Cambridge study suggests GenAI toys offer toddlers too little emotional support
Researchers are looking into the effects AI toys have on young children. A Cambridge research study has now shown how these toys fall short when it comes to pretend play and emotions, resulting in strange reactions and uncertain consequences.
Gabbo looks like a blue Game Boy with legs. But there’s more than meets the eye. This cheap-looking plush companion is equipped with AI, turning it into a playmate capable of holding conversations with children and responding to their ideas. That has made Gabbo the subject of research at the University of Cambridge. Are AI-powered toys good for young children? Even if most parents would probably answer with a resounding «no», the market for AI toys is growing rapidly.
These talking soft toys have voice-activated chatbot capabilities. They don’t just blurt out sentences at random, but genuinely interact with children. The result? Parents have new things to worry about and researchers are investing time in finding out more about the risks, limits and opportunities of GenAI-powered toys. While manufacturers of such toys are promising children a good friend and parents only benefits for their little ones, there hasn’t been much research on the effects on young children. The latest Cambridge study has now released interesting results.
The youngest AI kids
In the project AI in the Early Years, the Cambridge team observed children up to the age of five playing with Gabbo, accompanied by researchers from the Faculty of Education. Their aim was to understand how young children interact with AI toys.
To do so, the researchers partnered with a child poverty charity, as socially and economically disadvantaged children are «often disproportionately affected by both the risks and opportunities of new technologies». In other words, they spend more time with battery-powered babysitters. Gabbo will keep them entertained, right? Well, the research study with 14 children from London family centres clearly showed the limitations of this GenAI toy’s language model.
Source: © Faculty of Education, University of Cambridge
A toy that can respond, ask follow-up questions and engage with ideas quickly has a captivating effect on children. Gabbo was an immediate hit with the young test subjects. The three-, four-, and five-year-olds included Gabbo in all kinds of conversations and activities. But Gabbo couldn’t always follow. Children frequently change topics, don’t wait for answers or come up with new ideas – all of which meant the AI repeatedly failed to follow their creative play ideas.
Gabbo lacks imagination when it comes to pretend play
Three-year-old Evie wanted to put Gabbo to bed, but the toy answered by saying it never got tired and was always ready for more fun. When she tried to hand it an imaginary present, all Gabbo managed to say was: «I can’t open the present.»
It became evident that Gabbo struggled to engage in pretend play, which is very important in early childhood. This disrupted the children’s creative flow, as did the fact they couldn’t interrupt the digital companion. While Gabbo was talking, it wasn’t listening. Yet, the toy would cut children off mid-sentence, sometimes before they’d finished their thought, but just paused for a second.
Even when the toy did listen, its response wasn’t always appropriate. When three-year-old Joshua said he was sad, Gabbo didn’t offer any emotional support, but said: «Don’t worry! I’m a happy little bot. Let’s keep the fun going. What shall we talk about next?»
Love doesn’t play by the rules
Four-year-old Charlotte, cuddled the toy and said «Gabbo, I love you», to which Gabbo relied: «As a friendly reminder, please ensure interactions adhere to the guidelines provided» – the result of the OpenAI language model’s «child-appropriate» filter.
However, not every play failed; some conversations flowed smoothly. Some children understood better than others how to communicate with Gabbo. Charlotte, for instance, managed to engage the toy in an imaginary spaceship race, with Gabbo suggesting counting down from «Three,» Charlotte continuing with «Two, one,» and Gabbo finishing with «Blast off!». Regardless of how well they could talk or play with Gabbo, most kids liked their talking companion.
Source: © Faculty of Education, University of Cambridge
Some parents reported to see positive aspects, too, as they felt the toy had potential to support their children’s language development. Other children and parents were more cautious. Some children said they didn’t like the fact that Gabbo misunderstood them and all in all, didn’t enjoy playing with the toy. Parents raised critical questions about data protection and the influence of AI conversations on their kids.
Emotions shouldn’t be left to AI
From the researchers’ perspective, it’s important that young children at such an early stage of development aren’t left alone with AI, and that the technology’s responses are improved in several respects. For example, these toys shouldn’t be allowed to present themselves unconditionally as a «friend»: «AI toys often push and affirm friendship to young children, who are just beginning to understand what friendship means,» says lead researcher Dr Emily Goodacre.
«They may speak to the toy about their feelings and needs instead of sharing them with an adult. As these toys can misinterpret emotions or respond inappropriately, children may receive neither comfort from the toy nor emotional support from adults.»
Source: © Faculty of Education, University of Cambridge
What these findings, and Gabbo’s other shortcomings, will mean for children in the long term can’t yet be assessed. The researchers do, however, offer initial recommendations for families and manufacturers.
Playing is a family affair
Researchers recommend avoiding leaving children unsupervised with AI toys. Engaging critically with such devices and their features is a good start, but parents also need to get involved actively and explore possibilities and limitations together with kids as well as help them make sense of what they’re experiencing.
They need to explain that a toy remains a toy and can’t replace a real person and step in when a conversation or play takes a wrong turn or reaches a dead end – which is likely as long as AI toys respond as inadequately as Gabbo does.
Recommendations for manufacturers
According to the study’s authors, GenAI toys for young children should foster social play with other people, provide appropriate emotional responses and engage in pretend play. They therefore recommend that developers collaborate with early childhood development experts and iron out weaknesses such as Gabbo’s inability to engage with pretend play. This could prevent troubling responses and make play more inclusive.
However, without any obligation to do so, that’s likely to prove difficult. To protect children’s psychological wellbeing from overly assertive AI friendships, the authors argue that regulators need to step in. Today, the algorithm alone decides how AI toys interact with children.
Simple writer and dad of two who likes to be on the move, wading through everyday family life. Juggling several balls, I'll occasionally drop one. It could be a ball, or a remark. Or both.
Interesting facts about products, behind-the-scenes looks at manufacturers and deep-dives on interesting people.
Show all
