An image is generated by a surveillance system in a conflict zone. A satellite 36,000 kilometres from the earth can capture a person’s face. The photograph is sent in real time, and in encrypted format, to an NSA data centre, which receives 604800 terabytes of information a day. This image taken by satellite crosses paths with e-mails, Internet traffic, phone calls, biometric profiles, geolocation. To relate these different elements, a semi-supervised machine learning system is used.

//TAR: User is friend of X XXXXXXXXX is supplying to Garúa.

A Cascade algorithm processes information. It puts to one side the data not dealing with sensitive subjects and launches alerts whenever words like Taliban, Leader or Bomb appear in the messages of a targeted individual. Cross-referencing of this information leads to the creation of a target who is then monitored in fifteen-minute periods. It is considered that the person in the photo taken by the satellite has sent far too many messages containing these keywords and should not have been in the place where they were when the photo was taken. They become suspect even though no specialist has ever checked that the satellite facial recognition system has really taken the photograph of a human being and not a face form that appeared on the earth, in the form of a cloud or a reflection on a lake.