Farmers can diagnose cassava disease without waiting for scarce expert access or relying on unaided vision.
Executive brief
The model outperformed typical users, but the workflow still depends on owning a suitable phone, capturing six useful leaves, and knowing when to escalate.
AI value · Field diagnostic accuracy
VerifiedNuru 65% in 2020; 74-88% when six leaves per plant were assessed
Diagnostic accuracy is not crop-yield impact. Adoption research found 45% adoption, 65% smartphone unavailability, and 41% complexity constraints.
Before
Inspect visible leaf symptoms using personal knowledge. → Provide diagnosis when reachable.
After
Scans multiple leaves with an Android phone in the field without internet. → Classifies disease or pest symptoms and provides real-time guidance. → Decides whether to act, rescan, or seek expert confirmation.
Human boundary
Nuru returns a diagnosis and advice; the farmer or extension officer decides what action to take.
Why it matters
Farmers can diagnose cassava disease without waiting for scarce expert access or relying on unaided vision.
Before
How the work ran before the change.
Farmer or extension officer
Inspect visible leaf symptoms using personal knowledge.
ControlHuman diagnostic skill
Expert
Provide diagnosis when reachable.
ControlTravel and availability
What changed
Farmers can diagnose cassava disease without waiting for scarce expert access or relying on unaided vision.
Decision rightHuman moves from creator to judge
After
How the same work runs now.
Farmer or extension officer
Scans multiple leaves with an Android phone in the field without internet.
ControlImage-quality and six-leaf protocol
Nuru
Classifies disease or pest symptoms and provides real-time guidance.
ControlModel confidence and supported conditions
Human user
Decides whether to act, rescan, or seek expert confirmation.
ControlHuman authority
Exception path
Users can scan six leaves, repeat under better lighting, or escalate uncertain and unsupported cases to trained experts.
Decision authority
Nuru returns a diagnosis and advice; the farmer or extension officer decides what action to take.
| # | Actor | Action | Control |
|---|---|---|---|
| 01 | Farmer or extension officer | Inspect visible leaf symptoms using personal knowledge. | Human diagnostic skill |
| 02 | Expert | Provide diagnosis when reachable. | Travel and availability |
| # | Actor | Action | Control |
|---|---|---|---|
| 01 | Farmer or extension officer | Scans multiple leaves with an Android phone in the field without internet. | Image-quality and six-leaf protocol |
| 02 | Nuru | Classifies disease or pest symptoms and provides real-time guidance. | Model confidence and supported conditions |
| 03 | Human user | Decides whether to act, rescan, or seek expert confirmation. | Human authority |
Work that left the path
Human role before
Farmers and extension agents relied on uneven diagnostic expertise and delayed expert access.
Human role after
Users capture standardized evidence, interpret a model result, and retain crop-management decisions.
AI role
Offline deep-learning object detection for cassava mosaic disease, brown streak disease, and green-mite damage.
Farmers 18-31% and extension agents 40-58%Nuru 65% in 2020; 74-88% when six leaves per plant were assessed
2020 East Africa field evaluation · Field diagnosis of cassava disease and pest symptoms compared with farmers, extension agents, and experts
Diagnostic accuracy is not crop-yield impact. Adoption research found 45% adoption, 65% smartphone unavailability, and 41% complexity constraints.
Anti-pattern
Equating diagnostic accuracy with farmer income or ignoring access and usability barriers.
Questions
Portability conditions
Reputation risk
medium