Frederick Florin / Getty

Update: This article has been updated to remove identifying information about a teacher and to clarify the reporter’s classroom observations.

Four years ago, Paul France left a teaching job in the Chicago suburbs to move to San Francisco and be part of the so-called personalized-learning revolution in education. He joined a high-profile start-up called AltSchool whose investors include Mark Zuckerberg and the venture-capital firm Andreessen Horowitz. France was passionate about both education and technology and welcomed the opportunity to combine the two. But they would not prove to be as complementary as he thought.

AltSchool’s founder, a former Google executive named Max Ventilla, envisioned it as a place where students, aided by technology and tech-savvy teachers, would learn at their own pace, based on their interests and aptitudes. Teachers, using proprietary software to track student performance and help guide their learning, would be freed from time-consuming tasks like scoring exams and could devote more attention to students. The company’s mission was to help move the American education system from an industrial-age model “where schools were set up to resemble factories and students had a conveyor belt-like experience” to a more “learner-centric” approach that gives students a sense of agency and 21st-century problem-solving skills, Devin Vodicka, who is responsible for guiding the company’s personalized learning platform as it expands to more schools, told me.

The company opened small lab schools in San Francisco, Palo Alto, and New York. Students were provided with their own iPads or Chromebooks along with individualized “playlists” of learning activities to choose from. Cameras in classrooms videotaped students’ and teachers’ interactions, enabling teachers and engineers to analyze the footage and review what worked, a process AltSchool technologists hoped would eventually be aided by computers using machine-learning algorithms.

AltSchool was flooded with applicants willing and able to pay tuition of $30,000 or more for their children to attend, and became one of the hottest start-ups in educational technology; to date, it has raised more than $170 million in investment. It was part of a broad investor rush to ed tech. Last year, venture-capital investors put $2.7 billion into ed-tech companies, up from $1.6 billion in 2016, according to CB Insights, a software company that examines technology trends. The Chan Zuckerberg Initiative, the Netflix founder Reed Hastings, and the Gates Foundation have given millions of dollars to schools implementing technology-based personalized learning—many of them urban charter schools serving low-income children.

But in the midst of all the excitement, there’s little strong evidence that classroom technology, including personalized learning, is improving educational outcomes. A 2015 report from the Organization of Economic Cooperation and Development found that countries that invested heavily in computer technology for schools showed “no appreciable improvements” in reading, math or science, and that technology “is of little help in bridging the skills divide between advantaged and disadvantaged students.”

A recent Rand study evaluating the use of personalized learning found emerging signs of promise, but with plenty of caveats. Students in 40 schools with personalized-learning programs funded by the Gates Foundation scored a bit higher on standardized English and math tests over the course of one school year, with only the math gains reaching statistical significance. A “slight majority” of schools had positive gains, the report found. But because it didn’t have a randomized control group, “our study is not able to make a very strong statement about whether the personalized learning approaches actually caused the improvement,” Elizabeth Steiner, a study co-leader and an associate policy researcher at the think tank, told me.

The Rand report also found challenges: Helping students work at their own pace can make group projects and collaboration more difficult because students are at different places, while also making it harder to prepare kids for year-end standardized tests. Another review of the Gates-funded effort, by the Center for Reinventing Public Education, concluded that: “At the end of two years, despite some pockets of innovation, few schools had developed replicable strategies for personalized learning.”

Vodicka, aware of such findings, said the results of personalized learning depend on how it’s implemented—and that AltSchool was already addressing many of the concerns identified by the Rand review by, for example, creating flexibility in the pace at which students progress through a course and giving teachers time and resources to develop and evaluate new materials and approaches. “There’s a version of personalization which ends up as computer-based screen time in which the learner basically gives up their agency to the technology,” he told me. “That’s not what we’re trying to do at AltSchool.”

Another school network that has been using technology and personalized learning for years in an effort to eliminate the achievement gap, with mixed results, is Rocketship Public Schools a network of charter schools, which serves low-income Latino and African-American students, mostly in and around San Jose. It has won credit for pushing up test scores but has also been criticized for its heavy reliance on computer-based instruction.

At Rocketship Los Sueños in San Jose, students spend extended periods of time on laptops in their classroom and another 90 minutes in the Learning Lab—a large room where kids spend time sitting at long tables, wearing headphones and working on laptops, supervised by classroom aides. On a visit last June, I found that few things broke the silence: when kindergartners filed in from recess; when a staff member came into the room. Students scarcely talked and when they did, or their attention drifted too far, they were admonished. A supervisor prompted the kids to “focus” and “sit up”. She counted down—“8, 7, 6, 5, 4”—when it was time to switch from one software program to another. The kids looked zoned out, with blank expressions on their faces. “Technology is not a substitute for excellent instruction at Rocketship Public Schools,” a spokesperson for Rocketship said. “Our teachers lead the learning process for every single student we serve.”

When Paul France went to work at AltSchool, he told me, he believed using technology in schools could “minimize the complexity of the classroom ... and make us more powerful, allowing us to do other things with our time.” He helped open three Bay Area schools and worked closely with engineers, testing and developing software. As the school’s buzz grew, he became the company’s poster-teacher, appearing in stories in Wired,  The New Yorker, and Pacific Standard. “He is young, enthusiastic, and enterprising—the kind of teacher every parent would want for their child,” Issie Lapowsky wrote in Wired.

France and his fellow teachers prepared lengthy narrative reports on each child’s progress sent twice a month to parents. For many parents, that wasn’t enough. “Parents are coming to me asking, ‘What are you doing to individualize for my child?’” France said. “Or ‘Why is my child working on the same thing as that kid over there?’” He felt many parents seemed to put more faith in software like Lexia, a reading program, and DreamBox, a math application, than in teachers. They’d ask, for example, why a student who’d already passed a certain standard in DreamBox was still doing similar work in class. “They don’t know how mastery works with kids,” France said. “The programs aren’t really assessing for conceptual understanding, they’re assessing if they answered the question right.”

As time went on, France started questioning what he and his colleagues were doing. “The vision was a curriculum that catered to every child so they’re learning at their level all the time,” he said. “But when every child is working on something different, you’re taking away the most human component in the learning process, which is social interaction—learning from one another and collaborating to solve problems. They’re developing a relationship with their tablet but not with each other.”

For France, the turning point came one morning when he looked around a kindergarten classroom, “and the kids were staring at their tablets, engrossed by them. And I’m thinking to myself, ‘They could be building with blocks, they could be doing a number of different things that are more meaningful that also build social and emotional skills but they’re choosing not to. Why? Because the tool is so addictive, that’s all they want to do.’”

AltSchool representatives say that after four years of experimentation, and a fair bit of trial and error, they’ve solidified their approach and let go of concepts that don’t work, including some of those that bothered Paul France. Vodicka, a former superintendent of Vista Unified Schools in California, was hired last year to build AltSchool’s network of partnerships. The company closed its Palo Alto school and shelved plans to open more of its own schools in favor of providing its platform to more partners. It also ditched the idea of collecting data from classroom cameras. “That was an idea that unfortunately did not validate,” Vodicka says.

Today, 20 charter and private schools, along with five California school districts, are using AltSchool’s technology and receiving training and support. In his visits to partner schools that are using AltSchool’s technology, he says, “I’m seeing learners that are energized and inspired, that are highly collaborative. I’m encouraged by that.”

An AltSchool spokeswoman said its teacher-retention rate is 90 percent. In the summer of 2017, though, France quit AltSchool and returned to Chicago. He spent the past year  teaching at a traditional private school that’s been around for a century and, in his view, sees the “value in using a plain old notebook and pencil to engage in the writing process.” He still uses technology when it’s appropriate. His class last year did a project on Chicago neighborhoods and visited many. Since they couldn’t get to all of them, they used Google Earth to virtually walk through several. “I’m not anti-technology but I’m definitely for minimizing it,” France says. “You use technology to remove a barrier. And the question always should be: Is the tech in my classroom going to preserve or enhance human connection?”

We want to hear what you think about this article. Submit a letter to the editor or write to letters@theatlantic.com.