AI Lie Detector

A new artificial intelligence (AI) algorithm has been developed that can determine whether a robbery report is legitimate or fake, with a high degree of accuracy. Known as VeriPol, the system was developed to deal with a growing number of spurious theft reports that consume valuable police time and resources.

The reasons people file phony robbery complaints are many – some are looking to cash in on insurance claims, while others are simply trying to avoid disclosing the fact that they have lost or damaged some item of value. In any case, catching them requires hours of review by seasoned investigators. But the new AI system – which is based on natural language processing – analyzes a number of features in the text to identify which cases have likely been fabricated.

Historical police reports were fed into VeriPol so that it could “learn” specific patterns that were common among false reports. A number of themes emerged, including: a greater focus on the property than the incident, a lack of precise detail about the incident or the attacker, a lack of witnesses, and a failure to contact authorities immediately following the incident. It was then tested on more than 1,000 reports from the Spanish National Police and compared to analysis by two police experts. The results indicated that VeriPol correctly discriminated between real and false statements with greater than 80 percent accuracy, which was 15 to 20 percent better than the humans.

The new system is already being deployed throughout Spain in the hopes of deterring people from filing fake claims. Although it is currently specific to personal robberies, the developers hope that it will eventually be able to detect inconsistencies in other falsified crime reports, including car theft and home burglary.