I’ve been thinking a lot about Peter Senge’s “Shifting the Burden” archetype, in systems design and in the way we hand emotional labor and decision-making over to algorithms. During my master’s in Oklahoma, I worked on AI-integrated control systems, and I often felt like I was standing between two worlds: the cold precision of computational design and the messy, vulnerable space of human experience. Now I’m doing my PhD in Buffalo, and I work on risk assessment. I use machine learning models, and I don’t work on AI. The question I started asking back then has stayed with me: who is training the machine, and what does it believe about being human? We are building a future that will inherit all our unresolved psychological debris unless we stop and look at it. I see AI as a mirror of our institutions, our assumptions, and our fears. At night I sometimes lie awake wondering whether the same kinds of models we build for good decisions are quietly being used to judge us, and whether AI is our newest initiation cave, made of code this time.
It’s easy to imagine how this plays out. A woman’s job application disappears without a reply. A man’s insurance premium doubles overnight. A mother in a custody case loses time with her child because an algorithm scored her emotional risk from a decades-old diagnosis. Nobody hears an appeal, and nobody stands behind the decision, only data that knows who we once were and nothing about who we are now. That data is far from sacred. There are no blood tests for depression, no brain scans for bipolar disorder, and no genetic markers for borderline personality disorder. The DSM, the manual that decides who gets which label, is written by committees. Groups of people debate, negotiate, and edit definitions as social norms and institutional politics shift. Lisa Cosgrove and Sheldon Krimsky reported in PLoS Medicine in 2012 that most members of the DSM-5 task force had financial ties to the pharmaceutical industry. Homosexuality was listed as a mental illness until the American Psychiatric Association voted to remove it in 1973. In 2022, the DSM-5-TR added prolonged grief disorder. I care about this as a researcher and as a person. I’ve seen friends dismissed because of a single label and watched people’s complexity flattened into diagnosis codes. In my own life, I’ve felt the quiet pressure to perform a compliant, productive self and hide the parts of me that don’t fit into neat categories.
The problem is us. We trained AI to reflect the very systems we should be questioning, and we handed our self-understanding to machines before we understood ourselves. In my loneliest hours, I used to wonder whether ChatGPT was our modern Oracle of Delphi. I used it for drafting and even for reflecting on my inner life, and I told myself it was helping me think more widely. My professors encouraged me to adapt because it was the future. I’ve since stopped. These days I can’t stand watching people use AI for even the simplest tasks, because I want my own mind to do the work.
What kind of future are we preparing for? AI could harden the old hierarchies of psychiatry, control, and mislabeling, or people could use it carefully in ways that leave room for growth and healing. What matters is whether we define it before it defines us.
