Beth Semel

BETH SEMEL, then a graduate student at Massachusetts Institute of Technology, Cambridge, Massachusetts, was awarded funding in October 2016 to aid research on ‘Speech, Signal, Symptom: Psychiatric Diagnosis and the Making of Algorithmic Listening in the United States,’ supervised by Dr. Graham M. Jones. This multi-sited, ethnographic project investigated U.S. university-based research teams of psychiatric and engineering professionals collaborating to develop artificial intelligence (AI)-enabled speech analysis echnologies for mental health applications. Traditional techniques of psychiatric diagnosis in North America have depended on clinicians’ ability to interpret the semantic content of patients’ self-reported symptoms. Yet the neuroscientists, psychiatrists, and psychologists who make up the research teams upon which this dissertation focuses enlist the work of engineers and computer scientists to develop alternative means of deciphering the biomedical significance of patients’ spoken utterances, using AI to draw connections between acoustic features of speech and psychological pathology. By analyzing day-to-day activities associated with building and testing these technologies and tracking how researchers represent them to academic and general audiences, this dissertation illustrates that statements about AI’s transcendent objective capacities should be understood in their contexts of enunciation and materialization. The researchers’ attempts to use AI to cut through the sociocultural aspects of speaking in order to capture the biological basis of mental illness enacts ambivalence within Euro-American attitudes toward the value of speakers’ inner states in determining the meaning of speech, while also transforming conceptualizations of expertise, listening, and care within the culture of U.S. mental health.