VeriPol: Spotting false reports made to the police

VeriPol was created by an international team of researchers in order to help police identify fraudulent reports. The system uses natural language processing and machine learning to analyze a call and predict the likelihood that the report is fake. It focuses its analysis on three critical variables within a complaint: the modus operandi of the aggression, the morphosyntax (grammatical and logical composition) of the report, and the amount of detail given by the caller (Objetivo Castilla-La Mancha Noticias 2018). The system learns by being exposed to two different data sets, one formed by false complaints and the other formed by regular complaints (Pérez Colomé 2018). The system has been tested on over one thousand occasions since 2015 by the Spanish National Police and has earned a 91% accuracy rate. VeriPol was made in response to a recent increase of fabricated reports of violent robberies, intimidation, and theft (Peiró 2019). Researchers claim that this tool could help improve the efficiency of law enforcement and also discourage false complaints (Quijano-Sánchez et al. 2018).

Although the system appears effective on paper, potential discriminatory outcomes should still be assessed. For example, the assessment of the morphosyntax of a call could lead to unforeseen discriminatory outcomes. Since VeriPol analyzes a call using natural language processing and uses coherence and grammatical structure as a strongly-considered variable, the algorithm could be more likely to report a call from a less-educated or non-native caller as a fabricated report (thus, making a false-positive). Inquiries such as this ought to be investigated in order to safeguard vulnerable populations against automated discrimination.