As concern for AI’s impact on young people continues to mount, OpenAI has introduced an “age prediction” feature into ChatGPT that is designed to help identify minors and put sensible content constraints on their conversations.
OpenAI has been heavily criticized in recent years for the impacts that ChatGPT can have on children. A number of teen suicides have been linked to the chatbot and, like other AI vendors, OpenAI has also been criticized for allowing ChatGPT to discuss sexual topics with young users. Last April, the company was forced to address a bug that had allowed its chatbot to generate erotica for users who were under the age of 18.
The company has already been working on its underage user problem for some time, and its new “age prediction” feature merely adds to protections already in place. The new feature leverages an AI algorithm that assesses user accounts for particular “behavioral and account-level signals,” in an effort to identify young users, OpenAI said in a blog post Tuesday.
Those “signals” include things like the user’s stated age, the length of time an account has existed, and the times of day that the account is usually active, the company said. The company already has content filters designed to weed out discussions of sex, violence, and other potentially problematic topics for users that are under age 18. If the age prediction mechanism identifies an account as under 18, those filters are automatically applied.
If a user is mistakenly designated as underage, there is a way for them to re-establish their “adult” account. They can submit a selfie to OpenAI’s ID verification partner Persona, OpenAI says.
