The Weight of Words — Summary

The moment a person receives a diagnosis, language reshapes reality. “Dementia patient” is not the same person as “Eleanor.” The label precedes her into every room, every interaction, every assumption about capacity. We are building AI systems that learn and reproduce language. What happens when that language carries stigma?

Medical language developed for clinical efficiency — chart shorthand saves time, categories enable billing. But efficiency for whom? The chart becomes easier to read; the person becomes harder to see. Language models train on medical records containing decades of stigmatizing language: “combative,” “refuses medication,” “wandering.” The system does not know that “combative” often reflects overwhelmed staff rather than aggressive intent. It sees patterns and learns to reproduce them. Pattern without context becomes bias at scale, and it arrives with the authority of automation.

Deficit language dominates: “She can’t remember. He doesn’t understand. They are unable to.” But capacity is contextual and domain-specific. Margaret cannot recall what she ate for breakfast but can recall every verse of hymns learned in childhood. Her procedural memory for cooking remains intact. A system trained to see deficits will serve deficits — offering workarounds for weaknesses rather than scaffolding for strengths, protecting rather than enabling.

The stigma feedback loop: a person receives a diagnosis, clinical language enters their record, AI trains on records, AI generates deficit-framed outputs, caregivers treat the person according to the framing, the framing becomes reality. The loop closes.

Breaking it requires intentional design. Language auditing identifies stigmatizing patterns in training data. Reframing protocols transform “refuses” to “declines,” “aggressive behavior” to “distress response” — not euphemism but accuracy, since “refuses” implies willful resistance while “declines” describes choice. Temporal specificity recognizes that capacity fluctuates: the chart-Margaret and the morning-Margaret are not the same. The AI trained on charts will see the chart. The AI designed to see Margaret will notice her humor, her preferences, her morning clarity.

The words we teach machines to use will echo through every care decision, every moment of human contact mediated by AI. We should choose them carefully.