Elizabeth Olson, Ph.D.

About

I’m an applied science leader and clinical psychologist specializing in AI, NLP, and machine learning in healthcare settings. I currently serve as Vice President of AI/NLP Clinical Research Science at Crisis Text Line, where I lead a global, interdisciplinary team using large language models and transformer-based architectures to analyze over 12 million mental health crisis conversations.

My work focuses on the development and governance of ethically responsible AI tools in high-risk, high-sensitivity environments. This includes suicide risk detection, multilingual NLP, and real-time behavioral health insights. I also contribute to AI risk frameworks, IRB processes, and data privacy protocols at the organizational level.

Prior to joining Crisis Text Line, I spent a decade as an Assistant Professor in the Department of Psychiatry at McLean Hospital / Harvard Medical School, where I founded and directed a federally funded lab on trauma, resilience, and social behavior. My academic research used digital phenotyping, mobile sensing, and multimodal neuroimaging (fMRI, MRS, DTI) to study social withdrawal, reward learning, and mental health during critical developmental periods. In the earliest part of my career, I focused on understanding how brain development contributes to the development of decision-making during adolescence and the transition to adulthood. 

My work has been supported by the National Institute of Mental Health (K23, R01), the CDC, the American Public Health Association, and the Brain & Behavior Research Foundation. 

Selected Recent Projects / Publications

  • My team and I built a natural language processing pipeline to extract mentions of planned suicide methods from crisis conversation transcripts, providing a unique pathway to understanding changes in planned methods in near-real-time. (Tripodi et al., JAACAP Open, in press) 
  • We used NLP techniques to identify relationships between linguistic features of crisis counselor speech and the texter’s perception of genuine concern, opening pathways to improve the effectiveness of crisis conversations. (Buda et al., EMNLP 2024). 
  • We developed an NLP model that provides continuous, granular scoring of the emotional ‘heat’ of statements in conversations, which can serve as a tool to measure how effectively an intense conversation is de-escalated. (Tripodi et al., EMNLP 2025). 
  • In my prior role, my colleagues and I identified differences in the relationship between loss of pleasure (anhedonia) and decision-making in people with PTSD.  (Olson et al., Biological Psychiatry CNNI, 2024). 
 

Contact

elizabeth@elizabetholsonphd.com

LinkedIn