AI system
- System
-
nEmesis (SVM language model over geotagged tweets with a web interface for inspectors)
- Model
-
Not assessed
- Release
-
Not assessed
- Application developer
-
University of Rochester (Adam Sadilek, Henry Kautz and colleagues)
- Role
-
Not assessed
- Configuration
-
Not assessed
- Inputs and outputs
-
Not assessed
- Autonomy
-
Not assessed
- Workflow
-
Dense geotagged social-media posting across a metropolitan area. A health department willing to assign half of a matched inspection sample to algorithm-flagged venues. A graded inspection protocol (demerits, letter grades) that makes arms comparable.
- Human oversight
-
Not assessed
- Relation to AI
-
The AI chose which venues to inspect. The comparison was a controlled double-blind design with 142 venues split between nEmesis and standard protocol. So the difference in findings is attributable to the AI's venue selection. The developers (Rochester) are distinct from the deployer (SNHD), though SNHD staff co-authored the paper. The paper was peer-reviewed as an IAAI deployed paper.
- AI contribution
-
The AI chose which venues to inspect. The comparison was a controlled double-blind design with 142 venues split between nEmesis and standard protocol. So the difference in findings is attributable to the AI's venue selection. The developers (Rochester) are distinct from the deployer (SNHD), though SNHD staff co-authored the paper. The paper was peer-reviewed as an IAAI deployed paper.
Benefits
- Outcome identifiers
-
Not assessed
- Benefit
-
found 11 venues graded C or worse versus 7 for the control arm, a 64% improvement
- Category
-
Improved Task Accuracy
- 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
-
found 11 venues graded C or worse versus 7 for the control arm, a 64% improvement
- Comparison
-
Not assessed
- Results
-
found 11 venues graded C or worse versus 7 for the control arm, a 64% improvement
- 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: University of Rochester (Adam Sadilek, Henry Kautz and colleagues). Operator:
Southern Nevada Health District. - Jurisdictions
-
Benefit: Las Vegas,
United States. Developer: United States. Operator: United States. - Ownership and control
-
Not assessed
- Dependencies
-
Not assessed
- Cross-border connections
-
Not assessed
- Geopolitical implications
-
Not assessed
Follow-up
- Conditions for replication
-
Access to geotagged public posts and venue locations as available through the Twitter and Google Places APIs in 2015. Inspectors blinded to the source of the assignment. Enough flagged venues per day to schedule adaptive inspections.
- Replication attempts
-
1 follow-up entries. Benefit recurred: 1.
- Later findings
-
No later report of SNHD continuing or expanding nEmesis after the 2015 experiment was found; the 2016 FINDER deployment is a different model.
- Availability
-
Not assessed
- Institutional decision
-
Not assessed
- Grounds
-
Not assessed
- Action and result
-
Not assessed
- Success criterion
-
Not assessed
Evidence
- Source numbers
-
AIBD-0188-source-0001
- Citations
-
Numbered references below.
- Language
-
Not assessed
- Source relationships
-
Developer
- 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: Not assessed.
- Reasons
-
Organizations: In 142 blinded inspections the nEmesis arm found 9 vs 6 demerits per inspection (p = 0.019). It found 11 vs 7 venues graded C or worse, an unpermitted seafood establishment and a sick food handler. Society: The 71 infections and 4.4 hospitalizations said to be saved are estimates from CDC outbreak averages. The drop in foodborne-illness investigations (5 vs 11) is an uncontrolled year-on-year comparison, not a measured outcome.
- Observed or intended
-
Not assessed
- Rubric
-
Not assessed
- Assessor and date
-
Not assessed
- Alternative assessments
-
Not assessed
- Fill
-
The makers' claim
Entry details
- Kind
-
Not assessed
- Related entries
-
Not assessed
- Field
-
Food safety / public health. Restaurant inspection targeting.
- Purpose
-
Rank Las Vegas food venues by predicted foodborne-illness risk, inferred from Twitter posts, so that health inspectors can prioritize the riskiest venues.
- Dates
-
Date basis unassessed: 2015 (precision unassessed). Updated: 21 October 2026.
- Stage
-
Not assessed
- Version
-
SHA-256: aafd5697d61c…
References
- Other identifiers
-
Not assessed
- Category mappings
-
Not assessed
- Brief
-
Not assessed
- Revisions
-
4 revisions. Latest: 21 October 2026.
Sources
-
Peer-reviewed study Developer
Sadilek A, Kautz H, DiPrete L, Labus B, Portman E, Teitel J, Silenzio V. Deploying nEmesis: Preventing Foodborne Illness by Data Mining Social Media. Proceedings of the AAAI Conference on Artificial Intelligence 30(2):3982-3989 (2016). doi:10.1609/aaai.v30i2.19072
We determined that adaptive inspections uncover significantly more demerits: 9 vs. 6 per inspection (p-value of 0.019).