AI system
- System
-
PAWS (Protection Assistant for Wildlife Security), enhanced iWare-E predictive model with Gaussian-process uncertainty and game-theoretic patrol planning
- Model
-
Gaussian-process models within iWare-E.
- Release
-
Exact release not established.
- Application developer
-
Harvard University and University of Southern California (Milind Tambe and colleagues), with WWF, WCS and Uganda Wildlife Authority co-authors - Role
-
The models predicted poaching risk to guide patrol locations.
- Configuration
-
The iWare-E models were trained on dry-season data from January 2015 through April 2018.
- Inputs and outputs
-
The models used historical patrol data to estimate where poaching was likely.
- Autonomy
-
Not assessed
- Workflow
-
Historical SMART patrol data (2015–2018) sufficient to train a risk model despite 3.6% positive labels. Rangers willing to patrol the suggested blocks and record findings in SMART. Dry-season deployment when rivers are passable and poaching concentrates in the north of the sanctuary.
- Human oversight
-
Not assessed
- Relation to AI
-
The AI's output (risk-ranked 3 x 3 km blocks) determined where patrols went. Rangers were blinded to risk class. High-risk blocks yielded detections while low-risk blocks yielded none. The paper itself cautions that the first QENP test's increased findings “may be partially explained by factors beyond the predictive model”. The 1,000-versus-200–300 comparison relies on rangers' reported averages, not a controlled baseline. The peer-reviewed paper is written by the system's academic developers together with staff of the operating organizations (WWF, WCS, Uganda Wildlife Authority). There is no commercial vendor.
- AI contribution
-
The AI's output (risk-ranked 3 x 3 km blocks) determined where patrols went. Rangers were blinded to risk class. High-risk blocks yielded detections while low-risk blocks yielded none. The paper itself cautions that the first QENP test's increased findings “may be partially explained by factors beyond the predictive model”. The 1,000-versus-200–300 comparison relies on rangers' reported averages, not a controlled baseline. The peer-reviewed paper is written by the system's academic developers together with staff of the operating organizations (WWF, WCS, Uganda Wildlife Authority). There is no commercial vendor.
Benefits
- Outcome identifiers
-
Not assessed
- Benefit
-
removed over 1,000 snares in December 2018, versus 200–300 in a typical month
- Category
-
Improved Environmental Sustainability
- Status
-
Not assessed
- Beneficiary groups
-
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
-
removed over 1,000 snares in December 2018, versus 200–300 in a typical month
- Comparison
-
the rangers' usual monthly snare removal without PAWS
- Results
-
Reference: two or three hundred snares in an average month. Result: over one thousand snares in the first month of field tests.
- 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:
Harvard University and University of Southern California (Milind Tambe and colleagues), with WWF, WCS and Uganda Wildlife Authority co-authors. Operator: WWF Cambodia with Cambodia Ministry of Environment rangers, Srepok Wildlife Sanctuary. - Jurisdictions
-
Benefit: Srepok Wildlife Sanctuary (Mondulkiri),
Cambodia. Developer: 
United States. Operator: Cambodia. - Ownership and control
-
Not assessed
- Dependencies
-
Not assessed
- Cross-border connections
-
Not assessed
- Geopolitical implications
-
Not assessed
Follow-up
- Conditions for replication
-
Patrol capacity exists to act on predictions (SWS ranger numbers had more than doubled since 2016). Poacher behavior remains spatially predictable from past detections. Findings are measured against a comparable baseline. The paper notes increases may partly reflect factors beyond the model.
- Replication attempts
-
2 follow-up entries. Benefit recurred: 2.
- Later findings
-
5 November 2020. By November 2020 PAWS had been incorporated into the SMART conservation software used in parks worldwide, and Srepok rangers reported finding 5 to 20 times more snares since adopting it.
- Availability
-
Not assessed
- Institutional decision
-
Not assessed
- Grounds
-
Not assessed
- Action and result
-
Not assessed
- Success criterion
-
Not assessed
Evidence
- Source numbers
-
AIBD-0029-source-0001–0003
- 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
-
Organizations. Society.
- Depth
-
Organizations: 2. Society: 3.
- Reasons
-
Organizations: The ranger force's monthly snare removals rose from a reported two or three hundred to over one thousand. A material gain in patrol effectiveness for the operating team. A chi-squared test confirmed predictions discriminated risk (p = 0.023). Society: Over one thousand snares were removed from a sanctuary identified as a tiger-reintroduction site. The paper states each snare “represents the life of one animal potentially saved”. A material landscape-scale conservation benefit rather than a documented population-level change.
- Observed or intended
-
Not assessed
- Rubric
-
Not assessed
- Assessor and date
-
Not assessed
- Alternative assessments
-
Not assessed
- Fill
-
Independently reported
Entry details
- Kind
-
Not assessed
- Related entries
-
Same task and domain: AI-directed anti-poaching in a protected area; PAWS patrols removed 1,000 snares, while Ezemvelo KZN Wildlife's AI cameras failed to detect poachers and four rhinos were killed.
- Field
-
Conservation / anti-poaching law enforcement. Protected-area ranger patrols.
- Purpose
-
PAWS was built to predict where poaching will occur and to prescribe patrol routes that maximize detection and removal of snares. The snare removals are exactly this intended benefit.
- Dates
-
Date basis unassessed: 2018 (precision unassessed). Updated: 22 October 2026.
- Stage
-
Not assessed
- Version
-
SHA-256: faaa16a57c0d…
References
- Other identifiers
-
AI Incident Database: 219.
- Category mappings
-
Not assessed
- Brief
-
Not assessed
- Revisions
-
6 revisions. Latest: 22 October 2026.
Sources
-
Peer-reviewed study Developer
Xu L, Gholami S, Mc Carthy S, Dilkina B, Plumptre A, Tambe M, Singh R, Nsubuga M, Mabonga J, Driciru M, Wanyama F, Rwetsiba A, Okello T, Enyel E (2020). Stay Ahead of Poachers: Illegal Wildlife Poaching Prediction and Patrol Planning Under Uncertainty with Field Test Evaluations. IEEE ICDE 2020.
During the first month of field tests in SWS, park rangers detected poaching in 32 cells and removed over one thousand snares
-
Peer-reviewed study Developer
confirm that the predictive power of PAWS extends promisingly to multiple parks
-
News report Independent author
Geib C (2020). Where to patrol next: 'Netflix' of ranger AI serves up poaching predictions. Mongabay.
rangers in the park have reported finding between 5 and 20 times more snares since they started using the PAWS system