A tutor gives a student they are working with a pretty typical maths problem: “There are 40 metres of bunting for the school fair. You need to figure out how many metres would be in each section if you cut it into six pieces. What division calculation are you being asked to do?”

“Forty divided by six,” the student replies.

The tutor follows up, “What will you get if you take 40 and divide it by six? Think about how many groups of six you can take out of 40.”

A long pause from the student. Then, “Can you give me the answer, please?”

“Let’s think about it together, instead,” deflects Skye, an AI chatbot tutor that helps students get stronger in maths.

AI in education

Skye is the creation of Third Space Learning, which for the past decade has used humans to tutor mostly Stem university students from India and Asia. In 2025, it shifted almost all of its tutoring to being conducted by AI.

This AI tutor is now being used to support younger students across the UK and the United States, and is still following the gold-standard research on best practices for tutoring, the company says. But Third Space argues that running the AI tutor is more affordable, and therefore it can “scale” - reach more students who need extra help in maths.

“Human tutoring, if I look at the dosage, the number of sessions that schools could afford per student -even when they could afford it - was not enough to be practical, affordable and scalable,” says the company’s CEO, Tom Hooper. “We started looking at [AI] and thinking, ‘If this can deliver evidence-based [tutoring at scale], then that would be transformative.’”

Chatbot teaching

A friendly school-based chatbot and many other AI inventions are not theoretical concepts - they are fast becoming established in education.

Many students, teachers and schools are using AI already. Half of US students report that they’ve asked a chatbot for help with homework, and more than 60 per cent of UK students have done the same. Some of the biggest tech players, including Microsoft and Anthropic, have partnered with the US’ second-largest teachers’ union to instruct teachers on how to use AI in the classroom.

In England, the Department for Education is currently investing in its own AI tutors, due to roll out next year, just as a network of schools that completely replaces teachers with AI-based tutoring apps is exploding in popularity across the US.

The welcome for this AI innovation has not been universally warm. Parents and teachers are worried about how AI and chatbots are already changing, or maybe interfering with, student learning - helping students to avoid the hard work of thinking as they pass off their reading, writing and math problems to Claude or ChatGPT.

Across the US, parents and teachers are pushing back more generally against classroom technology, from phones to laptops and AI, while in England the government has recently announced a ban on smartphones in schools.

Experts say AI, with some guardrails in place, can be a useful tool for learning. Yet perhaps even bigger questions remain about how artificial intelligence could change teaching.

How are teachers using AI in the classroom?

Surveys show that teachers are rapidly adopting AI technology into their classrooms for a variety of purposes. AI participation for educators doubled between 2023 and 2025, according to US-based Education Week magazine, though it’s unclear how the tech is being used. Generative AI is already being employed in popular classroom apps like Canva and Khan Academy’s Khanmigo chatbot tutor, though some are using it in bigger ways, such as to aid in lesson planning.

Meanwhile, three-quarters of teachers in England are using AI in their classrooms, “primarily in resource creation (61 per cent of respondents) but also lesson planning (41 per cent) and administrative tasks (38 per cent)”, according to the NEU teaching union’s State of Education: AI report, published in April. Yet half of schools have no policy whatsoever on how AI can be deployed by either teachers or students, the report says, and while the majority of teachers are using AI themselves, most also do not support the government’s plan for AI tutors.

Some teachers say AI can assist by saving hours of time spent on tasks like making lessons stronger. Samantha Lippert, a third-grade teacher in Western New York in the US, has been using AI to help create lessons incorporating unit vocabulary to increase reading fluency, for example. The district curriculum “has a lot of gaps and pieces that need to be adjusted to effectively teach students”, Lippert says.

“This year I have used AI to take what [the district curriculum] has given us and make it better,” she says. “This has allowed me to also make teaching better because it is creating routine for both myself and the students.”

However, others caution that using ChatGPT to help plan lessons might skip over a key piece of thinking for teachers - getting beyond the structure of activities to the purpose behind the lesson. “Lesson planning is not the same as intellectual preparation,” writes author and educator David Didau. “Without this intellectual work, teachers are left performing lessons they don’t fully understand, no better off than students copying homework from ChatGPT.”

And while AI may be useful for saving busy teachers time on administrative tasks and lesson planning, some of England’s teachers appear alarmed at what the technology is doing to student thinking, and tell the NEU that they don’t want AI tutors in classrooms, even for children with learning gaps. “I am shocked by Ofsted’s recommendation that these technologies should be used when they are not shown to support or develop critical thinking or learning,” one teacher said.

Artificial intelligence as a teacher’s helper

It’s tutoring, though, where much of the excitement around AI currently sits.

Tutoring providers like Third Space Learning are increasingly deploying AI technology for both classroom and out-of-class programs, sold to schools as an assistant to busy teachers that gives targeted help to fill in student gaps in reading and maths. Human teachers still teach the main lessons students receive throughout the school day.

In the past, one-on-one tutoring, considered the gold standard for education, was only for families who could afford to go outside of school and hire a tutor, but now students from all backgrounds can sit down at a computer, put on some headphones and get the same help.

But what that tutoring actually involves - how the AI is being used - can mean a variety of things, says policy researcher Liz Cohen, vice president of policy at 50CAN and author of The Future of Tutoring.

Cohen spent several years studying more than 10,000 tutoring programs, many of them digital, and says, “Right now AI means whatever you want it to mean to whomever is speaking.”

Some tutoring programs merely have added interactive elements based in AI, while others use chatbots to have a conversation with students and reason with them, like Skye.

The quality of AI tutoring programs varies wildly, like with most education technology products, Cohen says. But some AI tutors, like Skye and a US-based literacy tutor called Amira, are based on the research behind “high-impact tutoring” that emerged after the pandemic to combat learning loss. High-impact tutoring recommends three 30-minute sessions per week in a school-based setting, either one-on-one or in small groups, under the supervision of a well-trained adult.

Recent research suggests that human-monitored AI tutoring, in which human tutors refine AI lessons before students use them by “training” the large language models - often referred to as “human-in-the-loop” tutoring - is more effective than human tutoring alone. But Cohen says what’s often being left out of the conversation in both tutoring scenarios is the role humans play generally in helping students to learn if they are behind in school.

“While I don’t have quantitative evidence to back this up, it’s hard for me to have seen all of the tutoring I have seen and not believe that the human relationship of the tutor to student is part of the magic of tutoring,” Cohen says.

Students who received high-quality human tutoring didn’t just improve academically, she says, they also came to school more and were more engaged in their regular classes. For some students, having an adult to connect with at school might have been just as valuable, maybe more so, as the instruction itself.

“What I think we don’t know is what we lose from the trade-off of something that’s more affordable and easier to scale, versus the human piece of it,” she says.

Certain aspects of teaching, like giving feedback, some experts say, may be more effective coming from a human - and students say they prefer it.

“A big risk is this overreliance on the mechanisation of tutoring. It’s very comfortable for everyone if kids are stuck in front of personalised learning,” says Mary Burns, who led the research and writing for the Brookings Institution’s report A new direction for students in an AI world: prosper, prepare, protect. “The danger is indiscriminate adoption and overreliance on these systems - the ‘platformisation’ of education where we increasingly outsource instruction and assessment to software.”

While schools might be rightly concerned about students offloading their thinking to AI, the same concern could be felt about teachers, she says, who could easily offload tasks like giving feedback to a chatbot as well.

“We know from online learning that being there and caring and presence - it’s critical to student learning and satisfaction with education,” she explains. “If teachers let AI give the feedback, it will reduce their role in that, narrowing the education experience.”

The end of teaching content?

Some schools are going further with AI, beyond tutoring to an entirely new model of schooling where the instructional part of teaching is being performed by artificial intelligence, with human teachers rebranded as “guides” who provide motivation and social-emotional support.

The new Alpha School model in the US has shortened the time students spend on academic learning to two hours per day, with the rest of the school day spent on life skills and interesting projects.

The private-school group, which costs between $10,000 and $65,000 (£7,400 and £47,800) per year to attend, has grown to 22 locations including Austin, Miami and Washington DC in the past few years. However, the school’s founders say a more affordable version is in the works, to help 1 billion children use their software to “reach their full potential through AI-powered tools, personalised learning and life skills that prepare them for real-world success”.

Students spend a couple hours each morning learning content entirely from AI-based apps on laptops or tablets, personalised to each student’s needs and learning gaps. The apps have been built using the principles of learning science, and include direct instruction and spaced retrieval practice, with built-in attention to cognitive load - the idea being that learning apps that follow evidence on learning will help students learn better and faster.

“The key insight around generative AI is where you generalise personalised lesson plans based on learning science, and where kids are motivated to engage with them,” says Joe Liemandt, Alpha School’s principal. Liemandt, who made his fortune in software, has committed $1 billion to developing Alpha School and its proprietary software. “There is good screen time and bad screen time, and bad and good AI. If chatbots are used as cheatbots, if that’s going to fix education, we are doomed,” he adds.

Time on academics has been compressed while student learning has increased. Liemandt says student growth is happening at twice the average rate for most students - a concept that might take eight hours to master in a typical classroom gets cut down to four hours, because apps are personalised to each student’s strengths and weaknesses, and because Alpha students are motivated to learn. In fact, he claims that technology is only responsible for 10 per cent of their success, while motivation accounts for the other 90 per cent.

Sparking that motivation is the role of the guides - many of them former teachers. No longer needed for content expertise, Liemandt says, guides can focus on interaction with the students, figuring out what motivates them and providing emotional support.

Education researcher Carl Hendrick, co-author of books like How Learning Happens and Instructional Illusions, has helped Alpha School design its own proprietary learning platform, called Timeback, which adheres closely to evidence-based instruction. He says the Alpha way might be solving the hardest problem for teachers: perfecting learning through appropriately paced, precise curriculum design in a way that humans can’t match.

“We have a system that’s kind of dependent on this one-in-20 hero teacher who’s killing themselves. What if we were able to deliver curriculum design? Forget about the human aspects of it, which are hugely important, by the way - the ability to scale up really high-quality instructional materials, I think, is really interesting,” he says.

When learning the content gets more efficient, Hendrick adds, teachers have more room to really get to know their students and work with them. The paradox of Alpha School, he explains, is that kids there are on screens less, and with humans more, than the average student. “They do two hours, and then they go, ‘I’m done,’” Hendrick says. “Then for four to six hours, they’re making stuff.”

People were turning to technology for a shortcut around the messy work of teaching and learning academic material long before the arrival of Alpha School, detractors say. For decades, tech companies and school systems alike have tried to reinvent, compress and remake what sceptics say is a non-linear process that fundamentally happens between humans.

Benjamin Riley, who left his job as executive director of science-of-learning non-profit Deans for Impact to dedicate his time to studying the effects of AI on learning, is now founder of Cognitive Resonance, focusing on building human understanding of AI and tamping down the hype. The public is, in his view, falling for the same old tech promises.

“The same premise that goes from these recent versions all the way back to the sort of big data analytics craze and the personalised learning craze of the 2010s is this notion that you can take the data, figure out what the gaps are, and then address those gaps. But that’s not how learning works,” he says.

“Even if we can assume that a kid actually wants to engage with the product, there’s no way that the technology can figure out what’s going on in the mind of a young kid who’s grappling with something and trying to make sense of a concept that is new to them. This is the whole skill and art of teaching,” Riley goes on. “So, you know, it’s doomed to fail.”

Others are less pessimistic. Former teacher George Greenbury, who helped to create strikingly human AI chatbot tutor Inkling, says their product was designed to support students not yet at the expected standard, or those in need of a study companion. But some schools are finding it comes in handy as a full-time teacher if a teacher leaves, or they struggle to find a replacement.

“You’ve got schools who are using the technology more widely with their students, because otherwise their students would not have a teacher,” Greenbury says. He believes an hour spent with the AI tutor is more effective for students than an hour spent in his classroom.

“Truthfully,” he says, “I did not expect to find something that could teach my students better than I can, but ultimately, it can teach my students better than I can.”

Whether or not the students would agree is perhaps the question we should be asking next.

Holly Korbey is a freelance writer