AI tutors are quietly changing the way kids study in the US, and the leading apps are from China

Evan, a high school sophomore from Houston, was stuck on a calculus problem. He opened Answer AI on his iPhone, took a photo of the problem from his Advanced Placement math textbook, and ran it in the homework app. Within a few seconds, Answer AI generated a response along with a step-by-step process to resolve the issue.

A year ago, Evan was searching for long YouTube videos in hopes of meeting the challenges of his assignment. He also had a private tutor, who cost $60 an hour. Now, the arrival of artificial intelligence robots poses a threat to long-established tutoring franchises like Kumon, the 66-year-old Japanese giant that has 1,500 locations and nearly 290,000 students across the United States.

“The hourly cost of the tutor is about the same as the entire year of Answer AI subscription,” Evan told me. “So I stopped doing a lot of things. [in-person] tutorships.”

Answer AI is among a handful of popular apps that are taking advantage of the arrival of ChatGPT and other big language models to help students with everything from writing history papers to solving physics problems. Of the top 20 educational apps on the US App Store, five are AI agents that help students with their schoolwork, including Answer AI, according to Data.ai data from May 21.

There is a perennial debate about the role AI should play in education. The advantages of AI tutors are obvious: they make access to after-school tutoring much more equitable. $60-an-hour tutoring in Houston is already much more affordable than services in more affluent and academically cutthroat regions, like the Bay Area, which can be three times more expensive, Answer AI founder Ric Zhou told me.

Zhou, a serial entrepreneur, also suggested that AI allows for more personalized teaching, something difficult to achieve in a classroom of 20 students. Chatbot teachers, who can remember a student’s learning habits and never get grumpy when answering questions, can replace the private coaches that wealthy families hire. Myhanh, a Houston-based high school student, said her math scores have improved from 85 to 95 in the six months since she used generative AI to study.

For now, AI tutors are limited primarily to text-based interactions, but soon they will be able to literally speak to students in ways that optimize each student’s learning style, whether it be a more empathetic, humorous, or creative style. OpenAI’s GPT-4o has already demonstrated that an AI assistant that can generate voice responses in a variety of emotive styles is within our reach.

When AI doesn’t help you learn

The vision of equitable learning powered by AI is yet to be fully realized. Like other apps that forward API calls to LLMs, AI tutors suffer from hallucinations and can spit out incorrect answers. Answer AI attempts to improve its accuracy through recall augmented generation (RAG), a method that tunes an LLM with some domain knowledge, in this case, a sea of ​​problem sets. But it still makes more mistakes than state-of-the-art homework apps that match users’ queries to an existing library of practice problems, because these apps don’t try to answer questions they don’t already know.

Some students are aware of the limitations of AI. Evan often compares Answer AI results with ChatGPT, while Myhanh uses Answer AI in an after-school study group to exchange ideas with her classmates. But Evan and Myhanh are the type of autonomous learners who are most likely to use AI as a learning aid, while some of their peers can comfortably delegate AI to do their tasks without learning anything.

Respond to AI Screenshot Feature via Your Chrome Extension / Image: TechCrunch
Image credits: Reply AI

For now, educators aren’t sure what to make of AI. Several US public school districts have banned access to ChatGPT on school devices, but enforcing a complete ban on generative AI becomes a challenge as soon as students leave school grounds.

The reality is that it is impossible for teachers and parents to stop children from using AI to study, so it may be more effective to educate children about AI’s role as an imperfect assistant that sometimes makes mistakes instead. to ban it completely. While it’s difficult to discern whether a student has learned to solve a math problem by rote based on the answer she writes, AI is at least good at detecting AI-generated essays. That makes it harder for students to cheat on humanities assignments that require more original thinking and expression.

Chinese dominance

The two most popular AI helpers in the US, as of May, are both Chinese-owned. The year-old Question AI is the brainchild of the founders of Zuoyebang, a popular Chinese homework app that has raised around $3 billion in capital over the past decade. Gauth, on the other hand, was launched by TikTok parent ByteDance in 2019. Since its creation, Question AI has been downloaded six million times on the Apple App Store and Google Play Store in the US, while its Rival Gauth has accumulated twice as many installations. since its launch, according to data provided by market research firm SensorTower. (Both are published in the US by Singapore entities, a common tactic as Chinese technology receives increasing scrutiny from the West.)

ChatGPT solving an arithmetic sequence problem / Image: TechCrunch

The success of Chinese homework apps is the result of their concerted effort to target the US market in recent years. In 2021, China imposed rules to clamp down on its burgeoning private tutoring sector focused on the country’s public school curriculum. Since then, many service providers, including traditional tutoring centers and online study apps, have turned to foreign users. Unsurprisingly, the United States is its most coveted international market due to its large size.

The fact that tutoring apps are likely to use similar fundamental AI technologies has leveled the playing field for foreign players, who can overcome language and cultural barriers by calling on AI to study user behavior. As Eugene Wei wrote in his canonical analysis of TikTok’s global success:[A] “A significantly sensitive and accurate machine learning algorithm can pierce the veil of cultural ignorance.”

Reliance on the same LLM pool also makes it difficult for these study apps to differentiate themselves solely by the quality of their answers. Some of the legacy players, such as Zuoyebang and PhotoMath, can use a combination of generative AI and search their extensive problem set libraries to improve accuracy. Newcomers will need to find alternative ways to differentiate themselves, such as improving user customization features.

“An AI agent needs to proactively interact with students and tailor its responses to individual learning needs,” Zhou said. “A plain language model is not an out-of-the-box AI agent, so we try to differentiate ourselves by fine-tuning our AI to teach more effectively. For example, our AI bot would invite students to ask follow-up questions after presenting an answer, encouraging deeper learning rather than simply letting them copy the result.”

Leave a Comment