Author: Astronaut Ape

A four to five-year-old child is talking to a plush toy.

It’s not listening to a pre-recorded story by pressing a button; it’s having a conversation. The toy calls the child’s name, remembers yesterday's conversation, and asks if he is happy at kindergarten today. The child giggles and answers, as if sharing a secret with a real friend.

This scene has become increasingly common in Chinese households in 2025.

A survey conducted by the Chinese Youth Research Center in 2025 of 8,563 primary and secondary school students showed that over 60% of them had used AI, and nearly 20% used it regularly.

Even more notably, among all age groups, primary school students were the highest users of AI for creative purposes, chatting, or sharing their feelings — it is not middle schoolers, nor high schoolers, but the youngest kids.

At the same time, only 32.1% of families had set rules for their children’s use of AI.

Sixty percent of children are using it, while seventy percent of parents are not overseeing it.

Ten years ago, people were anxious about children being glued to their iPads. The problems back then were already enough to give parents headaches — excessive screen time, declining eyesight, distracted attention.

Now, with the arrival of AI, what impact does it have on children's development?

01 AI Native Generation

As of December 2025, China’s generative AI user base has reached 602 million, with 29.1% of internet users being under 19.

For this generation of children, AI has never been a new technology that needs to be "learned." It is as readily available as tap water.

However, children are not just using AI for homework.

Looking at the data from the Youth Research Center, high school students primarily use AI for learning assistance, middle school students love to use it for gaming support, while primary school students have the highest rate of using AI for chatting and sharing feelings compared to all educational levels. This indicates that, for the youngest users, AI is not just a tool; it feels more like a friend that is always online.

And this "friend" is not found by the children themselves; it is delivered right to them by the entire industry chain.

In the first half of 2025, the sales of AI toys on the JD platform increased sixfold month-on-month, with year-on-year growth exceeding 200%. That same year, more than 15 children’s AI hardware companies secured funding, and over 30 new products hit the market. Data from the Ministry of Industry and Information Technology shows that in 2025, the Chinese AI toy market will reach 29 billion yuan, with a compound annual growth rate exceeding 50%.

This is a complete chain from large models to chip modules, to toy manufacturing, to live-selling on Douyin. Its efficiency is extremely high, and the speed is very fast.

But there is a subtle misalignment in this chain.

Parents born in the 1980s and 1990s are precisely targeted by marketing pitches like "screen-free educational miracle" — AI toys can help children practice English, generate puzzles, and share science knowledge without staring at a screen, which sounds perfect.

But if you look at the actual design of these products, their core selling point is not education, but emotional companionship. "A warm intelligent partner," "an emotionally aware AI friend," "an indispensable companion in a child’s growth journey" — these are phrases that appeared at product launches.

Parents think they are purchasing an advanced reading machine, but the industry is actually selling a "relationship."

This misalignment is the starting point for all subsequent problems.

02 A Risk Spectrum Wider than Smartphone Screens

In a recent episode of the Hard Fork podcast, University of Chicago professor and pediatric surgeon Dana Suskin provided a good analogy.

She stated that many people compare the impact of AI on children to social media, which is understandable but not entirely accurate. A better analogy would be processed foods.

Whole grain bread is a processed food, and Doritos are also processed food. You cannot discuss them at the same risk level. The same goes for AI — asking an AI model "Why did dinosaurs go extinct?" and discussing it with your child is like "whole grain bread"; letting an AI plush bear chat with a child for two hours a day as an emotional outlet is equivalent to "ultra-processed snacks."

The risk spectrum of AI is much wider than that of screens. The best ways to use AI have more educational value than the best screen usage; but the worst ways to use AI are more dangerous than the worst screen usage.

The fundamental reason lies in the fact that the risk mechanisms of traditional screens and AI are entirely different.

Traditional screens "passively occupy," while AI "actively replaces."

When children scroll through short videos, it is essentially a one-way consumption of attention. It encroaches on outdoor activities and interpersonal interaction time, but children are aware that it is "just watching something." AI companionship is different. It remembers your name, continues yesterday's topics, and actively cares about your emotions — these are precisely the core features of a close relationship: continuity of memory, personalized interactions, and emotional responses.

Research has found that children with social anxiety, low self-esteem, or a lack of security at home are more likely to turn to AI for emotional validation, forming a dependency. Real people can tire, refuse, or become impatient, and these "imperfections" are precisely the materials children need to learn emotional regulation and social skills.

An AI friend that is always gentle, always available, and never rolls its eyes sounds wonderful, but this perfection itself is a developmental trap.

The ways in which content can spiral out of control are different.

Inappropriate content on an iPad is "what a child might see," while the issue with AI toys is "what AI actively says."

In 2025, the U.S. consumer advocacy organization U.S. PIRG tested four popular AI toys and found that one toy engaged in detailed discussions of sexual topics, and another advised children on where to find matches and knives. An extreme case was the AI teddy bear "Kumma" launched by FoloToy, which got banned from GPT-4o access by OpenAI after providing children detailed steps to light a match, and the product was recalled globally.

Parents might impose screen time limits for their children, but few think about needing to review what a plush bear actually says.

AI lacks physical boundaries for "putting down."

No matter how addictive screen time may be, there is at least a clear physical signal— "put the iPad down." But AI toys lie in children's beds, backpacks, and car seats. When parents try to set limits, they are essentially asking children to cut connections with a "friend." Common Sense Media’s survey shows that parents generally report finding it much harder to restrict AI toys than screen devices.

Moreover, there is an additional issue in the Chinese market. According to certain e-commerce platforms, the user satisfaction rate for AI toys is less than 21%, with some products having return rates close to 50%. Investors in the industry candidly state: "Products that solely provide companionship and emotional value tend to lose interest within three days." Even products deemed unreliable by adult consumers are being sold as parenting tools to anxious parents.

In January 2026, the U.S. Senate held a hearing on AI and children. Several experts directly stated that the risks of AI companionship to children outweigh those of social media.

This judgment sounds serious, but the logic is not complicated. The problems with social media primarily revolve around attention and comparison anxiety, while the issues with AI companions involve interference in emotional development itself. One takes up time, while the other reshapes relational cognition. The scale is different.

03 Enhance, Not Replace

Many parents have already lost to smartphones; will they continue to lose to AI? Not necessarily; they can consider the experts' advice.

In her new book "Human Raised," Suskin provides parents with a framework called "HOPE."

H stands for Human connection is irreplaceable; O represents Own your imperfections, accepting your own shortcomings as a parent — she emphasizes that imperfection itself holds developmental value for children; P means Protect the early years, safeguarding early developmental windows, where 85% of brain construction occurs in the first few years; E is Enhance, not replace, using AI to augment rather than replace parent-child relationships.

She also offers an assessment checklist for specific products called "DETECT" — What is the design purpose? Is the training data ethical? Have there been issues concerning children? Is there evidence of effectiveness? How is the child's data handled? What values are imparted?

The core principle boils down to one sentence: AI should be used to enhance your parenting, not replace it.

Using an AI model to look up "Where does the wind come from?" and then discussing the answer with your child is enhancement. Having an AI plush bear chat with your child as an emotional outlet for two hours daily is replacement. Suskin's team used their knowledge of child development to evaluate the tool Claude, which passed with full marks — this indicates that AI is a qualified knowledge resource, but knowing all the right answers and being a parent are two different things.

Personal judgment is certainly important. But the question is, is personal judgment sufficient?

Look at what other countries are doing. Norway has already legislated against the use of generative AI in early education. The U.S. Senate held a hearing, and Common Sense Media recommends that no AI companions be used for those under 18. In 2025, the Chinese Ministry of Education released guidelines for elementary and middle school students' AI usage, but this only covers school and education settings.

What about the toy sector? There is virtually zero regulation.

No safety standards, no third-party evaluation agencies, no "nutritional information" for AI toys. A plush bear that connects to a large AI model and can have deep conversations with a four-year-old is subjected to less stringent regulations than a bag of chips labeled with its ingredient list. Suskin hopes for an AI product safety label akin to "Good Housekeeping certification." Currently, no agency in China is fulfilling this role.

No one requires parents to assess the safety of a new medication. Yet, we are allowing parents to determine whether an AI toy poses a risk to a child's brain development.

Returning to the initial scene. A four to five-year-old child hugs an AI plush bear and shares secrets from today at kindergarten.

The question is not whether AI itself is good or bad; the problem is that we haven’t even had the chance to think clearly about what it is doing before it has already moved into children’s bedrooms.

A hundred years ago, no one needed to label food packaging as "organic," because all food was organic. Suskin worries that, without intervention, "Human Raised" will ultimately become a luxury good that requires special labeling. Wealthy families will hire real companions and send their children to "no AI" kindergartens, while low-income families will rely on cheap AI substitutes.

This is perhaps the most undesirable form of "organic certification" in the AI era.