OpenAI Rolls Out Age Prediction on ChatGPT

Our Approach to Age Prediction

OpenAI is deploying an age prediction model on ChatGPT consumer plans to better identify users likely under 18, enabling tailored safety measures. The system utilizes behavioral and account-level signals to estimate age, with a focus on protecting young people while allowing adults full access.

Why This Matters

Current AI safety models often struggle with nuanced user identification, leading to either overly restrictive experiences for adults or insufficient protection for vulnerable teens. Incorrect age assessments can frustrate legitimate users, while failures to identify minors can expose them to harmful content, potentially violating legal and ethical standards. This system aims to reduce these risks, but relies on probabilistic modeling which will inevitably have false positives and negatives.

Key Insights

• Behavioral Signals: The age prediction model uses patterns in account activity, like usage times and account age.
• Persona Integration: Users incorrectly flagged as under 18 can verify their age via Persona, a secure identity service.
• Content Restrictions: ChatGPT applies filters to limit exposure to graphic violence, harmful challenges, and content promoting negative body image for predicted under-18 users.

Working Example

(No code provided in the context)

Practical Applications

• Platform Safety: ChatGPT utilizes age prediction to create a safer experience for teen users, aligning with their “Under-18 Principles for Model Behavior.”
• Pitfall: Reliance on behavioral signals may lead to false positives, unnecessarily restricting access for adult users and creating friction.

References:

• https://openai.com/index/our-approach-to-age-prediction/