AI, Risk, and Contemplative Science (ARC) Lab

We develop methods for evaluating the validity and reliability of text-based psychological assessments. Our research has identified which symptoms indicate the highest risk of suicide from crisis hotline conversations and social media posts. We created an open-source toolbox that uses LLMs to measure and validate symptoms like these.

We examine how voice changes across mental health conditions. Our work focuses on detecting speech changes that occur as hospitalized and recently discharged individuals become more suicidal, with the goal of improving risk assessments.

We study how to regulate and evaluate AI chatbots when users are suicidal. Our current work explores whether AI chatbot users are able to recognize when suicidal and delusional concerns should be challenged rather than validated during AI chatbot interactions.

The Child Mind Institute’s Mirror journaling app, which serves as a privacy-preserving phenotyping tool to assess suicide risk, circadian rhythms, energy, arousal, and mood, has been shown to reduce anxiety. Our ongoing research explores the similarities and differences between meditation and journaling.

Mindfulness and other psychological factors help explain changes in mental health following naturalistic psychedelic use. We are currently testing whether mindfulness meditation can reduce stress in low-income populations. We are also exploring the emergence of insight and its effects after psychedelic use, as well as strategies for reducing harms associated with its use.

Chatbot users often turn to AI for support with existential and spiritual concerns related to meaning, purpose, and death. We are investigating the potential and limits of AI in providing guidance in these domains.