Andrea Hinojosa can no longer imagine doing her teaching job without the assistance of artificial intelligence tutoring.
The tool has been a game changer: Her students have more opportunities to practice their writing, they receive swift and thorough feedback, and her multilingual students demonstrate their understanding of content across a language barrier.
“AI has really just changed how we can do our jobs,” said Hinojosa, a history teacher at Copper Hills High School in Utah. “I think it’s really improved how we can do our jobs and the things we can offer our students.”
Since the explosion of large language models like ChatGPT, the technology has begun to seep into classrooms across the nation. Advocates say schools can use it to enhance course offerings and downsize workloads for overworked and underpaid educators.
Some are using it even more aggressively: In at least one cyber charter school network, the goal is for the technology to lead classes, adapting lessons in real time in response to how students are understanding the material.
A growing number of districts are beginning to embrace AI tutoring, as the technology allows for teachers to double the practice they can offer, without adding to the grading they must then do. Students, like those in Hinojosa’s class, are able to write more, building up those skills ahead of Advanced Placement tests. And Hinojosa doesn’t have to spend hours reading and penning feedback, getting exhausted in the process.
Though the concept of AI tutoring is not new, the leap in accessibility in large language models has made using AI much simpler—it allows for more organic interactions with AI that hadn’t existed before, said Zachary Pardos, an associate professor of education at the University of California, Berkeley studying adaptive learning and AI.
For example, students can get instant feedback on their writing. An AI tutor can define a word or explain context a student doesn’t understand. And a math tutor can quickly adapt to a student’s skillset if the problems are too challenging.
“It seems like this is really low-hanging fruit, in a way it wasn’t before,” Pardos said.
In Hinojosa’s class, that’s Class Companion, an AI platform that scores students’ writing. It’s revolutionized what she’s able to assign to help students prepare for the AP exams, which are composed of essay segments. MagicSchool AI provides feedback, allowing students to adjust their writing and try again, before they submit it to Class Companion for a grade.
“I think the most important thing is they’re doing a lot more writing, and they’re getting a lot more instant feedback than ever before,” Hinojosa said.
AI tutors add efficiency, but educators should still be cautious
Pardos, who leads a team that developed the AI tutor OATutor, said studies show AI helps improve classroom efficiency, compared to manually trying to create new questions or material. It ultimately can save time and money for districts—not by doing away with teaching roles but perhaps by cutting down on technology vendors and instructional materials, Pardos said.
In one study, a random trial of 1,800 K-12 students from underserved communities found that, when human tutors used an AI app, Tutor CoPilot, their students were 4 percentage points more likely to master mathematics topics than students whose tutors did not have access to the app.
Tutors with the app were also more likely to ask more guiding questions, rather than giving away the answer. The app also helped human tutors respond better to student needs, the study found.
There were some flaws with the app—like the AI generating questions above a student’s grade level. But largely, the research found that it was helpful for equitable tutoring.
There are things for educators to be wary about, though, Pardos added. For one, when you incorporate technology from large companies, it begs the questions: Who is setting the curriculum and who is dictating pedagogy?
“If something gets too deeply adopted from a company, and they’re not listening to the district or the students or the teachers anymore, it kind of feels like co-opting,” he said.
There’s also the possibility for bias within AI. AI tools are trained by humans, and can reflect societal biases that can shape AI’s responses to people from different backgrounds. Experts say it’s important for educators to be aware that the technology can have bias against race, gender, or household income—as well as if student is an English learner, has learning differences, or performs on grade level.
There has been less research on what large language models, like ChatGPT, can do for learning, particularly when it comes to people from different backgrounds, Pardos said.
“I’m not a pessimist here. I think the promise is greater than the risk in this regard,” he said. “I hope what we find is that we’re delighted that actually this benefits those who most need educational benefit, but I’m eager to see what the answer is to that question and how we can evaluate that in a more uniform way.”
In one district, AI tutors help kids with math and reading
As generative AI exploded in access, there were early adopters who began integrating the technology into their classes almost immediately. And there were the skeptics.
The Sante Fe public schools rolled out a two-year plan for integrating AI into its processes, trying to embrace both groups of educators. In year one, which the district is now in, the work has focused largely on professional development and training teachers in how to use AI in the classroom. A small pilot group—those early adopters—has been testing some AI-powered tutoring resources in class.
“Artificial intelligence is moving way faster than we can move as a district, and so we needed to be flexible,” said Neal Weaver, the New Mexico district’s chief information and strategy officer. “Yet, we also needed to be very cognizant of the demands and expectations on our teachers.”
Some of the professional development has helped educators more effectively use software the district had purchased years before. One is an adaptive math tutor, which responds to how students answer math problems. Another, Amira Learning, helps students in grades K-6 with reading instruction, said Vanessa Romero, the district’s deputy superintendent for teaching and learning.
Amira records students as they read and provides immediate feedback. If a student mispronounces a word, or doesn’t understand the context, an AI character helps them work through it. It provides teachers with students’ reading levels, and tailored lessons, so teachers can focus on what students are struggling with.
“In the past, myself as a teacher, we did running records, where we had to actually sit and listen to every kid read to determine where the reading level was,” Romero said. “I’d sit there for hours. It would take me a week to assess all my students’ reading.”
AI tutoring opens up new space and time in the classroom, said Kate Grunewald, the innovation coordinator in digital learning for the district, who is overseeing the pilot programs. With personalized, customizable technology, students can work in small groups or at stations, freeing up a teacher’s time to work one-on-one or with another small group.
But there are still some risks educators have to consider: These tools can generate responses based on outdated information or fabricate facts, they can reflect societal biases and amplify harmful stereotypes, and there are data concerns regarding student privacy.
The phased approach is giving leaders time to fully understand the technology before it can become incorporated into classrooms in year two.
“I think with any kind of ed tech, there’s this tension of: We want to be mindful and move intentionally and then we feel like, ‘Oh my gosh, things are advancing so fast,’” Grunewald said. “We don’t want to just adopt something just to adopt it. [We’re] really asking ourselves: ‘Is this working for our students? Is it working for all of our students? And, how do we know?’”
In Utah, Hinojosa is able to see the payoff for her students. She is not a Spanish speaker, and one of her students doesn’t speak any English. But with the AI tutoring tool, she’s been able to see that he’s learning and able to communicate what he’s learning, even when he writes in Spanish.
In the past, she might not have assigned a written response, given the language barrier. But now, English learners are able to write, and she can see what they know and where their skills are—immediately.
“It’s amazing,” she said. “I still have to teach my students those explicit skills and reinforce what that looks like. But then, the AI enables me to quickly assess whether they’ve got those skills.”
