Abstract
“Technology doesn't just do things for us. It does things to us, changing not just what we do but
who we are.” - Sherry Turkle
Treatment of mental health in America now is faced with a disparity in access, with millions of Americans unable to get effective, timely treatment due to economic barriers, rural isolation, and a shortage of trained professionals. Machine learning is emerging as a possible bridge to close this gap and enhance the current psychotherapy field by increasing diagnostic accuracy and providing alternate methods of care, such as chatbots, to increase availability of mental health services beyond the traditional clinic. My technical project explored these capabilities through a meta-study of various machine learning applications in treating mental health, while my STS research paper investigated how such technologies aid in the datafication of human experiences using a technological momentum approach. Together, these studies attempt to understand the potential that machine learning holds and the society machine learning can create for itself.
In my technical research, I studied how various machine learning applications are being integrated into mental health treatment in a comprehensive meta-study of academic research and real-world implementations. My research revealed that deep neural networks (DNNs) and natural language processing (NLP) have the potential to improve diagnostic performance for conditions like ADHD and Alzheimer's disease, in some cases surpassing the performance of experienced clinicians'. Mental health chatbots showed significant reductions in depressive symptoms and improved well-being across diverse populations, from college students to individuals with chronic conditions like migraines. These technologies provide those with financial limitations, and those isolated from areas with available mental health treatment, with an available solution, along with improving the existing diagnostic power of therapists. However, challenges with machine learning must be addressed before widespread implementation can be recommended, specifically the "black box" nature of many machine learning models and inconsistent user engagement patterns.
My STS research revealed the societal impacts of integrating machine learning into mental health care. With a technological momentum framework, I examined how machine learning technologies create a self-reinforcing datafication loop that uses patient data as training material for machine learning models. My research uncovered how these technologies, once established, gain momentum that reshapes the social and institutional structures around mental health care. As ML models require large amounts of data to function properly, previously private information like social media posts, conversations with therapists, and emotions becomes redefined as necessary training data. This shifting expectation of privacy has implications for how patients engage with therapy and how society understands mental health treatment. The power dynamics among therapists, patients, and machine learning technologies become reconfigured, with developers of machine learning technologies gaining direct control over therapeutic interactions through algorithm design without direct accountability to patients.
The dynamics between my technical and STS research demonstrate what can be gained by considering technical, organizational, and cultural variables simultaneously. My technical research highlighted the capabilities of machine learning in mental health care, while the STS perspective investigated how the same technologies might fundamentally change power structures within a human-centered environment like psychotherapy, influence national standards around privacy, and ultimately impact the human experience of therapy in general. The concept of technological momentum allows engineers to recognize that their design decisions made early in development can make it increasingly difficult to redirect trajectories as technologies become embedded within social systems. This concept encourages engineers to determine not only if a technology works or is efficient, but also to be aware of how it might impact social structures and human experiences over time.
Together with these perspectives, I have developed an understanding of how machine learning technologies could responsibly be developed for use in mental health. More than a focus on improving algorithmic accuracy or the user interaction experience, engineers should also consider the effects on patient autonomy, expectations for data privacy, and the patient therapist relationship. Ultimately, a technology development perspective that considers social processes also facilitates ethical accountability in engineering because it requires developers to think about potential consequences on society and engage in critique before technologies achieve momentum that makes intervention difficult.
