What kind of protections does ResourceRouter have to minimize bias or discrimination?
SoundThinking seeks to reduce crime with its solutions, but that capability alone is not enough to make a public safety solution of benefit to a community. The product must provide more benefit than harm. The company has incorporated technology and policy protections to mitigate bias and discrimination yet still yield a significant public safety benefit. These protections include:
- We use crime data that is least susceptible to bias – Our models only use data for crime types that are typically called in from the community and not driven by police presence. We exclude misdemeanor and nuisance crimes that can create negative feedback loops with enforcement bias. These loops can occur in other modeling approaches when police presence in an area can repeatedly return police to the same area.
- We supplement crime data modeling with non-crime data and exclude people data – We also work to reduce bias by supplementing reported data with multiple sources of relevant data from independent, open sources. Typical examples include seasonality, time of month, day of week, time of day, holidays, upcoming events, weather, and locations of liquor establishments.
- We maximize the reduction of harm – We do not make predictions about the actions of people – that means no arrests, social media, or personal data is used. We limit the time an officer patrols and the occurrence of patrol assignments in the same location to prevent over-patrolling.
- We prioritize oversight and accountability – We log data input used and outputs generated by each model. We also log patrol activities including time, place and tactics used.
- We are proactively transparent – We are committed to being transparent about how our system works and use third parties to provide objective assessments.
Comments are closed.