What it takes to trust AI in mental healthcare

I recently watched Forbes’ interview on how Spring Health is setting the standard for AI in mental health, and it has stayed on my mind since. I will always advocate for a world where everyone receives truly personalized, continuous mental healthcare. At the same time, we’re facing massive provider shortages and access barriers that impact our most vulnerable community members, and those gaps are not going to close on their own any time soon.

While AI steps in to help fill these gaps and support providers, we need to be honest about the stakes. When it comes to mental health, we are working with people’s most vulnerable moments, in life and in health. A tool that misses a sign of risk or responds without care is not a small technical error, because it can shape whether someone decides to reach out for help again.

This interview, especially the introduction of VERA-MH, spotlights the possibility of rigorous and scalable tools for measuring safety outcomes. VERA-MH, which stands for Validation of Ethical and Responsible AI in Mental Health, is an evaluation framework that Spring Health built together with practicing clinicians, suicide prevention experts, ethicists and AI developers. It looks at whether mental health chatbots can recognize risk, respond with empathy and know when to bring a human into the conversation. I was glad to see it released as an open-source evaluation, so that others in the field can use it, test it and push it further.

Patients can’t afford for ethics to be an afterthought, and this interview centers the need for transparency about what these tools can and can’t do. I am impressed with Spring Health’s ability to match the scale of the issue to the quality of potential solutions, and I hope to see more organizations hold their tools to this kind of standard.