What's this about?
Mental health workers report that algorithmic triage systems — software that sorts and prioritises patients — are harming the people they are meant to route.
What is triage software actually optimising?
Not patient outcomes, usually. These systems are procured to allocate scarce appointments, and what they can measure is throughput, wait time and coverage. Clinical severity is inferred from proxies: intake questionnaire scores, diagnostic codes, prior utilisation. A patient who under-reports on a form, or whose distress does not fit the scored categories, is routed by the proxy rather than the condition — and the clinician who can see the difference is downstream of the decision.
Why does this matter to you?
HN 55 points, 19 comments
- It is the clearest current example of algorithmic decision-making where the affected party cannot appeal and the operator cannot easily audit.
- The workers reporting it are the people with the ground truth, and they are outside the feedback loop by design.
- Directly relevant to how any automated triage — including security finding triage — should be evaluated: on what it misses, not on what it processes.