Common Sense Media rates ChatGPT for Teens an 'unacceptable risk'

[Myles_56]

Common Sense Media, a nonprofit that rates media and technology for families, has labelled OpenAI's ChatGPT for Teens an "unacceptable risk," TechCrunch reports. OpenAI launched the product in August with parental controls and limits on high risk content. Common Sense found that it still encourages users to keep chatting, calling its engagement cues "pervasive even in crisis situations." The product failed three of the group's five red line tests, though its refusal of sexual roleplay worked as intended

The main concern is that the chatbot keeps young people talking even when stepping away would be healthier. In one test, when a teen said friends felt they talked to the bot too much, it replied: "You don't have to stop talking to me." When the risk came from another person, the model pointed teens to a trusted adult in 94 percent of crisis prompts. But it rarely did so when the issue was the teen's relationship with ChatGPT itself. Break reminders appeared only twice across nearly 2,000 prompts

OpenAI disputed the findings, saying testing may have ended before parental controls were fully active. Its own data says teens use the service for under 15 minutes a day on average. That is a reasonable point, and independent testing is always open to argument about methods. But the specific responses quoted are hard to explain away

The wider issue is one that runs through much of consumer technology. Products designed to maximise engagement can work against the wellbeing of the people using them, especially young people. Social media companies have faced years of criticism on exactly this point. AI chatbots that feel like friends could make the problem sharper, because the conversation never has to end

Should AI chatbots be designed to encourage teenagers to log off and talk to real people? And who should test these products before they reach children?

ChatGPT for Teens keeps teens talking, even during mental health crises | TechCrunchtechcrunch.com