Artificial intelligence can recognize and differentiate between child and youth pornography and non-criminal adult pornography with an accuracy of over 90 percent. This is what North Rhine-Westphalia’s Justice Minister Peter Biesenbach (CDU) reported in Düsseldorf on the results of a research project with Microsoft that was started around a year ago.
The specific image content is anonymized on local computers so that neither persons nor criminally relevant child pornographic content can be identified in the files. Then the material is sent to the cloud, where algorithms are supposed to recognize content in the deconstructed recordings.
An AI algorithm pre-sorts the deconstructed image files and classifies them according to four categories: whether the images show criminal depictions of abuse by children or adolescents, whether it is permitted adult pornography or other images, explains Microsoft.
Constant AI training
So far, the AI has been able to correctly categorize images 92 percent of the time. Suspicious images can thus be processed with priority. In the event of incorrect results, investigators can manually flag images retrospectively. This shows an advantage of centralized technology: the assessments of all experts flow into the constant training of the AI; the more feedback the system gets, the more the AI learns.
In addition to the classification, printed or handwritten texts can also be recognized directly on the images via OCR and automatically compared with keyword lists. For example, watermarks, such as those used by perpetrators, or chat histories can be evaluated more quickly.
The prototype developed together with scientists and Microsoft does not replace human evaluators and legal evaluators, said Biesenbach. At a very early stage of the investigation, however, he could quickly and effectively filter out the evidence from the amount of data that the criminal prosecutors needed to examine the urgent suspicion required for pre-trial detention. This will revolutionize the work of public prosecutors, said Biesenbach.
Unique in the world
The research project with Microsoft’s technology started in August 2019. The hybrid cloud model used is so far “unique in the world” in law enforcement. Ultimately, only a mains connection and a power plug are necessary for the evaluation, said Chief Public Prosecutor Markus Hartmann, the head of the central and contact point Cybercrime North Rhine-Westphalia (ZAC NRW). Data protection and data security are guaranteed. The evidence is only under the control of the law enforcement authorities. External would have no access.
Without artificial intelligence, the sheer flood of data in the field of child pornography and child abuse could no longer be mastered, said the Minister of Justice. A task force set up for this area in mid-2020 has so far led to more than 1,600 preliminary proceedings against more than 1,800 suspects.
The reason for the research project were large abuse complexes with an investigation starting point in Lüdge in Lippe, in Münster and Bergisch Gladbach.