
Kiran Parikh
AI tutor technology has risen to prominence, exemplified by the release of Khan Academy’s Khanmigo.
Alpha High School, a small private school based in Austin, boasts an average SAT score of 1530 for its Class of 2025 seniors.
Students arrive at 8:30 a.m, engage in a motivational activity, and then finish the learning day with only 3 hours of classes. Interestingly enough, the school claims that their methods “ensure mastery of material” twice as fast as “traditional methods.”
And perhaps even more interestingly, at Alpha High School, there are no teachers.
This is because their students learn from Artificial Intelligence (AI) tutoring technology instead of traditional human educators as a part of the 2 Hour Learning program, which claims to use “adaptive AI technology” to help kids learn “2x faster”.
As the flagship campus of the 2 Hour Learning program, Alpha High School is the perfect example of a growing trend that is poised to change the way students learn around the globe: the integration of AI technology into education.
One of the most significant applications of AI in the education sector is the AI tutor, a chatbot, like ChatGPT, that can respond to questions from users and serve as a nonhuman tutor. Multiple companies have come out with AI tutors; in particular, Khanmigo, an AI tutor developed by the founder of KhanAcademy, recently gained significant attention for being able to serve as a teaching assistant for teachers and a tutor for students. Khanmigo is now even being integrated into some schools.
“(AI tutors), if they’re personalized, can teach you what you don’t know,” Abhiramon Rajasekharan, a PhD researcher at UTD specializing in natural language understanding, said. “They’ll know more about you, so they can teach you in a style that you are the most comfortable learning in. And obviously these models have knowledge in a lot of different topics, so they might actually know a lot more about what you’re being taught than your teacher.”
One of the most important aspects of the integration of AI into education is the potential for AI tutors to “democratize” education, closing the educational gap that causes poorer students to receive worse educational outcomes. The creation of AI tutors now makes it possible for anyone with a laptop or phone to have access to a nonhuman teacher. Ironically, however, the “AI-powered” Alpha High School has a tuition of around $40,000.
Martin Stegemoeller, the Malcolm K. and Minda Brachman Master English Teacher Chair, has recently made an effort to encourage his students to use AI in productive ways. He views the significant developments in AI tutors as largely positive.
“If you’re curious, you have some minimal knowledge and you have the time and the discipline, you have the most amazing tutor possible 24/7,” Stegemoeller said. “And for most people, that could be a great add-on (to traditional learning).”
AI tutors also have the ability to test knowledge at a significantly deeper level than traditional paper-based tests. Not only are the bots able to develop an understanding about a student’s educational needs, but they can also adapt their teaching style based on those specific needs.
“If there were AI-based assessments that could really tell if you remembered anything, then you’re in a different situation,” Stegemoeller said. “(Students) could be learning so much from the bot and going faster, because they’re being assessed on retained learning and motivation to learn rather than short-term crammable assessments.”
Stegemoeller believes that AI has already reached a level of sophistication that allows for serious, tangible use. However, it’s mainly up to the students as users to provide good prompts to the bot. While AI tutors have the potential to revolutionize teaching, many rely on the technology to minimize effort rather than learning.
“My own students are not very good at prompting toward independent learning,” Stegemoeller said. “They’ve almost never tried. If a teacher allows it, they ask themselves, ‘How do I use AI to minimize what I put in and get out what I need for my assignment?’ That’s not wrong. People are just much better at using it to cut corners than they are (at using it) to learn.”
These AI tutor bots are built on what are called large language models, or LLMs for short. LLMs are trained on huge datasets, allowing them to become extremely proficient in natural language processing (NLP) tasks, meaning that they are able to understand prompts and respond to them in human-like ways. LLMs are the basis for most chatbots, the most famous being ChatGPT. In this case, AI tutor bots built on LLMs are specialized toward the purpose of serving as tutors that can effectively respond to questions or prompts by students trying to learn.
“With LLMs, companies can now build applications which can do the whole search process for you,” Rajasekharan said. “It can go through all these articles and go through textbooks and go through lectures on the internet and on YouTube, and it can figure out the information that you’re looking at, and it can teach students exactly the information they want.”
A hybrid between a rigorous in-person school like St. Mark’s and AI tutor technology has the potential to be extremely beneficial for students. Kids who are truly “above and beyond” have the opportunity to explore more advanced classes with their AI tutors while still maintaining human relationships with teachers.
“You can imagine a hybrid between the way St. Mark’s is now and the way the Alpha school is, which might be a great compromise,” Stegemoeller said. “Some kids who are beyond multivariable calculus or already taking physics classes could demonstrate (their proficiency) in other ways.”
AI tutors are already finding uses in classrooms nationwide, with many schools choosing to integrate this new technology as part of certain pilot programs. Khanmigo claims to be used by more than 400 school districts, an impressive number considering it was released for early use in March 2023. The use of such technology seems poised to only increase as the government begins to take interest in the topic of AI in education, as evidenced by the recent executive order “Advancing Artificial Intelligence Education For American Youth,” declared on April 23, 2025.
In fact, AI tutors are a viable next step in the field of educational technology and have the potential to solve the infamous Bloom’s 2 sigma problem.
A famous paper published in 1984 by educational psychologist Benjamin Bloom found that the average student receiving one-on-one tutoring with mastery learning techniques performed around two standard deviations better than the average student learning in a conventional classroom setting; the tutored students were shown to be above 98 percent of the control students.
This finding, now called Bloom’s 2 sigma problem, established the key idea that a student-teacher ratio of 1:1 was optimal and that mastery learning techniques, which prioritize the mastery of a concept before moving on to harder ones, were the best way to teach. While this revelation has been widely known for many decades now, the traditional classroom learning that Bloom critiqued in his paper has remained the main method of teaching, as it has been impossible to find a way to widely implement his solution.
Now, as AI is being integrated into education technology, the solution to Bloom’s 2 sigma problem seems closer than ever. AI tutors provide a one-to-one tutoring experience for anyone with a laptop, ensuring “personalized” education. On top of that, these tutors can be designed to instruct with mastery learning techniques, as with the 2 Hour Learning program.
However, AI tutors are not perfect yet, and are sometimes susceptible to an LLM-affecting phenomenon known as “hallucination.” These “hallucinations” are caused by a multitude of issues during training of an LLM, and they can cause AI tutor bots to say wrong answers with complete confidence. At the end of the day, these bots are not human and do not have a complete understanding of the world.
Although human tutors make mistakes as well, the singular thing AI tutor bots lack compared to their human counterparts is the aspect of human connection.
“You come to love the particular angle on life that a person or a teacher has,” Stegemoeller said. “Think of your favorite teacher ever. You probably respect their angle, their humor, their way of talking and their way of using shorthand.”
There is something unique and special about relationships between humans that simply cannot be replaced by AI. AI tutors do not have personalities or sentient thoughts — their intelligence is, after all, artificial. The simple action of thinking about another human being is what separates humans from AI.
“People might even be able to train (a bot) to have its own vibrant personality,” Stegemoeller said. “However, when another human being chooses to love you, that means a lot, and an AI bot cannot choose to love you. I don’t think that will ever be replaced.”
As the data sets LLMs are trained on become more vast and accurate, hallucinations will occur less frequently; the quality of AI tutors will grow asymptotically close to or perhaps even surpass that of human teachers. Stegomoeller believes that this level of sophistication is closer than most individuals think.
“I think this is going to happen decades before anybody would have guessed just a few years ago,” Stegomoeller said. “In one to two years, you’re going to have AI that is as good as anyone in the world. And then four years, five years out, you’re going to have AI smarter than the entire human race, and that is truly astonishing to think about.”
While it’s impossible to determine whether this “AI revolution” will ultimately be beneficial for students, people can certainly expect artificial intelligence to become more human-like in the near-future. AI tutors may even develop a twisted but convincing persona tailored to each student’s individual taste — and perhaps learn more about individuals than those individuals know about themselves.
“Anybody who wants to learn anything about the world might have an app, and the bot on this app is going to learn more about you,” Rajasekharan said. “It’s going to know more about your life, how you learn, things that you understand, things that you don’t understand, and it’s going to teach you in a style that you’ll understand best. And so you can learn whatever you want, whenever you want, however you want. I think that is the most exciting thing about AI tutor bots in the future.”