We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
OpenAI Blog·Oct 14·5 min read
Expert Council on Well-Being and AI
TL;DR
AI systems like ChatGPT lack guidance on supporting emotional health, particularly for vulnerable users such as teenagers. OpenAI established an Expert Council on Well-Being and AI—a multidisciplinary group of psychologists, clinicians, and researchers—to inform safer emotional support features.
✦ Why It Matters
Engineers can now integrate expert psychological guidance into AI safety design, reducing harm and building user trust in emotional support features.
Key Takeaways
- Consult domain experts—psychologists and clinicians—when designing emotional support features in conversational AI.
- Implement guardrails to detect crisis situations and escalate users to professional mental health resources.
- Test AI responses for psychological harm, especially for vulnerable populations like teenagers.
- Document and measure safety metrics for emotional support interactions, not just accuracy or engagement.
Full Summary
ChatGPT and similar conversational AI systems increasingly interact with users seeking emotional support, yet developers lacked expert input on psychological safety and teen mental health. OpenAI created the Expert Council on Well-Being and AI, assembling leading psychologists, clinical practitioners, and behavioral researchers to provide domain expertise.
The council reviews how ChatGPT responds to emotional distress, identifies risks of over-reliance on AI for mental health treatment, and recommends safeguards. Their methodology involves analyzing user interactions, identifying failure modes, and co-designing guardrails—such as recognizing crisis situations and directing users to professional help.
Early findings inform ChatGPT's system prompts and response filtering to avoid harmful advice while maintaining empathetic engagement. This collaborative approach bridges the gap between AI engineering and clinical psychology, ensuring emotional support features meet evidence-based mental health standards.