Machine Learning, Mental Health and Eugenics.

Machine learning is set to become deeply embroiled in mental health diagnosis. The so-called digital phenotype is dizzyingly broad, so that changes in our smartphone usage, data from fitness trackers and the tone of words we use on Twitter will become factors in predictive diagnosis. But this is a collision between computational exactness and the inconsistencies concealed by psychiatric labeling. While the idea of 'precision psychiatry' promises to detect early onset of psychosis before services or the individual themselves are able to, user movements argue that a lot of what is being medicalized is the expression of trauma or distress. The volumes of data needed for AI will require a pervasive surveillance that will amplify the anxiety already instilled by a distrustful and punitive benefits system. AI is not a neutral actor but will weigh in heavily on the side of biological reductionism, reinforcing the understanding of mental health problems as disorders of circuits in the brain rather than accounting for adverse life events. This parallels the emerging field of sociogenomics and the use of Genome Wide Association Studies (GWAS) to correlate social status with distributed genetic factors, re-opening eugenicist narratives that had been thought of as buried in history. Both precision psychiatry and GWAS class mental health problems as innate tendencies and act as smoke-screens to obscure social and political conditions. To overcome the onlooker consciousness of AI, we need critical technical practices that can unlink vectorial distances from social differences. This can come through a feminist AI that draws on standpoint theory and feminist approaches to science, combined with collective structures of research that include those who are most affected in the process of inquiry. We need an alternative psychopolitics of machine learning.