tryout periods, to maximize students’ opportunities feel like part of a team. Leaders also offered stipends for coaching and encouraged their best community builders to coach, by explicitly valuing expertise at culture building alongside expertise at the sport. The school engaged audiences by inviting families to vote on a mascot and created an engaging game-day experience with a cheerleading squad, songs, and chants. This attracted a substantial audience, so student-athletes could compete in front of more people and fans could build community through gathering and cheering together.

It’s important to have high-quality extracurriculars that aren’t based on years of prior experience. It’s hard for a student to decide in grade 8 that they would really like to be a part of the basketball team if they haven’t already spent years playing. But that’s not true of the debate team or the Spanish club. Those activities should be as well run as any others, rather than a lonely space with obligatory supervision where the connections are peripheral at best.

Schoolwide rituals are also important to fostering a sense of belonging. For example, Frazier’s school has a regular meeting circle where the entire school is present. Students are publicly honored for their academic excellence or for being positive members of the school community.

Character education and social and emotional learning programs can also play a role. But my advice is to build a few priorities into the fabric of the school rather than buy a program to use in an isolated manner. Positive character traits should be “caught, sought, and taught,” according to my co-author Hilary Lewis. Gratitude is a great example. When students make a habit of concretely expressing gratitude to other people in the school community, it confers mutual benefits. Expressions of gratitude make the recipients feel more connected while also confering status on the giver, because their appreciation is a thing worthy of sharing deliberately.

And, as Shawn Achor explains in his book The Happiness Advantage, expressing gratitude regularly has the effect of calling students’ attention to its presence. Repetitive thinking causes a “cognitive afterimage” where we continue to see whatever it is we’re thinking about, even when we’ve shifted focus. In other words, if you continually share and expect to be sharing examples of things you are grateful for, you start looking for them. You begin scanning the world for examples of good things to appreciate and notice more of the good that surrounds you. Gratitude is a well-being builder.

Open-ended opportunities to relax and connect outside of the classroom also foster connectedness and belonging. At Cardiff High School in Wales, for example, school leaders filled a common area with games that are easy to join. They added chess boards, card tables with decks of cards, and even a ping pong table to create opportunities for engaging, positive social interaction in between classes.

Saying No to Cellphones

These innovations can be powerful—but not on their own. The pull of smartphones and social media apps designed to distract is bound to undermine any expression of support, after-school sport, or card table. The single most important thing schools can do is to restrict cellphone access for large parts of the school day. This doesn’t mean banning phones, it just means setting rules. These can take different forms, like setting up cellphone lockers at the main entrance, requiring students to use cellphone-collection baskets at the classroom door, or limiting use to cellphone-approved zones in the school building. My personal preference is a simple policy: You can have your cellphone in your bag, but it must be turned off and cannot be visible during the school day. Not during lunch, not in the hall, not anywhere until after the last bell rings. If there’s an emergency and you need to contact your parents, you may use it in the main office. That’s it.

Schools must create blocks of time when students can work in a manner that allows them to rebuild their attentional skills and experience the full value of connected social interaction. They must also protect students’ opportunities to socialize with one another. Allowing students to use their phones as classroom tools (for quick research or as calculators, for example), or to leave them turned on (but with silent haptic notifications that distract nonetheless), or to turn them on during lunch or other learning breaks keeps them connected to their devices and disconnected from one another.

It won’t be easy, but it can be done. France has done it. The state of Victoria in Australia has done it. Some American public schools and districts have done it, in Missouri, Pennsylvania, Maine, and New York.

These bans are often followed by remarkable and instantaneous change. “It has transformed the school. Social time is spent talking to friends,” a teacher from Australia told my colleagues and me. “It is so nice walking around the yard seeing students actually interacting again, and no distractions during class,” said another.

The change, teachers told us, was quick—so long as you could get the adults to follow through. That is, if the rule was consistent and enforced, then students adapted quickly and were happy, even if they fought it at first. If the ban didn’t work, the problem was usually that some of the adults didn’t follow through. “Consistent enforcement from all = key,” one teacher explained in a note. “Can’t be ‘the cool teacher’.” The problem, of course, is that there’s a strong incentive to be “the cool teacher,” so schools must spend time making sure teachers understand the reasons for the rule and holding them accountable for supporting it.

School and district leaders should be prepared for doubts, skepticism, and pushback. We’ve seen this at the state level already. In 2019, lawmakers in four states proposed legislation to ban cellphones in school. But the bills, in Arizona, Maine, Maryland, and Utah, failed to advance. A rule that barred students from bringing cellphones into New York City public schools was ended in 2015, because then-Mayor Bill DeBlasio said “parents should be able to call or text their kids,” though individual schools may choose to limit phone access.

Two comments I often hear: “it’s an infringement on young people’s freedom” and “the role of schools is to teach young people to make better choices. We should talk to them about cell phones, not restrict them.”

The first response makes two assumptions: first, that there is no difference between young people and adults, and second, that there is no difference between the people who run a school—and therefore are responsible to stakeholders for outcomes—and the young people who attend the school. Both are mistaken. The purpose of a school is to give young people the knowledge and skills they require to lead successful lives. This always involves an exercise of social contract. We collectively give up something small as individuals and receive something valuable and rare in return as a group. It is impossible to run a school without this sort of give-and-take. Suggesting that we give students “freedom” to use cell phones whenever they want is trading valuable and enduring freedom that accrues later for a self-destructive indulgence in the present.

The argument that “schools should teach young people the skill of managing technology” is patently unrealistic. Schools are not designed to address, much less unravel, psychological dependence on portable supercomputers designed to disrupt and hold our attention. Teachers already have a daunting list of educational priorities. They are not trained counselors, and the school counselors on staff are in woefully short supply.

It’s magical thinking to propose that an epidemic that has doubled rates of mental health issues and changed every aspect of social interaction among millions of people is going to go away when a teacher says, “Guys, always use good judgment with your phones.” We’re not really wrestling with the problem if our response assumes that the average teacher, via a few pithy lessons, can battle a device that has addicted a generation into submission.

Restriction is a far better strategy. These efforts won’t be simple to execute, but the alternative is simply too damaging to students’ learning and well-being. Keep cellphones turned off and out of sight during the school day—and give students and educators a fighting chance to focus, reconnect, and build school cultures that nurture belonging and academic success.

Doug Lemov is founder of Teach Like a Champion and author of the Teach Like a Champion books. He is a co-author of the forthcoming book Reconnect, from which this essay is adapted. He was a managing director of Uncommon Schools, designing and implementing teacher training based on the study of high-performing teachers.

This article appeared in the Fall 2022 issue of Education Next. Suggested citation format:

Lemov, D. (2022). Take Away Their Cellphones … So we can rewire schools for belonging and achievement. Education Next, 22(4), 8-16.

For more, please see “The Top 20 Education Next Articles of 2023.”