From Anthropic · · 1m 30s
Anthropic funds independent AI wellbeing research
A $5 million grant program aims to build better tools for measuring how AI affects users' mental health and emotional wellbeing.
In brief
A $5 million grant program aims to build better tools for measuring how AI affects users' mental health and emotional wellbeing. Independent researchers will create open-source evaluations to help the AI industry understand and measure the wellbeing impact of conversational AI systems. Originally reported by Anthropic.
Five million for wellbeing research
Anthropic is launching a $5 million grant program to fund independent research into how AI impacts users' wellbeing, with direct funding, model access, and technical support for grantees.
AI as emotional support
AI systems now serve as conversational partners and sources of emotional support during difficult times, but the industry lacks clear standards for how models should behave in these sensitive interactions.
Wellbeing is hard to measure
Evaluating wellbeing requires understanding context across long conversations. A response that's helpful for one user might be harmful for another, like diet advice given to someone with a history of disordered eating.
Anthropic's current safeguards
Anthropic develops safeguards to identify sensitive conversations and publishes research on how people interact with Claude to inform safety measures and protect user wellbeing.
What makes a strong evaluation
Rigorous wellbeing evaluations should clearly define what they measure, involve clinical experts, test both overcaution and harm, reflect real multi-turn conversations, and validate graders against subject-matter experts.
How to apply
The program invites clinicians, psychologists, and methodologists to build open-source evaluations. Applications close September 21, with full proposal notifications by October 5.
The key point
Independent researchers will create open-source evaluations to help the AI industry understand and measure the wellbeing impact of conversational AI systems.
