AI Learning Machines Soaring, "Education Dilemma" Under Algorithm Hunting

Editor’s Note: As “intelligence” becomes the standard in educational hardware and “algorithms” attempt to take over children’s development, we need to consider not only what technology can do but also what it shouldn’t do. Qilu Evening News · Qilu One Point launches the “Deep Observation of AI Education” series, piercing through market noise to explore the boundaries between technology and humanity.

When technology deeply intervenes in education—one of the most complex human activities—under the guise of “personalization,” the chain reactions it triggers are far-reaching, more so than false advertising and “pseudo-intelligence.” In this era of AI frenzy, what we need is a cool-headed reflection on the essence of AI and education.

At the start of the new school term, learning devices marketed as “score-boosting tools” or “AI private tutors” once again become the focus of family education consumption. From entry-level models costing a few hundred yuan to flagship models nearing ten thousand yuan, AI learning machines have almost become a “standard” for families with primary and secondary school students. However, accompanying them are waves of user complaints and controversies.

Behind prices often exceeding thousands or even ten thousand yuan, the widely promoted features like “precise academic diagnostics” and “one-on-one AI tutoring”—are they truly black technologies capable of tailored teaching, or just “pseudo-intelligent” marketing tricks that harvest parental anxiety? A reporter from Qilu Evening News · Qilu One Point investigated.

Parent Complaints:

Additional paid expansion courses

Are worth less than a few dozen yuan of supplementary books

Ms. Chen, a parent of a sixth-grader living in Guangdong, bought a major brand AI learning device for her child at the start of this semester, spending 4,800 yuan, driven by the marketing claims of “one-on-one AI tutoring” and “precise score improvement across all subjects.” She initially hoped it would ease after-school tutoring pressure, but just one month of use made her feel she had “fallen into a big trap.”

“The core homework correction function is completely useless,” Ms. Chen told reporters. The child’s handwritten math homework often confuses basic digit recognition—mistaking a “7” for a “9,” unfairly deducting points; the child’s essays only receive generic comments like “well-structured, fluent language,” with no targeted suggestions or guidance, making it impossible to improve writing skills; when the child asks about advanced math problems, the AI simply marks them as “beyond scope,” unable to break down solutions step-by-step or even provide basic answer explanations.

Ms. Chen said she has now stopped her child’s use of the device and started over with paper textbooks, gradually cultivating reading habits.

What disappointed her even more was the “precise learning” and vast question bank touted as core features. The manufacturer claimed a “1.6 billion question bank covering all grades and subjects,” but in practice, most questions are old publicly available problems found online. The high school bridging courses are missing chapters and do not match the new school textbooks. The so-called “precise identification of weak knowledge points” is ineffective—her sixth-grade child repeatedly reviews basic questions from fourth grade, missing gaps in learning, wasting time.

Even the extra paid expansion courses are out of sync with the school curriculum and teaching schedule. “I spent nearly 5,000 yuan on the device and courses, but in terms of tutoring accuracy and content usefulness, it’s no match for a few dozen yuan I spent on synchronized supplementary books. Now, the device just gathers dust at home—it’s basically a tax on intelligence,” Ms. Chen said helplessly.

Her experience is not unique. The huge gap between promotional promises and actual experience of AI learning devices has become a widespread consumer pain point across the industry. On the well-known domestic consumer complaint platform 【Download Black Cat Complaint App】, as of March 2026, there have been 10,089 complaints related to “learning machines,” covering almost all major brands such as Xiaoyuan AI, iFlytek, BBK (Vivo), Wenkuxing, Xiaodu, and Readboy, with none exempt.

Complaints span the entire chain—from product promotion, functionality, content services, to after-sales support: issues like frequent errors in AI correction, incorrect tutoring, ineffective pseudo-intelligence, slow content updates, unsynchronized textbooks, and inaccurate question banks; as well as consumer traps like intrusive ads, inducements for minors to pay, and refusal to refund after activation. Many parents accuse brands of exaggerated marketing and false advertising, packaging basic functions as AI black technology to harvest family anxiety.

Investigation:

Claimed to be equipped with self-developed educational large models,

but reveals flaws when solving complex problems

With many parental doubts, a reporter visited mainstream AI learning device stores, dissecting the real quality of these “score-boosting” tools from the underlying AI technology, course content, to practical functions. The findings show that the core competition in the industry revolves around “AI intelligence capabilities” and “question bank resources,” but whether it’s the claimed “self-developed large models” or “exclusive question banks,” industry truths remain hidden.

At a store claiming “21 years dedicated to education,” staff emphasized that their device is powered by a “self-developed nine-chapter educational large model,” and openly stated that “most other brands use general large models like Doubao or DeepSeek,” adding that they “are not certified by the state, so their accuracy will be low.” This confirms the industry reality: many so-called “AI tutoring” products are just third-party general models wrapped simply, without deep optimization for educational scenarios.

“These general models fall apart at middle and high school levels,” an industry insider who wished to remain anonymous revealed. When dealing with complex derivation problems in physics and math, these models often cannot provide clear step-by-step solutions and may even produce logical errors. More critically, most third-party models lack national certification for large models, posing risks in data security and content compliance—what is called “educational intelligence” is actually “pseudo-intelligence” with limited capabilities.

Different brands’ AI learning devices have similar core functions

During testing at a brand store, the upgraded device with a larger screen showed no substantial improvement in core “AI companion learning” functions. When asked a complex fourth-grade math application problem of over 50 words, the machine repeatedly said “I don’t understand the question” and failed to recognize the core information; staff admitted that this AI voice interaction is more for “function search and navigation,” capable of handling simple questions but unable to process complex problems.

Regarding course resources, almost all brands promote “exclusivity”: some claim deep cooperation with publishers to develop synchronized courses, others boast self-developed “school-based question banks,” and some say they can fetch real questions from top local schools in real-time. But when asked to define “exclusivity,” each brand pointed fingers at others, claiming their “exclusive” content is not genuine.

“Those so-called exclusive question banks are just repackaged publicly available exercises,” a salesperson from a certain brand privately complained. The “real-time search of school questions” is essentially scraping open educational resources, with no original content; courses developed with publishers are mostly just digitized versions of textbooks, not customized for learning scenarios.

This homogeneity is especially evident in essay guidance. When asked to write about “spring,” two different brands’ devices generated highly similar ideas and material types, almost identical templates with slight variations.

Staff explained, “Content generated by large models tends to be similar, and question banks are all open resources, so true differentiation is difficult”—meaning parents paying high prices for “exclusive courses” are essentially getting content no different from free online resources.

“AI” becomes the main selling point for many brands

During visits, it was noted that the pricing of AI learning devices heavily relies on “version iterations”—each system upgrade or hardware tweak becomes a reason for price hikes.

For example, a flagship model’s new version only increased the screen size but did not significantly improve core AI functions, yet its price rose by several hundred yuan; other brands’ updates mainly focus on appearance and storage, with minimal enhancement of educational features.

What parents criticize most is that some core functions are deliberately split into paid upgrades. For instance, to address “cheating on homework,” some promote a “answer locking” feature via mobile app, but tests show that if a child rephrases the question, the AI still provides answers directly, making the “anti-cheating” feature ineffective; even more, such functions are often only available in high-priced upgraded models, meaning extra money is needed to use them.

“Updates are more frequent than smartphones, each iteration costs more, but the actual functionality hasn’t improved much,” a parent accompanying their child lamented. The AI learning device has fallen into a cycle of “iteration—price increase—stagnant features,” with brands constantly creating anxiety through “new functions” and “new versions,” yet failing to address the core needs of “score improvement” and “tutoring.” This makes the frequent version upgrades and paid upgrades in the industry seem more like a capital game around the concept of “intelligence.”

Teacher’s advice: AI learning devices are only auxiliary tools

Parents should not be “hands-off”

Beyond obvious complaints about “malfunctioning features and poor experience,” the hidden damage to children’s thinking and autonomous learning abilities raises greater concern among parents. Mr. Zhang, a middle school parent, said in an interview that his main criterion for choosing a device is whether it can truly “guide children to think gradually, rather than just give answers.” Unfortunately, products capable of genuine heuristic guidance rather than rote memorization remain rare.

Such worries have already evolved into real educational crises in many families. Ms. Liu, a primary school parent, shared a chilling experience: “I still feel scared—almost ruined my child’s learning.” She bought a mainstream brand learning device to foster independent homework habits, thinking it would be a helpful learning assistant, but it instead facilitated “fake learning.”

“At first, I thought my child was very self-sufficient; she finished her homework early on weekends, and her exam scores were stable. I never suspected anything was wrong,” Liu said. But one weekend, she left the house with her child still having an unfinished essay. When she returned, the child claimed to have finished it but refused to show it. Curious, Liu checked the child’s notebook and found a well-written essay titled “My Sister,” full of details—yet her child has no sister.

Following the clues, Liu discovered the essay was identical to a model answer stored in the device. She immediately asked her child to rewrite it. The more shocking truth was hidden in the device’s question-scanning records: the child relied heavily on the search function for homework, with some answers in math and almost all answers in Chinese copied directly from solutions. “I was devastated—our child was pretending to learn, but in reality, she was just copying. She seemed to understand but hadn’t truly learned anything.”

Afterward, Liu sold the device at a low price. “Compared to the money lost on the device, I’m more worried that my child has developed the habit of copying answers and not thinking, which could ruin her foundation in learning. Now I have to put everything aside and help her rebuild independent thinking and self-completion of homework.”

On secondhand platforms, many AI learning devices are also being sold.

Such cases are common in primary education. Teacher Zhan Fei (pseudonym), with 12 years of elementary teaching experience, advises parents not to blindly buy various AI learning devices. “Many parents buy devices mainly to ‘free themselves,’ thinking that handing their children over to machines can save tutoring effort and improve scores. That’s a fundamental misconception,” he said. He emphasizes that learning devices are only auxiliary tools and must be used under full parental supervision and proper guidance. Most parents, however, buy with a “hands-off” attitude, turning these tools into shortcuts for laziness and homework avoidance, which can lead to bigger educational risks.

“The core goal at primary school is to cultivate children’s independent thinking and autonomous learning habits—things no AI can replace. If children get used to not thinking and just copying answers, their thinking and expression skills will gradually decline. This hidden, irreversible damage is far more serious than a few poorly done homework.”

On social media, opinions on learning devices are mixed.

Expert opinion:

Don’t let the rapid rise of AI learning devices turn into a new “question-solving frenzy”

Contrasting sharply with the frequent pitfalls faced by parents and the high volume of complaints is the explosive growth of the AI learning device industry.

A report by Zhiyan Consulting in June 2025 shows that in the first quarter of 2025 alone, the total sales volume of AI learning devices across all channels in China reached 1.265 million units, a year-on-year increase of 29.4%. Industry sales are widely expected to surpass 100 billion yuan in the future. However, behind this booming market is a stark set of user behavior data: according to CCIDNet’s “2024-2025 China K12 AI Learning Device Industry Research Report,” 45% of users only utilize the basic photo search function, while the high-level AI features promoted heavily by manufacturers are used by less than 30%.

In a mall in Jinan, multiple AI learning companies compete fiercely.

The underlying logic of this seemingly hot industry has long been exposed by insiders.

“Most AI learning products on the market are still aimed at exam-oriented purposes—just a way to do more questions,” said Ni Minjing, a member of the National Committee of the Chinese People’s Political Consultative Conference and director of the Shanghai Science and Technology Museum, during this year’s two sessions. He described the industry’s rapid growth as a “latent tendency toward involution” and warned it could evolve into a new “question-solving carnival.”

This is not alarmist. The history of Chinese family educational tools is filled with similar stories: from early electronic dictionaries, to reading pen devices, to today’s AI learning machines. Hardware has evolved, but the underlying logic remains unchanged—helping children score higher in exams. The difference is, in the past, it was a manpower war; now, it’s an algorithm war.

But in the face of parents’ most urgent need—“score improvement”—manufacturers find it hard to break out of their comfort zone. Machines capable of quickly pushing vast amounts of practice questions, grading instantly, and providing answers are far more marketable than partners that require patience and encourage thinking. Therefore, under AI, “personalization” largely translates to algorithmic distribution of homogeneous question banks; “intelligent tutoring” is simply moving offline drilling online, dressed up with AI.

Ni Minjing’s question prompts reflection: “No matter how fast or accurate you do questions, you can’t beat a machine or AI itself. If the goal of education is merely to cultivate ‘question-answering experts,’ then the arrival of AI marks the end of that goal.”