Back to Challenges
DIGITAL PUBLIC INFRASTRUCTURE
under_review
AI GeneratedAI Model Bias in Low-Resource Contexts: Failing for 90 Percent of Global Population
Problem Definition
AI fails in low-light conditions, non-English languages, and non-WEIRD cultural contexts. Medical AI shows 10-20 percent degradation on darker skin tones. Facial recognition has 35 percent error in African populations. Agricultural AI misclassifies 60 percent of tropical diseases.
Root Causes
Training data biased toward WEIRD populations
Models not validated in target deployment contexts
No diverse data collection mandates
Deployment happens before bias testing
Scope
Share
Priority
Evaluation Scores
Complexity7.0
Priority8.0
Interconnected8.0
Risk Level8.0
Clarity8.0
Composite Score
7.8
Metadata
Linked Ideas:No ideas yet
Evaluations:2
Version:1