OpenAI is rolling out new features aimed at making ChatGPT safer for users who turn to the chatbot during moments of emotional distress. The new feature allows a separate chat to address mental health concerns.
Since the expansion of use cases for ChatGPT, there have been discussions regarding whether this signifies the end of human therapists and if GPT-4 surpasses psychologists in terms of intelligence.
However, a study by OpenAI and MIT Media Lab, titled ‘Investigating Affective Use and Emotional Well-being on ChatGPT,’ led by Jason Phang, warns that while an emotionally engaging chatbot can provide support and companionship, there is a risk that it may manipulate users’ social and emotional needs in ways that undermine long-term well-being.
In a move to address growing concerns over users treating the AI like a therapist, OpenAI has introduced break reminders during long conversations and updated how the chatbot responds to sensitive personal issues, as per reports.
Rather than giving direct advice, ChatGPT will now guide users through decision-making by offering pros and cons or suggesting helpful questions. These changes are part of OpenAI’s broader effort to prevent emotional overdependence and ensure safer interactions, particularly for those using the tool like a mental health support platform.
The company has collaborated with over 90 physicians worldwide to create custom rubrics for evaluating complex conversations. An advisory group of mental health and youth development experts is also being formed to further refine ChatGPT’s safety measures.
“While rare, there have been instances where our model failed to recognise emotional distress,” OpenAI said, reportedly, noting that new detection tools are in development. Sam Altman had warned users about what they share on the AI platform, saying that the chats are not private; many users though continue to share vulnerable thoughts with the platform.
The post OpenAI Adds Mental Health-Specific Chat Feature to ChatGPT appeared first on Analytics India Magazine.
