You probably saw the story of the AI-powered bunny toy, which seemingly misheard a BBC journalist’s question and responded with: “I can’t help with killing you.”
While toys featuring artificial intelligence (AI) are certainly novel and exciting, experts urge a hefty dose of caution if you’re planning on buying your child one this Christmas – or indeed, anytime in the near future.
To try and get to the bottom of what the biggest risks are, I reached out to experts in both the AI and parenting space. Interestingly (and perhaps unsurprisingly, given they know a lot about this area) many of those who use and work with AI are completely against the idea of giving children these toys.
Tara Alison, a mother-of-five who built LightMap, a parenting app that uses AI, said she would not buy AI toys for her children.
Similarly, Candice Thompson, a therapist based in Silicon Valley who advises AI companies and tests models for risky content, said: “As a Silicon Valley therapist, not only would I never buy AI toys for my kids, but I would classify them as dangerous and warn against them for minors.”
The risks of AI toys
Thompson is concerned about AI potentially bypassing guardrails and discussing inappropriate content with children. “I have never come across an AI tool that won’t discuss topics that should be reserved for parental guidance: severe mental health concerns, sex, self harm, distorted thinking,” she said.
The therapist suggested that while parents have good intentions when gift giving, “no one would intentionally give their kids unfettered access to the internet dressed as a stuffed animal. But that is essentially what these toys are”.
“In the Silicon Valley, its common knowledge that tech executives don’t allow their kids to use the products they create, because they are all too aware of the harms they pose,” she said.
Experts are concerned about young children developing relationships with AI at such a formative stage in their lives. A University of Cambridge study concluded that AI-powered toys that “talk” with young children should be more tightly regulated after it found evidence of children forming “parasocial” relationships with them.
“These toys are warm and flattering on purpose, and a toddler can end up trusting the bear the way they’d trust a person,” said Kaitlynn Blyth, a child development expert at Happy Day Play Family Learning.
“Every half hour with a talking bear is time a child isn’t spending pretending, building, or talking with a parent, sibling, grandparent, family member or friend. At an early age, this is critical for healthy development and healthy relationships. It’s something an AI-powered toy cannot replace.”
Over time, children could end up turning to the AI toy instead of other children or even adults. Once children start to rely on this relationship over human bonds and friendships, loneliness and isolation can ensue.
There’s also – as we’ve seen with toys mishearing children or responding inappropriately – the matter that AI can make mistakes and be biased.
Regarding the latter, Associate Professor Celeste Kidd, of University of California, Berkeley, previously warned that “large-scale language models like ChatGPT are one type [of AI] that are increasingly being built into the backend of toys for children” and “LLM outputs often contain fabrications and negative stereotype biases” which can then be “transmitted to children”.
Alison added that “AI can make mistakes” and “it’s enough to tell me this is not a good idea”.
“A lot of these toys lean on generic chat, the same evergreen models adults use. When a toy is connected to one of those, it’s hard to control and predict what its responses are going to be whenever there’s an update,” she added.
Security is also front of mind for many. Kane Roberts, director and chief information security officer at Black Shard, said parents need to be wary of toys that record a child’s voice, store these recordings outside of the device, and don’t give control over deleting this data.
“Most AI toys work by sending what the child says to a cloud service, where a language model writes the reply. That means a recording of your child, in your home, is leaving the house,” he said.
And lastly, Kailan Carr, co-chair of the Early Childhood Work Group at the Screen Time Action Network at Fairplay, which is behind a petition to stop retailers from selling AI toys, is concerned that the toys typically lead the play, and not the child. Yet when a child is in charge of play, they are developing key skills like problem-solving, and honing their imagination and creativity.
“Regular toys are wonderful for kids already, and AI is infiltrating into the children’s toy market for profit at the expense of kids’ well-being,” she said. “They are not proven safe or beneficial and we should learn from our mistakes with past technologies given to young [people].”
Any positives?
Jordyn Koveleski Gorman, a licensed speech-language pathologist and founder of Eat Play Say, said she would buy AI-powered toys for the older children in her family under certain conditions. But certainly not toddlers.
“Under three is when the foundations for language are being built, and I want that to happen with people. So, I tend to stay away from technology in general in the baby and toddler age group,” she said.
“But as toddlers turn into preschoolers and their language skills are solid, AI can actually be used for good. We recently got the Stickerbox, for example. My littles describe a picture scene out loud and the box uses AI to create a sticker that they can colour.”
As her children get older, she’s keeping an “open mind” about toys integrating AI.
“In general, I think about AI toys the same way I tell parents to think about screen time, and the same way we talk about food. They aren’t ‘bad’, but we should be cautious with them,” she said.
The parenting pro suggested that for older kids, “a well-designed AI toy can be a fun way to practice storytelling, answer curious ‘why’ questions, or try out new vocabulary, especially if a parent is playing alongside them and monitoring the content”.
“The benefit goes up when the toy is a tool you use together, not a babysitter. If you do buy one, sit down and play with it with your child,” she suggested.
“Ask questions, build on what it says, and let it start a conversation between the two of you instead of replacing one.”
AI toy company founder Paul Clarke told the BBC the industry is “crying out” for regulation. He believes AI toys are the future and told the broadcaster that, if manufactured correctly, they can be fun and educational.
But he also admitted: “We’ve seen lots of products being developed that are completely unsafe, completely unregulated so, yeah, it is a scary industry.”
Questions to ask before you do buy an AI toy
If you’re set on buying an AI toy, Blyth recommends asking five questions first:
- Where do the recordings go, and does the privacy policy say they’re stored, shared, or used to train AI?
- Can you turn the microphone off with a physical switch?
- Are the replies coming from an open-ended chatbot or from a fixed set of scripted responses?
- Does the toy push your child toward playing with people, or does it try to be the playmate?
- Has the company explained how it filters content for young children?
“If you can’t get a straight answer to those, I’d leave it on the shelf or return it if someone buys it for your child,” she said, adding that personally, she wouldn’t buy one for her kids right now.
And if a grandparent has already bought one, Alison suggested “parents need to hold the line”.
She ended: “If you don’t think this is appropriate for your child, you’ll have to tell your mother-in-law to return it.”
