Algorithmenbasierte Risikoerkennung

With increasing digitization and technical progress, it seems logical that the state should also increasingly exploit technical possibilities to perform its tasks better or at least more efficiently. One popular topic of the present debate in Germany is the increasing use of algorithms for Predictive Policing, i. e. the data-based prediction of a increased risk of crime associated with a situation or person. But the possible use for a algorithm-based risk detection is not limited to the area of predictive policing. A functionally comparable use of algorithms is also possible in other areas of state activity (e. g. automatic analysis of satellite images to detect illegal constructions or use of risk management systems in the taxation process to assess the risk of tax evasion). However, despite the great practical importance and the existing commonalities, there has not been a cross-cutting approach from a legal perspective to algorithm-based risk detection yet. This contribution is intended to close the gap by providing an in-depth analysis of the use and legal classification of algorithm-based risk detection beyond a specific area of use. The article describes and analyses the basic features of the phenomenon and the constitutional standard against which algorithm-based risk detection used by the state must be measured across all subject areas. The applicable standards are not only presented as the state of legal doctrine, but also critically deconstructed to the extent necessary. Additionally the legal requirements by the European AI‑Act for the use of algorithm based risk detection are analysed. Beyond the individual consideration recently a cumulative approach has been proposed. Some scholars think there should be a limit for state surveillance measures in total. But such an approach relies on an oversimplified concept of freedom and therefore should be rejected. The article concludes that the use of algorithms for risk detection has a considerable potential to improve administrative abilities of risk detection. Such a use can be permissible, as long as it refers to sufficiently error-free algorithms and is accompanied by strict organizational and procedural measures to mitigate the risk of misuse.