Evan, a high school sophomore from Houston, was grappling with a calculus problem. He used Answer AI on his iPhone, took a snapshot of the problem from his Advanced Placement math textbook, and fed it into the homework app. In mere seconds, Answer AI provided a solution along with a detailed, step-by-step explanation.
A year ago, Evan would be sifting through lengthy YouTube videos, hoping to resolve his homework issues. He also had a private tutor, who charged $60 per hour. Now, AI bots are challenging established tutoring franchises like Kumon, the 66-year-old Japanese giant with 1,500 locations and nearly 290,000 students across the U.S.
“The tutor’s hourly cost is about the same as Answer AI’s whole year of subscription,” Evan shared. “So I stopped doing a lot of [in-person] tutoring.”
Answer AI is among a few popular apps utilizing advancements in ChatGPT and other large language models to aid students with tasks ranging from writing history papers to solving physics problems. According to data from Data.ai on May 21, five out of the top 20 education apps in the U.S. App Store are AI agents assisting students with their schoolwork, including Answer AI.
Debate continues over the role of AI in education. AI tutors clearly offer advantages: they make after-school tutoring more accessible. The $60-per-hour tutoring in Houston is already cheaper than in more affluent, competitive regions like the Bay Area, where rates can be triple, remarks Answer AI’s founder Ric Zhou.
Zhou, a serial entrepreneur, also noted that AI provides more personalized teaching, difficult to achieve in classrooms with 20 students. Chatbot teachers, which can memorize a student’s learning habits and are always patient, can replace private coaches that wealthy families employ. Myhanh, a high school junior from Houston, saw her math grades rise from 85 to 95 after six months of using generative AI for study.
Currently, AI tutors mainly interact through text, but soon, they will be able to speak to students in a way tailored to individual learning styles — whether empathetic, humorous, or creative. OpenAI’s GPT-4o has already shown that an AI assistant capable of generating voice responses with varying emotional tones is within reach.
When AI doesn’t help you learn
The vision of equitable, AI-driven learning isn’t fully achieved yet. Like other apps using API calls to LLMs, AI tutors can hallucinate and give incorrect answers. Answer AI uses Retrieval Augmented Generation (RAG) to improve accuracy by fine-tuning an LLM with specific domain knowledge — in this case, numerous problem sets. However, it still makes more errors than older homework apps that only respond with answers from an existing library of practice problems.
Some students recognize AI’s limitations. Evan frequently cross-references results from Answer AI with ChatGPT, while Myhanh uses Answer AI in an after-school study group to exchange ideas with peers. However, Evan and Myhanh are inherently motivated students who use AI as a supplemental learning tool. Some of their classmates may rely on AI to do their homework without understanding the material.
Educators are currently uncertain about how to handle AI. Several public school districts in the U.S. have banned access to ChatGPT on school devices, but completely banning generative AI becomes challenging once students leave school.
The reality is that it’s impossible for teachers and parents to stop kids from using AI for studying. A more effective approach might be to teach students about AI’s role as an imperfect assistant that sometimes makes errors instead of banning it entirely. While discerning whether a student learned to solve a math problem from their answer is difficult, AI is good at detecting AI-generated essays, making it harder to cheat on assignments requiring original thought and expression.
Chinese dominance
As of May, the two most popular AI tutors in the U.S. are Chinese-owned. Question AI, launched by the same minds behind Zuoyebang, a prominent Chinese homework app that has raised about $3 billion in equity over the past decade, is one such tutor. Gauth, launched by ByteDance in 2019, is another. Since its inception, Question AI has been downloaded six million times on Apple’s App Store and Google Play Store in the U.S., while Gauth has achieved double the downloads since its launch, as per market research firm SensorTower. (Both are published in the U.S. by Singaporean entities, a common strategy as Chinese tech faces increasing scrutiny from the West.)
The success of Chinese homework apps stems from their focused efforts to penetrate the American market in recent years. In 2021, China imposed regulations curtailing its booming private tutoring sector focused on the national school curriculum. Consequently, many service providers, including physical tutoring centers and online study apps, shifted their focus to overseas users, with the U.S. as a prime market due to its extensive size.
Given that tutoring apps likely use similar foundational AI technologies, the playing field has leveled for foreign competitors, who can overcome language and cultural differences by leveraging AI to analyze user behavior. As Eugene Wei wrote in his seminal analysis of TikTok’s global success, “[A] machine learning algorithm significantly responsive and accurate can pierce the veil of cultural ignorance.”
Using the same LLMs makes it difficult for study apps to stand out based purely on the quality of their answers. Established players like Zuoyebang and PhotoMath can combine generative AI with search functionalities in their extensive problem set libraries to enhance accuracy. New players must explore different ways to differentiate themselves, such as improving user personalization features.
“An AI agent needs to proactively engage with students and tailor its answers to individual learning needs,” said Zhou. “A raw language model isn’t a ready-to-use AI agent, so we try to differentiate by fine-tuning our AI to teach more effectively. For instance, our AI bot would invite students to ask follow-up questions after presenting an answer, encouraging deeper learning rather than just letting them copy the result.”
