by Dr. Nadja Braun Binder German

Research Institute for Public Administration, Project Cluster Transformation of the state in the digital age

Some thoughts with regard to public administration

Just like private individuals and companies, public administration also uses computational methods. The aim is not only to utilise software to help make processes quicker, less complicated and easier to handle but also to support the evaluation of already existing data. Algorithms are used to interpret the collected data and can be found in a variety of different administrative areas. Tax authorities aim to utilise algorithms, for example, for fully automated tax assessments. In future, the tax assessment process will be completed in a fully automated manner, i.e. without the use of human resources, on the basis of data collected from electronic tax returns submitted via a portal, as well as using data transferred electronically by third parties. A risk management system should filter out those tax returns with significant risk, as well as some randomly selected cases, so that they can be processed manually. Although the question of how these algorithm-based risk management systems should be controlled is being discussed in this context, no concrete solutions are yet in sight.

In the German energy transition, there has also been an increasing use of algorithms such as those found in smart meters which have already become obligatory in certain areas. A smart meter makes it possible to evaluate the electricity, water, gas and district heating requirements of a household in real time. The data that is collected and saved during this process could also be evaluated by energy supply companies, including public utilities, with the aid of algorithms. This type of evaluation is necessary in order to operate a smart grid (intelligent electricity grid) that is capable of compensating for the special fluctuations in utilisation through the use of renewable energies. The algorithms analyse real-time data and thus simplify planning for the required utilisation of the grid. The analysis of this data is viewed critically from a data privacy perspective because the algorithms use personal measurement data that make it possible to derive the daily routines of the individual households and to identify the electronic devices found in the household. It is not possible for private households to control the transferred measurement data or the subsequent algorithm-based evaluation. This lack of a control option over the process increases the conflict between digitalisation and privacy.

Something that the aforementioned examples have in common is that the utilised software is or is due to be mostly programmed, implemented and maintained by private companies on behalf of the relevant authorities. The result of this programming and at the heart of this software are algorithms. These algorithms not only help to automatically evaluate raw data but, depending on the algorithm's level of complexity, also to interpret this data. Recent developments in the area of algorithm research also make it possible to develop self-learning algorithms, which optimise their programming themselves in the event of missing or incorrect results. The fact that algorithms are externally programmed means that the authorities can not necessarily comprehend which methodology is being used. The responsible employee only receives a result. The precise process behind the calculation of the results cannot be viewed by either the user nor the government who has commissioned the external programming. And neither is it possible for a citizen affected by an administrative decision to understand how this decision was made. A decision that was based to a large extent or even exclusively on algorithms. This is problematic because, amongst other things, the algorithms also evaluate personal data and a decision is made in individual cases based on this evaluation

Science needs to devote itself to addressing this problematic situation in more detail in the future. In order to implement algorithm-based processes for government, for example, the question of how the legality of this administrative activity can be guaranteed through effective control options needs to be raised. At the same time, it must be asked what control limits can be set on the utilisation of government algorithms, such as if they presuppose the disclosure of personal data. These questions need to be investigated in combination with the issue of how expectations for efficiency and control by the government bodies in their use of IT-based services are perceived. The most suitable solution is an interdisciplinary approach, such as that on which the conference on 10 November 2016 on Computational Methods in Law in Non Common Law Jurisdictions was based!