Don’t Throw Ed Tech Babies Out With the Bathwater

When used wisely, technology, including AI, can be helpful in the classroom.

Photo by Rona Vorontsova on Unsplash

---

A backlash against technology in the classroom is sweeping the country, part of a general wave of alarm about data centers, screen addiction, privacy, and the risk that unfettered access to generative AI will lead to a decline in students’ cognitive abilities.

I get it. I’ve been concerned for years about giving kids access to devices, and those concerns have only increased with the advent of ChatGPT and its ilk. Still, there are uses of technology, and of AI in particular, that can actually improve teaching and learning rather than undermining it. And a blanket call to rid classrooms of technology could sweep those beneficial uses out the door along with the pernicious ones.

How can you tell which technology is going to be helpful rather than harmful? It helps to understand, as I’ve argued before, that some cognitive effort is essential to learning—and that subcontracting out that effort to a device will also subcontract out the learning.

That suggests that we need to be much more cautious about student-facing technology than about technology used by teachers. Students, by definition, are in school to learn.

Taking Burdens Off Teachers

Teachers, on the other hand, are there to do a job—teaching—that can be overwhelming. We may not want AI handling tasks like grading or determining whether a student can graduate—especially if it’s a generic tool like ChatGPT or Claude. But if a specialized AI tool can take some burdens off teachers’ shoulders, in a way that has adequate safeguards, it can make their jobs more manageable and improve their effectiveness.

A case in point is “Judy,” an AI-powered tool developed by The Writing Revolution, an organization that trains teachers in the Hochman Method of writing instruction. (I am the co-author of a book describing the method and serve on the organization’s advisory board.) The method is designed to be adapted to the content of whatever curriculum is being used, which means that teachers often need to create the writing activities for their students. That can be challenging and time-consuming.

But “Judy”—named after Judith Hochman, the method’s creator—can do that work for teachers. A teacher could, for example, ask Judy to create a sentence expansion activity for fifth-graders embedded in content about the human digestive system. Of course, the teacher still needs enough understanding of the method to be able to check whatever activities Judy provides. But the tool has been carefully trained on Writing Revolution materials and (I’m told) generally does an excellent job.

On the reading side, some teachers are using specialized AI tools to create “decodable” texts that are accessible for beginning readers. While the tools haven’t been studied much and may not be perfect, they at least allow students to practice reading connected text as opposed to isolated words. Researchers have found that it’s crucial for early readers to practice on connected text—and many don’t get the opportunity as often as they should.

Student Use Is Trickier

It’s also possible to design truly beneficial AI tools for students to use, but it’s trickier. AI tutors can be programmed to ask students questions that guide their learning rather than just spitting out finished answers or essays. The tools can also be designed to provide the kind of repeated explanations and practice that can lead to mastery of a specific skill or body of information. But students are likely to be less motivated to use a tool that requires them to expend effort, and they’re often adept at getting around whatever safeguards have been built in.

A generative AI tool may initially refuse to do the work for a student but, after repeated requests or even flattery, eventually give in. As New York Times reporter Dana Goldstein discovered, it’s possible to get around a bot’s refusal to write a specific essay by just asking it for an example of what such an essay might look like.

Some school systems—notably New York City and Los Angeles—have recently banned or paused AI use by students to varying degrees. That may make sense as a temporary measure, giving officials time to identify what AI tools are being used in classrooms and evaluate the evidence behind them and the safeguards they incorporate. But bans of tech or AI that fail to distinguish between helpful and harmful uses can end up holding students back.¹

Identifying Dyslexia More Accurately

One AI-powered tool that has recently come under attack from parents and policy makers is called Amira. Grounded in decades of research that began at Carnegie Mellon University, Amira listens to children reading aloud and analyzes their performance. It has two versions: a “tutor” that provides kids with personalized practice and a “screener” that flags them for reading difficulties.

According to Amira CEO Mark Angel, the screening tool has been trained on a huge variety of children’s speech and reading patterns, enabling it to distinguish between children who exhibit true signs of dyslexia—and therefore need specialized support—and those who simply need better mainstream instruction.

Some school systems screen kids by having teachers listen to individual students read. In those cases, Angel said, the advantage of a tool like Amira is that it saves massive amounts of time. But many schools have switched to computerized systems that do the listening for teachers, as Amira does. One problem with those systems is that they’re often biased against non-native English speakers or children with non-standard accents, erroneously identifying them as dyslexic—with potentially life-altering consequences.

But a tool like Amira, with its wide training, is less likely to exhibit that bias. A video produced by Amira provides the following example: A child who is a Spanish bilingual speaker reads the word violin as bee-oh-leen. Amira is able to recognize that the child is simply transposing Spanish pronunciation into English. But in the same sentence, the child reads the word her as here. Amira flags that misreading, which doesn’t align with the rules of Spanish pronunciation, as a true problem.

Amira is now being used in every state in the U.S.—and in some other countries—and has been adopted in 14 states, meaning that the state will cover the costs of using it. According to Amira, studies show “significant reading growth for students from a wide range of backgrounds, with particularly strong results for English Language Learners and other at-risk populations.” Fewer such students are being misidentified as dyslexic.²

Numerous studies have been conducted, including by third party researchers, showing Amira’s benefits when compared to other approaches. A study done in Louisiana public schools has met the federal government’s standards for evidence of effectiveness. And an independent review of Amira, including observations on how the tool is used in classrooms and a summary of the research, was recently published as part of an “AI in Action Learning Tour” report from Instruction Partners. The report observes that when implemented well, Amira can save teachers time and provide students with more reading practice and more personalized feedback than they would otherwise receive.

Pushback Over Privacy Concerns

But in some places—particularly New Mexico, which recently started requiring schools to use Amira for literacy screening in K-2—parents and legislators have been pushing back against the technology. Parents are particularly concerned that Amira records students’ voices, a fact that was apparently not communicated to them beforehand, and stores them.³

Last month, a bipartisan group of 30 state lawmakers sent the New Mexico department of education a letter asking that schools obtain parental consent before allowing students to use Amira. The letter asserted that “Amira has access to students’ names, gender, birthdays, locations, and other demographic and sensitive information” that could be “weaponized” against the state’s “vulnerable populations.”

The legislators also claimed that children’s voice recordings were being used to “train the model and influence its future output” in ways that are not “appropriate.” One parent told a reporter that Amira was “using our children as free labor to train their tool for corporate profit.”

In response, some New Mexico districts have discontinued their use of Amira—even as others say they’ve seen success with it. The Albuquerque school district reported that after it adopted Amira last year, first and second graders there received the highest reading scores since at least 2021. According to Angel, Amira’s CEO, the rate of dyslexia identification in the state has decreased from over 40%, based on other voice recognition screeners, to 16.3%, a figure that is in line with the national average.

And the claims about Amira collecting and potentially misusing student data are, according to the company, simply not true. Amira doesn’t collect biometric or any other kind of data, Angel told me. The Amira website’s “trust center” page says the voice recordings are de-identified and encrypted, never provided to any commercial third parties, and stored only long enough for educators to review and learn from them. The company has never had a data breach.

While the tutoring version of Amira—which is not required by the state—does use students’ voices to train its model, that’s not true of the assessment version. In any event, Angel said, Amira is now giving New Mexico districts the option to have the recordings “zapped” immediately, after 48 hours, or at the end of the month.

A Failure of Communication

The controversy over Amira appears to stem largely from a failure of communication. Amira has ample information about its privacy and other protections on its website, along with evidence of efficacy, but the company has no way of communicating with parents or other stakeholders directly. And in the case of New Mexico, at least, government authorities didn’t do much to make accurate information generally available.

Of course, there are some ed tech and AI tools that do present risks to student privacy—and many that have little evidence behind them. The questions that parents are asking are “valid questions,” Angel said, “and we can do better to make sure the answers are clear.”

But in the current atmosphere of suspicion of technology, and particularly AI, even better communication might not be enough to allay the public’s fears. Angel says that while Amira has been around for a while, the company hasn’t seen pushback until the last three months, as the general ed tech backlash has gathered steam.⁴

Angel compares a tool like Amira to the kinds of supports available in the medical profession. You wouldn’t expect a doctor to personally draw blood or conduct blood tests; their job is to focus on diagnosis. But, he says, “it seems more difficult for people to apply the notion to education.”

Amira isn’t replacing teachers, he points out, nor is it taking away a core part of their job. It’s enabling them to be more effective. “It’s a giant step up from what has been out there before,” he says. “Just like a doctor with an MRI can do a better job than a doctor without an MRI, a teacher with Amira is better than a teacher without one.”

We do need to be cautious about AI in schools, especially when it’s used as a shortcut that enables students to avoid learning. At the same time, it’s simplistic to adopt policies riding on a wave of generalized suspicion of technology, even if there are many instances where that suspicion is justified. Armed with evidence and an understanding of the science that explains how learning works, we need to be able to recognize when AI is capable of transforming students’ trajectories for the better.

Update, 9.28.26: Information about Amira from Instruction Partners’ recent report on AI usage was added.

1. The bans apply only to in-school AI use. A much more difficult issue to address is the rampant use of AI by secondary students to do their homework, including writing assignments.

2. Angel told me that the New Mexico English Language Learner data is still being reviewed, but a whitepaper on Amira’s website, dated May 2026, reports that “Amira is flagging 26% FEWER ELLs than the keyboard test” used in previous years. Under the previous test, the document says, 54 percent of English-learner students “were flagged as struggling readers” at mid-year. “Under Amira, that figure fell to 40%. For non-English learner students, the rate also fell, from 29% to 24%.”

3. New York’s moratorium on AI use will require schools to disable Amira’s AI component, according to a city education official.

4. The recent alarms about the risks of unfettered AI research will likely add fuel to the backlash, even though they have nothing to do with the tools used in classrooms.