Kids aren’t as sneaky as they think they are. 

They do try, as Holly Distefano has seen in her middle school English language arts classes. When she poses a question to her seventh graders over her school’s learning platform and watches the live responses roll in, there are times when too many are suspiciously similar. That’s when she knows students are using an artificial intelligence tool to write an answer. 

“I really think that they have become so accustomed to it, they lack confidence in their own writing,” Distefano, who teaches in Texas, says. “In addition to just so much pressure on them to be successful, to get good grades, really a lot is expected of them.”

Distefano is sympathetic — but still expects better from her students. 

“I’ve shown them examples of what AI is — it’s not real,” she says. “It’s like margarine to me.”

Educators have been trying to curb the use of AI-assisted cheating since ChatGPT exploded onto the scene. 

It’s a formidable challenge. For instance, there’s a  reserved for tech influencers who rack up thousands of views and likes teaching students how to most effectively use AI programs to generate their essays, including step-by-step instructions on bypassing AI detectors. And the search term for software that purports to “humanize” AI-generated content spiked in the fall, , only to fall sharply before hitting the peak of its popularity around the end of April.

While the overall proportion of students who say they’ve cheated , students also say . 

But there may be a solution on the horizon, one that will help ensure students have to put more effort into their schoolwork than entering a prompt into a large language model.

Teachers are transitioning away from question-and-answer assignments or straightforward essays — in favor of projects. 

It’s not especially high-tech or even particularly ingenious. Yet proponents say it’s a strategy that pushes students to focus on problem-solving while instructing them on how to use AI ethically. 

Becoming ‘AI-Proof’

During this past school year, Distefano says her students’ use of AI to cheat on their assignments has reached new heights. She’s spent more time coming up with ways to stop or slow their ability to plug questions and assignments into an AI generator, including by giving out hard copy work. 

It used to mainly be a problem with take-home assignments, but Distefano has increasingly seen students use AI during class. Kids have long been astute at getting around whatever firewalls schools put on computers, and their desire to circumvent AI blockers is no different. 

Between schoolwork, sports, clubs and everything else middle schoolers are juggling, Distefano can see why they’re tempted by the allure of a shortcut. But she worries about what her students are missing out on when they avoid the struggle that comes with learning to write. 

“To get a student to write is challenging, but the more we do it, the better we get.” she says. “But if we’re bypassing that step, we’re never going to get that confidence. The downfall is they’re not getting that experience, not getting that feeling of, ‘This is something I did.’” 

Distefano is not alone in trying to beat back the onslaught of AI cheating. Blue books, which college students use to complete exams by hand, have had a  as professors try to eliminate the risk of AI intervention, reports The Wall Street Journal. 

Richard Savage, the superintendent of California Online Public Schools, says AI cheating is not a major issue among his district’s students. But Savage says it’s a simple matter for teachers to identify when students do turn to AI to complete their homework. If a student does well in class but fails their thrice-yearly “diagnostic exams,” that’s a clear sign of cheating. It would also be tough for students to fake their way through live, biweekly progress meetings with their teachers, he adds. 

Savage says educators in his district will spend the summer working on making their lesson plans “AI-proof.” 

“AI is always changing, so we’re always going to have to modify what we do,” he says. “We’re all learning this together. The key for me is not to be AI-averse, not to think of AI as the enemy, but think of it as a tool.”

‘Trick Them Into Learning’

Doing that requires teachers to work a little differently. 

Leslie Eaves, program director for project-based learning at the Southern Regional Education Board, has been devising solutions for educators like Distefano and Savage. 

Eaves authored the board’s , released earlier this year. Rather than exile AI, the report recommends that teachers use AI to enhance classroom activities that challenge students to think more deeply and critically about the problems they’re presented with. 

It also outlines what students need to become what Eaves calls “ethical and effective users” of artificial intelligence. 

“The way that happens is through creating more cognitively demanding assignments, constantly thinking in our own practice, ‘In what way am I encouraging students to think?’” she says. “We do have to be more creative in our practice, to try and do some new things to incorporate more student discourse, collaborative hands-on assignments, peer review and editing, as a way to trick them into learning because they have to read someone else’s work.”

In an English class lesson on “The Odyssey,” Eaves offers as an example, students could focus on reading and discussion, use pen and paper to sketch out the plot structure, and use AI to create an outline for an essay based on their work, before moving on to peer-editing their papers. 

Eaves says that the teachers she’s working with to take a project-based approach to their lesson plans aren’t panicking about AI but rather seem excited about the possibilities. 

And it’s not only English teachers who are looking to shift their instruction so that AI is less a tool for cheating and more a tool that helps students solve problems. She recounts that an automotive teacher realized he had to change his teaching strategy because when his students adopted AI, they “stopped thinking.” 

“So he had to reshuffle his plan so kids were re-designing an engine for use in racing, [figuring out] how to upscale an engine in a race car,” Eaves says. “AI gave you a starting point — now what can we do with it?”

When it comes to getting through to students on AI ethics, Savage says the messaging should be a combination of digital citizenship and the practical ways that using AI to cheat will stunt students’ opportunities. Students with an eye on college, for example, give up the opportunity to demonstrate their skills and hurt their competitiveness for college admissions and scholarships when they turn over their homework to AI. 

Making the shift to more project-based classrooms will be a heavy lift for educators, he says, but districts will have to change, because generative AI is here to stay. 

“The important thing is we don’t have the answers. I’m not going to pretend I do,” Savage says. “I know what we can do, when we can get there, and then it’ll probably change. The answer is having an open mind and being willing to think about the issue and change and adapt.”

Students who attend charter schools in Newark, including English learners and those with learning disabilities, are less likely to transfer out within two years than their peers at the district’s public schools, according to a new study. Children were also significantly more likely to shed their special-needs classification while enrolled at a Newark charter, the authors note.

The study, released as this week and Education Next, offers more perspective on the long-running debate over admissions and retention in the charter sector. Critics of the publicly funded but privately operated schools that they push out kids with learning, language or behavioral challenges like suspension or expulsion.

Co-author Marcus Winters, a professor of educational leadership and policy studies at Boston University, said in an interview that he believed individual schools of all kinds “inappropriately” encouraged some students to leave. But the Newark study, along with looking at schools in Tennessee and North Carolina, has disproven the notion that charters routinely engage in the practice, he argued.

“I do think it’s fair to say that our paper … has now sufficiently debunked the myth that charter schools — at least in these areas that have been studied — are systematically pushing these students out,” Winters said.

Winters and co-author Allison Gilmour, a professor at Temple University, set out to compare enrollment trends in Newark, a city with one of the largest charter school sectors in the country. To do so, they used data from the Newark Enrolls assignment system, which allows families to select among their choice of traditional schools and approximately 70 percent of the city’s charters. (Not all local charters participate in Newark Enrolls, but those that do account for about five-sixths of charter students.)

Several variables in the Newark Enrolls formula determine which students are assigned to certain schools, including each child’s rank-ordered school preferences; individual factors prioritized by various schools (such as sibling preference); and randomized lottery numbers that are used in case a given school is overenrolled. By gathering administrative data between the 2014-15 and 2017-18 school years, Winters and Gilmour were able to compare patterns of school entrance and exit for nearly 14,000 students.

In all, children attending Newark charter schools were 22 percentage points less likely to leave that school within two years than substantially similar students who were instead assigned to traditional public schools. English learners were 16 percentage points less likely to transfer out of a charter, and students with a disability nearly 11 percentage points less likely. The difference for Hispanic students was not statistically significant.

The smaller chances of transfers may be attributable to the system’s format of ranked school preferences. In a model that controlled for families’ ranking of schools, charter students were still less likely to leave within two years, but only by about 10 percentage points; that suggests that a sizable portion of the charter school effect is simply a reflection of students attending the school they wanted to go to in the first place, Winters said.

“You might just be more willing to stick it out with a school that you originally had as a higher preference,” he said. “If you’re attending a school that you went to on purpose, you’re just less likely to leave it. And if you’re going to a charter school in a place like Newark, where several of the charter schools are among the most popular choices … you’re probably going to one of your most highly preferred schools.”

By tracking the same students over time, the study also observes gradual movement within individual subgroups. Specifically, children with a special-needs classification at charter schools are much less likely to still have an Individualized Education Program a few years later — a phenomenon that may help explain why the percentage of students receiving services is lower in charters. The effect is particularly notable for children entering charter schools between kindergarten and third grade (31 percentage points more likely to lose a special-needs classification within three years) and between grades four and six (20 points). Those findings dovetail with research pointing to similar trends in special-needs assignment at Boston charter schools.

Comprehensive examinations, including from the federal Government Accountability Office, have shown that charters generally teach smaller numbers of kids with disabilities than district schools. More recent evidence indicates that those gaps may be shrinking, though it’s unknown how the huge upheaval triggered by COVID-19 may have shuffled enrollment trends.

If the study raises doubts about the claim that charter schools consistently work to remove struggling or hard-to-teach students from their classrooms, it offers little clarity about how they approach recruitment. At least that charters in multiple states were less likely than district schools to respond to application inquiries from parents of children with severe disabilities.

Winters concluded that the population differences between sectors could arise from only one of three sources: recruitment of students, mobility of students once enrolled and (in the case of English learners and disabled students) changes in status classification such as those detected in the Newark study. More investigation was needed into how different schools attract families, he said.

“It’s clear to me, at least, that the major driver in these enrollment gaps is who’s enrolling in the first place. We need more information about the enrollment side.”