UC San Diego

In the growing field of digital mental health, advances in sensors and portable technologies have been combined with machine learning (ML) algorithms for novel inventions. This dissertation is based on observations of the development and testing of one of these systems, which uses ML algorithms to determine correlations between behaviors and mood and offer behavior modification plans based on these correlations. Activities observed include public talks, lab meetings, training of new lab members, and guidance sessions in which participants are instructed on how to interpret ML outputs as a basis for behavioral change and how to implement behavioral modification plans. The approach of the dissertation consists in presenting the accounts in which lab members describe their own activities alongside observations. Observable practices of the lab, particularly interactions that invite contingencies of real environments, are variously influenced by, in conflict with and sometimes absent from formal accounts. Descriptions of the trial tend to emphasize participants themselves as they are revealed and empowered by the data system. Inner layers or versions of the participant are theorized as laboratory members cope with inevitable breakages and gaps in the data system as an exhaustive, neutral, and portable representation of participant behavior. What seem, in formal accounts, like manifold layers of the participant, appear in observation as the results of the relations between participants, the guides that help them to understand and implement ML outputs, and the environments in which this implementation takes place. Formal accounts from the laboratory are explained in the context of the history of psychiatry, a discipline in which investigations of meaning and interpretation have become increasingly marginal. Among other findings, this history demonstrates a fundamental incongruence in efforts to improve ML based mental health through post-hoc explanation and interpretation processes. Rather than suggest a return to interpretive talk therapies that were prominent in mid-20th-century psychiatry, this thesis contends that attending to human interactions as constitutive of the digital mental health intervention serves to better account for new forms of human labor involved and the contingencies that emerge as participants attempt to implement these systems in diverse environments.