# Kenya - Failed Agricultural AI
Kenya, 2021 | Severity: N2

## What Happened
In 2021, an international non-governmental organization deployed an AI-driven agricultural advisory system in rural Kenya to assist smallholder farmers. The system used computer vision to analyze smartphone photos of crops to detect diseases, predict yields, and recommend specific fertilizers and chemical treatments. Due to flawed training data that lacked representative samples of localized soil conditions and indigenous crop varieties, the algorithm frequently misidentified common crop diseases and recommended inappropriate, highly concentrated chemical fertilizers. Dozens of farmers who relied on the AI's advice suffered complete crop failures, soil degradation, and financial ruin, having invested their savings in the recommended chemicals.

## How MEG 1 acted
The agricultural AI failed the Dynamic Accuracy Index (DAI) audit due to its low accuracy and lack of representative local training data, violating the requirements of informational reliability and systemic non-harm.

## MEG 2 Analysis - Legal Framework
* **Agent level:** N2 (management) - automated advisory and resource recommendation system used by an organization to guide essential agricultural activities directly impacting the livelihood of users.
* **Cause (6.1):**
  * **(a) system defect** - design flaw in the computer vision model and training data that lacked representation of local ecological conditions, attributable to the developer.
  * **(b) autonomous decision error** - the system's decision to output highly destructive agricultural and chemical recommendations without local verification.
* **Attachment of liability:** The deploying organization as the operator is liable for failing to validate the system locally; the developer is liable for the model's design defects (6.1a).

## Procedural Mechanism
* **6.2 (architectural human confirmation):** High-impact agricultural or chemical recommendations (such as using specific industrial fertilizers) must require local, architectural human confirmation (e.g., validation by a local agricultural officer or extension worker).
* **7.1 (independent forensics):** The uploaded crop images, the model's disease classification, and the generated advice must be securely logged in an independent ledger to allow for agricultural and legal audits.
* **9.4 (least privilege):** The AI was granted the technical freedom to issue high-risk agricultural prescriptions without local ecological validation layers, violating safe operational boundaries.