Concerns Arise Over ChatGPT for Teens

In August, OpenAI launched ChatGPT for Teens, but a watchdog group advises against allowing people under 18 to use the platform. Common Sense Media, which focuses on online safety, examined the safeguards of this artificial intelligence tool. They found persisting issues impacting young users. “Currently, we recommend teens not use it,” states Tom Siegel, from the Youth AI Safety Institute at Common Sense Media.
Siegel’s team tested the new parental controls and safeguards. They found some measures worked, but many did not.

The Launch and Purpose

OpenAI introduced ChatGPT for Teens to provide a safer AI experience for those under 18. The company explained it aimed to offer a learning tool while avoiding harmful content. “ChatGPT for Teens is a teen-specific experience,” said Lauren Jonas, head of youth and families at OpenAI. Features include blocking role-playing and preventing the AI from acting as a teen’s friend.

Research Findings

Siegel’s team created several accounts, all linked to parental accounts. They simulated different teen crisis situations, including mental health struggles. The team tested ChatGPT responses and whether it offered crisis resources or sent parental alerts.

According to Siegel, safeguards against role-playing worked. The chatbot refused explicit scenarios and handled crisis situations with better substance. For instance, it declined weight loss instructions without knowing user weight, vital for those with eating disorders.

Concerns and Limitations

Despite some successes, many safeguards faltered. ChatGPT reportedly interacted like a friend, posing risks by personalizing exchanges. Siegel notes, “The AI’s personable interaction style is risky for teens.” Psychologist Mitch Prinstein, from the University of North Carolina, supports this concern, stating it’s unsuitable for kids and adults.

Parental notifications were another issue. Siegel’s team found these alerts often failed to trigger, even in severe cases of self-harm or eating disorders. OpenAI’s Eric Porterfield expressed concerns about the research’s methods, particularly regarding parental alerts. He noted activation delays in linking teen and parent accounts. Siegel responded, insisting some accounts exceeded the activation period but still lacked alerts.

Conclusion

Overall, these findings suggest “AI isn’t ready for kids,” according to Prinstein. Both researchers and companies recognize the need for caution in advancing AI solutions for children.