Various iterations of AI “hold tremendous promise” to help personalize counseling for individuals grieving the loss of a loved one, according to a scientific-literature review slated for publication in the February 2022 edition of Current Opinion in Psychology.
In introducing their findings, the study’s authors note that, contrary to conventional assumptions, people mourn in widely dissimilar ways.
Given the updated consensus, they suggest, AI holds promise for both guiding one-to-one care and, at the macro level, illuminating the very nature of grief.
Along with trajectory modeling and network analyses of grief symptoms, Matteo Malgaroli, PhD, of New York University and colleagues discuss recent advances in AI-driven approaches that offer new opportunities for teasing out actionable features of grief in all its forms.
“These methods capture high-dimensional nonlinear relationships that can associate different information levels (e.g. biological, multiomics) with mental health outcomes,” they write.
The authors point to one recent study proving out a deep neural network that can prospectively—and accurately—connect hereditary risk scores with long-term bereavement outcomes.
Here the authors comment that the exploration of population heterogeneity has become “more empirically grounded with the adoption of unsupervised learning approaches.
“One such approach, latent growth mixture modeling (LGMM), tests whether the sample is best represented by a single response trajectory or several discrete populations, each characterized by a different growth curve pattern.”
Another study the team synopsizes used LGMM to track PTSD symptoms over one year following initial presentation in the emergency department with a potentially traumatic event.
An independent sample of trauma survivors treated at an unaffiliated institution externally validated this latter model’s algorithm.
Toward treatment ‘firmly rooted in data’
“Beyond their use in more sophisticated analyses, AI-derived methods including computer vision and natural language processing hold tremendous promise as empirical markers of mental health status,” Malgaroli and colleagues write.
Such methods offer the ability to objectively tease out subtle signs of symptoms such as facial expressions, vocal patterns and psychomotor traits.
Malgaroli et al. conclude:
[C]omputational approaches have the advantage of transparency, given their data-driven nature and the ability to share analytic codes and methods. These characteristics meet the fundamental requirements needed to improve replicability in science, offering insights on grief that are clear and can be re-evaluated in multiple groups and contexts. Given these considerations, we believe that embracing computational methods is an important step to make our field as scientific as possible, with the ultimate goal of providing treatment and informing policies that are firmly rooted in data.”
The study’s senior author is George Bonanno, PhD, head of the Loss, Trauma and Emotion Lab at Columbia University in New York and author of The Other Side of Sadness and The End of Trauma.
Abstract posted here (full study behind paywall).