Mental Health Workers Say Algorithmic Triage Is Hurting Patients

TEXT START: When Kaiser Permanente triage clinician Harimandir Khalsa began working in the psychiatry department at Kaiser’s Walnut Creek Medical Center in Northern California, she was on a team of nine people.

THE DISSECTION

The article documents automation being used as a rationing gate, not merely as a clinical aid. Patients are pushed through apps, call centers and algorithmic filters, then fewer human workers are left to repair the damage. “Service recovery” becomes the hidden labor of automation: clinicians absorb the anger, missed risks and emotional fallout while institutions retain the efficiency gains and deny replacement.

The corporate defense—licensed clinicians still make the final decisions—is technically narrower than the workers’ complaint. The algorithm may not sign the order, but it can determine who reaches a clinician, how quickly, and at what level of care. That is functional control disguised as human oversight.

THE CORE FALLACY

The article correctly identifies patient harm but misdiagnoses the underlying mechanism as hasty or poorly governed deployment. Under the Discontinuity Thesis, the deeper driver is structural: cognitive and administrative labor is treated as a cost center, staffing is compressed, and automation is inserted wherever it can reduce human throughput. “Better AI,” transparency and override rights are lag defenses. They may slow the damage; they do not reverse the economic pressure producing it.

The faith that licensed human judgment can remain permanently central also ignores P1–P3. As automation improves and coordination becomes more machine-mediated, human professionals are retained increasingly as exception handlers, liability shields and relationship labor—not as the governing core of the system.

HIDDEN ASSUMPTIONS

• Human clinicians will remain economically available at the scale required for safe care.
• “Assist, not replace” has operational meaning while staffing falls from nine triage workers to three.
• Guardrails and legislation can preserve a stable human-only domain despite institutional cost pressure.
• Better data or models can solve failures rooted in understaffing, fragmented responsibility and throughput incentives.
• Depressed, manic, suicidal or otherwise impaired patients can navigate digital systems like competent administrators.
• Human therapeutic relationships will remain accessible rather than becoming a scarce premium service.
• Legal liability will force institutions to improve care instead of merely transferring blame among providers, vendors and clinicians.

SOCIAL FUNCTION

Primary classification: partial truth. Secondary classification: transition management and ideological anesthetic.

The article is valuable because it records the early injuries: delayed access, inappropriate placement, displaced coordination work and intensified clinician exhaustion. But its proposed remedy channels the conflict toward guardrails, transparency and preserving “the uniquely human.” That gives workers a defensible bargaining position while allowing institutions to continue the transition under stricter supervision. It exposes the machinery without confronting the economic engine driving it.

THE VERDICT

The article is right about the harm and wrong about the remedy. This is an early autopsy of productive participation collapse in a high-stakes sector. The system is not yet fully automated; it is already being reorganized around automation, with humans left to mop up its failures. The patient is being denied care by a machine that officially makes no decisions. That distinction is bureaucratic camouflage. Under DT logic, the human clinician’s future is not secure centrality but narrower exception handling—unless they gain ownership, control, or indispensable leverage over the systems replacing their work.