AI system
- System
-
Ward-level poverty maps from supervised machine learning (deep learning) on satellite imagery, used for geographic targeting in the Rapid Response Register
- Model
-
Not assessed
- Release
-
Not assessed
- Application developer
-
Joshua Blumenstock (UC Berkeley) and Isabella Smythe (
Columbia University), using wealth estimates by Chi et al. (2021), with World Bank economists Jonathan Lain and Tara Vishwanath, for the Government of Nigeria - Role
-
Not assessed
- Configuration
-
Not assessed
- Inputs and outputs
-
Not assessed
- Autonomy
-
Not assessed
- Workflow
-
Deep-learning poverty map trained on geo-located household surveys. Government decision to adopt the ward ranking for eligibility. SMS blasts and USSD self-registration. In-person verification, de-duplication and payment onboarding.
- Human oversight
-
Not assessed
- Relation to AI
-
Decisive at the geographic stage: the ML poverty map alone chose which 1,163 wards were eligible. Individual inclusion depended on USSD self-registration and in-person verification, not AI. Payment and registration counts come from World Bank operational documents, not from the map developers. The PNAS paper and World Bank blog are co-authored by the developers.
- AI contribution
-
Decisive at the geographic stage: the ML poverty map alone chose which 1,163 wards were eligible. Individual inclusion depended on USSD self-registration and in-person verification, not AI. Payment and registration counts come from World Bank operational documents, not from the map developers. The PNAS paper and World Bank blog are co-authored by the developers.
Benefits
- Outcome identifiers
-
Not assessed
- Benefit
-
49,000 beneficiary households of the government's COVID-CT as of 15 October 2021
- Category
-
Increased Scalability
- Status
-
Not assessed
- Beneficiary groups
-
People. Organizations. Society.
- Beneficiaries
-
Not assessed
- Reach
-
Not assessed
- Distribution and access
-
Not assessed
- Duration
-
Not assessed
- Expected?
-
Yes
- Costs and harms
-
Not assessed
Measurement
- Studies
-
Not assessed
- Study design
-
Not assessed
- Measurement
-
Not assessed
- Setting and sample
-
Not assessed
- Measures
-
49,000 beneficiary households of the government's COVID-CT as of 15 October 2021
- Comparison
-
survey-based poverty map benchmark (DHS/NLSS data)
- Results
-
Reference: 0.79 precision at the ward level (survey benchmark). Result: 0.92 precision at the ward level (ML-based maps).
- Additional value from AI
-
Not assessed
- Limitations
-
Not assessed
- Denominators
-
Not assessed
- Materials for replication
-
Not assessed
Money
- Kind of value
-
Not assessed
- Amount
-
Not assessed
- Currency and prices
-
Not assessed
- Period and unit
-
Not assessed
- Realization
-
Not assessed
- Recipients and payers
-
Not assessed
- Gross or net
-
Not assessed
- Included costs
-
Not assessed
- Comparison
-
Not assessed
- Method
-
Not assessed
- Return and payback
-
Not assessed
People and places
- Participants
-
Developer: Joshua Blumenstock (UC Berkeley) and Isabella Smythe (
Columbia University), using wealth estimates by Chi et al. (2021), with World Bank economists Jonathan Lain and Tara Vishwanath, for the Government of Nigeria. Operator: National Social Safety Nets Coordinating Office (NASSCO), Federal Government of Nigeria. - Jurisdictions
-
Benefit: Urban wards across
Nigeria. Operator: Nigeria. - Ownership and control
-
Not assessed
- Dependencies
-
Not assessed
- Cross-border connections
-
Not assessed
- Geopolitical implications
-
Not assessed
Follow-up
- Conditions for replication
-
Ground-truth survey data for training labels. Ward boundaries compatible with map resolution. Mobile-phone access in target wards. Government funding for transfers and identity verification capacity.
- Replication attempts
-
1 follow-up entries. Benefit recurred: 1.
- Later findings
-
29 January 2025. The World Bank's 2021 scale-up project adopted the same machine-learning poverty maps to choose expansion wards for the RRR, and its January 2025 status report counts over 5 million households supported by the direct benefit transfer program after delays and a switch to biometric-verified payments, without breaking out the RRR-targeted share.
- Availability
-
Not assessed
- Institutional decision
-
Not assessed
- Grounds
-
Not assessed
- Action and result
-
Not assessed
- Success criterion
-
Not assessed
Evidence
- Source numbers
-
AIBD-0094-source-0001–0005
- Citations
-
Numbered references below.
- Language
-
Not assessed
- Source relationships
-
Developer, Independent
- Original evidence
-
Not assessed
- Access and versions
-
Not assessed
- Supported findings
-
Not assessed
- Evidence links
-
Not assessed
- Evidence strength
-
Not assessed
- Missing evidence
-
Not assessed
Grades
- Assessed benefits
-
People. Organizations. Society.
- Depth
-
People: 2. Organizations: 2. Society: 3.
- Reasons
-
People: Households in selected wards were beneficiaries of the government's COVID-CT. The PNAS authors say disbursement began mid-January 2021. The World Bank ISR records 49,000 beneficiary households as of 15 October 2021 without saying how many had been paid. Organizations: NASSCO stood up an urban registry within months through USSD self-registration guided by the ML poverty map. Over 1.75 million registrations succeeded (PAD, Nov 2021). Society: Material to many: 1,163 wards with over 20 million residents made eligible. PNAS simulation finds the ML maps would cut inclusion errors by 61.7% versus the survey-based benchmark.
- Observed or intended
-
Not assessed
- Rubric
-
Not assessed
- Assessor and date
-
Not assessed
- Alternative assessments
-
Not assessed
- Fill
-
Independently studied
Entry details
- Kind
-
Not assessed
- Related entries
-
World Bank-developed poverty-targeting algorithm for cash transfers (Takaful) oversimplified vulnerability and excluded poor Jordanians; Nigeria's register used World Bank poverty maps.
- Field
-
Social protection and cash transfers. Government social assistance.
- Purpose
-
Geographically target a COVID-19 urban cash transfer to the poorest wards where no social registry existed.
- Dates
-
Date basis unassessed: 2021 (precision unassessed). Updated: 21 October 2026.
- Stage
-
Not assessed
- Version
-
SHA-256: 2c38cee688ad…
References
- Other identifiers
-
AI Incident Database: 546.
- Category mappings
-
Not assessed
- Brief
-
Not assessed
- Revisions
-
7 revisions. Latest: 21 October 2026.
Sources
-
Peer-reviewed study Developer
Smythe, I. S., Blumenstock, J. E. (2022). Geographic microtargeting of social assistance with high-resolution poverty maps. PNAS 119(32).
the government elected to use our high-resolution poverty maps to target the COVID-19 Rapid Response Registration (RRR) Cash Transfer Project
-
Report Developer
1,163 of the poorest urban wards with a total population of over 20 million people were selected based on the ward-level poverty maps produced by applying cutting-edge machine learning algorithms
-
Report Developer
Baseline numbers based on the 49,000 beneficiary HHs of the COVID-CT implemented by the government as of October 15, 2021.
-
Report Independent author
Human Rights Watch (2021). Between Hunger and the Virus: The Impact of the Covid-19 Pandemic on People Living in Poverty in Lagos, Nigeria.
a new 'rapid response' cash transfer program, which began in January 2021 and is intended to reach 1 million people nationwide, primarily in urban poor communities
-
Press release Developer
Blumenstock, J., Lain, J., Smythe, I., Vishwanath, T. (2021). Using Big Data and machine learning to locate the poor in Nigeria. World Bank Data Blog.
which is being guided by the mapping exercise described below