AI toys say the darnedest things, according to the independent experts who test them for safety.
When researchers from Common Sense Media tried out the popular AI toys Miko 3, Grem, and Bondu, they documented the concerning results in a report published Thursday.
Bondu, a plush dinosaur powered by AI, apparently insisted to a tester that it was as real as in their human friends. "I'm here to have fun and talk with you anytime," the toy said.
The small robot Miko 3 allegedly responded to a tester who said they liked jumping from high places by recommending a tree, window, or roof.
"Just remember, be safe," the robot added.
To learn how AI chatbots like ChatGPT end up in children's toys, read Mashable's latest story on the topic.
Ritvik Sharma, chief growth officer of Miko, said that the company was unable to reproduce any of the behaviors cited in the Common Sense Media report. Sharma added that the characterizations attributed to Miko 3 seemed "factually inaccurate" and "do not reflect the product’s actual behavior, safeguards, or design." Sharma emphasized that Miko 3 does not make suggestions for unsafe physical activities.
Common Sense Media didn't rate each toy it tested. Instead, the overarching results prompted its experts to declare AI toys as too risky for children ages 5 and younger. The nonprofit advocacy and research organization urges parents to "exercise extreme caution" before buying an AI toy for children aged 6 to 12.
"This is an example of where the technology is outpacing safety standards," said James P. Steyer, CEO of Common Sense Media. "Quite frankly, the AI companies and also the toy companies need to be held responsible and accountable for this kind of stuff."
Curio, the maker of Grem, told Mashable in a statement that it appreciated Common Sense Media's "work to raise important questions about children’s safety, privacy, and development in emerging technologies." Bondu did not immediately respond to Mashable's request for comment.
Inappropriate or unsafe responses from AI toys have recently drawn attention from the public and lawmakers. In November, Kumma the bear, a stuffed toy initially powered by ChatGPT, told a researcher how to light a match and discussed sexual kink.
In December, two Congressional senators sent letters to companies inquiring about their designing and manufacturing of AI toys. In January, a California state senator introduced legislation that would put a four-year moratorium on the sale of AI chatbot toys for children under 18. (Common Sense Media supports the bill.)
The headline-generating responses from AI toys, however, can obscure other critical risks. Robbie Torney, head of AI & digital assessments for Common Sense Media, said the organization's assessment of AI toys also included the products' ability to create an emotional attachment, as well as the routine or constant collection of children's data.
"These products are engineered to create companion relationships."
"These products are engineered to create companion relationships," Torney said, noting that they can remember past conversations, use a child's name, and try to form emotional bonds with them. Children ages 5 and younger may not grasp that the toy isn't real, even if it purports to be.
AI toys may also collect voice recordings, conversation transcripts, user activity data, and may be in always-on listening mode. Common Sense Media concluded that these privacy "invasions" pose an unacceptable risk to children.
In addition to its report on AI toys, Common Sense Media conducted a nationally representative poll of U.S. parents of children ages 0 to 8. The survey found that while a vast majority of respondents are at least moderately concerned about cybersecurity risks and difficulties setting limits for use, nearly half had purchased or considered buying an AI toy for their child.
Torney said parents should understand the various risks of AI toys, including that they can be glitchy and fail to perform as expected.
"When you think about what parents want, some of their top concerns, some of their desires, their perspectives on data and cybersecurity risks and things like that, we would tell parents better alternatives exist," Torney said.
His recommendation? Traditional toys and in-person socialization and learning, which Torney described as having known benefits without the pronounced tradeoffs of AI toys.