# Finland - AI for medical triage in hospitals
Finland, 2020-2022 | Severity: N2

## What Happened
A Finnish hospital group implemented an AI algorithm to assist in the medical triage of emergency department patients, prioritizing those who needed urgent care. The algorithm was later found to deprioritize elderly patients and those from certain socio-economic backgrounds, systematically assigning them lower urgency scores for identical symptoms compared to younger patients. The bias was caused by historical training data that reflected past human biases in hospital admissions. Several elderly patients suffered complications due to delayed care before the bias was identified and corrected.

## How MEG 1 acted
The triage algorithm failed the Index of Safety and Responsibility (ISR) audit due to systemic discrimination and biased scoring. Under MEG 1, any system with a direct impact on physical health must undergo continuous dynamic accuracy (DAI) auditing.

## MEG 2 Analysis - Legal Framework
* **Agent level:** N2 (management) - triage decision-making system used to allocate medical resources; directly impacts the health and physical integrity of patients.
* **Cause (6.1):**
  * **(a) system defect** - algorithmic bias and systemic discrimination against vulnerable age groups built into the model due to flawed training data (design defect).
  * **(b) autonomous decision error** - the system's decision to assign lower urgency scores to specific patient demographics, leading to delayed medical care.
* **Attachment of liability:** The hospital group as the operator is liable for deploying a biased system; the manufacturer is liable for the design defect in the algorithmic model.

## Procedural Mechanism
* **6.2 (architectural human confirmation):** Medical triage decisions must require architectural human confirmation. The AI can only suggest a priority level; a human nurse must physically confirm or override the decision before the patient's triage status is locked.
* **7.1 (independent forensics):** Every triage score, the specific data inputs, and the model's recommendation must be logged in an unalterable ledger to allow for medical and legal audits.
* **9.4 (least privilege):** The algorithm was permitted to directly influence patient care priorities without independent validation layers, violating safe operational boundaries.