About

I am a computational social scientist in psychology, an Assistant Professor in Psychology, and the Shriram Faculty Fellow at the Institute for Human-Centered Artificial Intelligence. At Stanford, I co-direct the CREATE Center for AI therapy, and direct the Computational Psychology and Well-Being Lab. In 2011, I co-founded the World Well-Being Project at the University of Pennsylvania, which has since become a big-data psychology consortium. How can Large Language Models (LLMs) be deployed for better mental health and well-being? One of the main directions of our lab is to determine the safe and responsible conditions under which LLMs can deliver psychotherapy and well-being interventions. Over the last decade, we’ve pioneered methods of psychological text analysis. Specifically, we use social media (Facebook, Twitter, Reddit, …) to measure the psychological states of populations and individuals. We use this to understand the thoughts, emotions, and behaviors that drive physical illness

Research topics

• Sociology
• Computer Science
• Psychology
• Gerontology
• Medicine
• Demography
• Artificial Intelligence
• Data science
• Natural Language Processing
• Engineering
• Social psychology
• Linguistics
• Cartography
• Statistics
• Psychiatry
• Econometrics
• World Wide Web
• Mathematics
• Geography
• Developmental psychology

Selected publications

• Assessing personality using zero-shot generative AI scoring of brief open-ended text

  Nature Human Behaviour · 2026-01-30 · 2 citations

  articleOpen access
  PublisherOA PDFDOI

• PsychAdapter: adapting LLMs to reflect traits, personality, and mental health

  npj Artificial Intelligence · 2026-03-02 · 1 citations

  articleOpen accessSenior author

  AI language generators are now ubiquitous but typically produce generic text that fails to reflect individual differences. Here, we introduce PsychAdapter, a lightweight LLM architectural modification that uses empirically derived links between language and personality, demographic, and mental health traits to generate trait-reflective text, regardless of prompt. PsychAdapter was applied to GPT-2, Gemma-2B, and LLaMA-3, and expert raters confirmed that the generated text matched the specified traits: it produced Big Five personality traits with 87.3% and depression and life satisfaction with 96.7% accuracy. PsychAdapter is a novel method for embedding psychological behavioral patterns into language models by conditioning every transformer layer, without relying on prompting. Beyond personality-conditioned generation, this approach has potential uses for simulated patients reflecting psychopathology and translation tailored to reading or educational level. It also enables generation of characteristic sentences for studying the language of traits, expanding the language processing toolkit for psychology.

  PublisherOA PDFDOI

• Risk Communication in Real Time: Examining Individuals’ Smartphone Use During the 2021 Texas Winter Disaster With Mobile Sensing

  Risk Analysis · 2026-03-31

  article

  Risk communication can support people in making timely, well-informed behavioral decisions during extreme weather events. Today's information environment has revolutionized how individuals engage with risk communication, as they see and share more information than ever before. It is not yet known how much risk communication any specific individual encounters, where it comes from, what messages it contains, and when it is received over the course of evolving extreme weather events. In this study, we provide a detailed description of how individuals encounter risk communication during extreme weather events by analyzing comprehensive records of all the content that 11 adults viewed on their smartphones (n = 162,418 screenshots collected via mobile sensing) over 5 days of the winter disaster that struck Texas in February 2021. Participants viewed substantial amounts of risk communication, which comprised, on average, 21% of their total smartphone use during the winter disaster. Most risk communication was accessed through participants' social networks, primarily through personal messaging apps (M = 33%) and social media apps (M = 22%). Risk messages contained updates, actionable advice, social support, sense-making, and recreation and humor, the specifics of which evolved over the course of the winter disaster. Taken together, our results demonstrate how individuals engage with risk communication in dynamic, heterogeneous, and socially embedded ways during extreme weather events-opening new possibilities for risk communication theory and practice.

  PublisherDOI

• AI-Generated Messages Can Be Used to Persuade Humans on Policy Issues

  2025-04-22 · 2 citations

  preprintOpen access

  The emergence of large language models (LLMs) has made it possible for generative artificial intelligence (AI) to tackle many higher-order cognitive tasks, with critical implications for industry, government, and labor markets. Here, we investigate whether existing, openly-available LLMs can be used to create messages capable of influencing humans’ political attitudes. Across three pre-registered experiments (total N=4,829), participants who read persuasive messages generated by LLMs showed significantly more attitude change across a range of policies - including polarized policies, like an assault weapons ban, a carbon tax, and a paid parental-leave program - relative to control condition participants who read a neutral message. Overall, LLM-generated messages were similarly effective in influencing policy attitudes as messages crafted by lay humans. Participants’ reported perceptions of the authors of the persuasive messages suggest these effects occurred through somewhat distinct causal pathways. While the persuasiveness of LLM-generated messages was associated with perceptions that the author used more facts, evidence, logical reasoning, and a dispassionate voice, the persuasiveness of human-generated messages was associated with perceptions of the author as unique and original. These results demonstrate that recent developments in AI make it possible to create politically persuasive messages quickly, cheaply, and at massive scale.

  PublisherOA PDFDOI

• Big team science reveals promises and limitations of machine learning efforts to model the physiological basis of affective experience

  2025-07-10

  preprintOpen access

  Researchers are increasingly using machine learning to study physiological markers of emotion. We evaluated the promises and limitations of this approach via a big team science competition. Twelve teams competed to predict self-reported affective experiences using a multi-modal set of peripheral nervous system measures. Models were trained and tested in multiple ways: with data divided by participants, targeted emotion, inductions, and time. In 100% of tests, teams outperformed baseline models that made random predictions. In 46% of tests, teams also outperformed baseline models that relied on the simple average of ratings from training datasets. More notably, results uncovered a methodological challenge: multiplicative constraints on generalizability. Inferences about the accuracy and theoretical implications of machine learning efforts depended not only on their architecture, but also how they were trained, tested, and evaluated. For example, some teams performed better when tested on observations from the same (vs. different) subjects seen during training. Such results could be interpreted as evidence against claims of universality. However, such conclusions would be premature because other teams exhibited the opposite pattern. Taken together, results illustrate how big team science can be leveraged to understand the promises and limitations of machine learning methods in affective science and beyond.

  PublisherDOI

• LLM-Generated Messages Can Persuade Humans on Policy Issues

  2025-06-19 · 3 citations

  preprint

  The emergence of large language models (LLMs) has made it possible for generative artificial intelligence (AI) to tackle many higher-order cognitive tasks, with critical implications for industry, government, and labor markets. Here, we investigate whether existing, openly-available LLMs can be used to create messages capable of influencing humans’ political attitudes. Across three pre-registered experiments (total N = 4,829), participants who read persuasive messages generated by LLMs showed significantly more attitude change across a range of policies - including polarized policies, like an assault weapons ban, a carbon tax, and a paid parental-leave program - relative to control condition participants who read a neutral message. Overall, LLM-generated messages were similarly effective in influencing policy attitudes as messages crafted by lay humans. Participants’ reported perceptions of the authors of the persuasive messages suggest these effects occurred through somewhat distinct causal pathways. While the persuasiveness of LLM-generated messages was associated with perceptions that the author used more facts, evidence, logical reasoning, and a dispassionate voice, the persuasiveness of human-generated messages was associated with perceptions of the author as unique and original. These results demonstrate that recent developments in AI make it possible to create politically persuasive messages quickly, cheaply, and at massive scale.

  PublisherDOI

• TherapyTrainer: Using AI to train therapists in written exposure therapy

  2025-06-25

  preprintOpen access

  Though evidence-based treatments for mental disorders are effective, existing implementation efforts are expensive and difficult to scale. Novel solutions— especially those that offer active learning strategies, repeat skill practice and personalized feedback to therapists — are needed to fill this gap. We developed TherapyTrainer, which uses large language models (LLMs) to allow therapists to practice delivering written exposure therapy (WET) for PTSD to AI-Patients while receiving expert feedback from an AI-Consultant. Here we present initial feasibility, acceptability and usability data for TherapyTrainer gathered from therapists, supervisors, and WET expert-consultants across iterative rounds of development. In Phase 1, we rapidly prototyped and developed TherapyTrainer based on ongoing feedback from WET clinicians and experts (n = 4). In Phase 2, mixed methods data from therapists engaged in an otherwise-routine WET workshop (n = 14) indicated that TherapyTrainer is feasible and acceptable and may help therapists feel prepared to deliver WET. In Phase 3, therapists (n = 6) completed structured user testing interviews to identify key issues impacting usability for subsequent rounds of development. AI and large language models hold potential to provide ongoing support to therapists in a cost-effective and scalable manner, and may help close the research-practice gap.

  PublisherOA PDFDOI

• Structured AI Dialogues Can Increase Happiness and Meaning in Life

  2025-10-02 · 2 citations

  articleOpen access

  Millions of people now use AI-powered chatbots to support their mental health, yet little is known about whether such interactions can effectively enhance psychological well-being. We conducted a preregistered experiment on a large, diverse sample (N = 2,922) to test four AI chatbots, each prompted to employ a multi-step strategy drawn from prior psychological research on sources of happiness and meaning in life. Chatbots encouraged participants to either (a) savor positive life experiences, (b) express gratitude toward a friend or family member, (c) reflect on sources of meaning in their life, or, (d) reframe their life story as a “hero’s journey.” All four chatbots led to improvements on a broad range of psychological well-being outcomes – including affective well-being, meaning in life, life satisfaction, anxiety, and depressed mood – relative to a control chatbot condition. These results generalized to key subpopulations, including those with high baseline levels of anxiety or depression. Chatbot interactions increased interest in seeing a human therapist, including among those who were previously unwilling or had never attended therapy. A separate, nationally representative survey (N = 3,056) found that half of U.S. adults expressed interest in using empirically validated AI chatbots for mental health support. These findings demonstrate that AI-driven well-being chatbots grounded in psychological research offer a scalable and effective way to produce short-term increases in several aspects of psychological well-being. Importantly, these results do not generalize to all AI-based emotional support.

  PublisherOA PDFDOI

• Monitoring the opioid epidemic via social media discussions

  npj Digital Medicine · 2025-05-15 · 5 citations

  articleOpen accessSenior author

  The opioid epidemic persists in the U.S., with over 80,000 deaths annually since 2021, primarily driven by synthetic opioids. Responding to this evolving epidemic requires reliable and timely information. One source of data is social media platforms. We assessed the utility of Reddit data for surveillance, covering heroin, prescription, and synthetic drugs. We built a natural language processing pipeline to identify opioid-related content and created a cohort of 1,689,039 Reddit users, each assigned to a state based on their previous Reddit activity. We measured their opioid-related posts over time and compared rates against CDC overdose and NFLIS report rates. To simulate the real-world prediction of synthetic opioid overdose rates, we added near real-time Reddit data to a model relying on CDC mortality data with a typical 6-month reporting lag. Reddit data significantly improved the prediction accuracy of overdose rates. This work suggests that social media can help monitor drug epidemics.

  PublisherOA PDFDOI

• Current Real-World Use of Large Language Models for Mental Health

  2025-06-23 · 7 citations

  preprintOpen accessSenior author

  There have been growing reports of people using general-purpose large language models (LLMs), like those offered through ChatGPT, for their mental health—but the extent of this use is not known. We conducted a survey of U.S. adults – stratified to ensure proportional representation across age, sex, and race – to quantify such use and understand users’ motivations and perceptions. A striking 24% of surveyed participants use LLMs for mental health. These users are more likely to be young, male, and Black and have poorer mental health and quality of life. They report difficulty accessing traditional mental health treatment – particularly due to cost and insurance coverage – and use LLMs because they are free, convenient, and available when needed. Users rely on LLMs for social and emotional support, to learn therapy skills and tools, and to supplement existing therapy. Non-users of LLMs for mental health expressed doubts about LLMs’ empathy and trustworthiness. Our sample likely overrepresents technology adopters, conservatively adjusting for this based on address-based estimates of population LLM use suggests that 13-17 million US adults may use general-purpose LLMs for mental health. These findings highlight the unmet need in mental health care and the urgency of providing scalable solutions. Though our study suggests widespread interest in using LLMs for mental health, including as part of LLM-human blended care, research should first establish that LLM systems can deliver safe and clinically effective care.

  PublisherOA PDFDOI

Frequent coauthors

• Lyle Ungar

  California University of Pennsylvania

  95 shared

• Margaret L. Kern

  University of Melbourne

  77 shared

• David B. Yaden

  Johns Hopkins University

  66 shared

• H. Andrew Schwartz

  51 shared

• Alejandro Adler

  University of Pennsylvania

  38 shared

• Fernando Ribeiro

  Universidade Federal do Rio Grande do Sul

  37 shared

• Salvatore Giorgi

  University of Pennsylvania

  32 shared

• Sharath Chandra Guntuku

  University of Pennsylvania Health System

  24 shared

Labs

• Stanford Computational Psychology & Well-Being LabPI

Education

• Ph.D., Psychology

  Stanford University

  2015

• M.A., Psychology

  University of California, Berkeley

  2010

• B.A., Psychology

  University of California, Berkeley

  2008

Awards & honors

• John Philip Coghlan Fellowship, Stanford (2023-2025)
• Rising Star, Association for Psychological Science (2022)
• Early Career Researcher Award, International Positive Psycho…
• Emerging Leader in Science & Society, American Association f…