Will it soon be possible to detect a lie in a handwritten message? In Spain, since 2017, the police have been testing a new artificial intelligence program called VeriPol, able to detect false complaints according to various criteria. Having been put into service for just over a year, the program has had a success rate of 83%.
First tested in the cities of Murcia and Malaga, VeriPol was developed by researchers from Cardiff University and Charles III University in Madrid. The latter created the program's algorithm by going through more than 1,000 written reports of violent or threatening robberies provided by the Spanish National Police.
The sample contained 534 true complaints and 588 false ones, which the program had to learn to distinguish by analyzing several elements, such as punctuation, the size of the text, and also the length of the complaints or their content.
AI learned, for example, that lack of detail about the incident or thieves was a common feature of many false complaints, as was the absence of witnesses or evidence. Complaints that place undue importance on stolen property are also considered suspicious.
It's a system that can have its limits but has had a convincing success rate in its first year of existence. Jose Camacho-Collados, co-author of the study, believes that this work 'gives fascinating information on how people lie to the police, and provides a tool that can one day be used to dissuade people from doing so to the police.'