The Federal Commissioner for Data Protection and Freedom of Information

Navigation and service

The BfDI’s Symposium: Experts shed Light on Artificial Intelligence and Data Protection

Bonn/Berlin, 01 October 2019

The current technology debate is mainly shaped by the concepts of artificial intelligence (AI) and of neural networks. They usually require a large number of data, which are often personal data. In order to shed more light on the resulting problems from different angles, more than 150 guests participated in the BfDI’s Symposium on “Chances and Risks for the privacy-friendly Use of Artificial Intelligence” on 24 September 2019 in the Harnack House of the Max Planck Society in Berlin.

In keeping with the venue, the Federal Commissioner for Data Protection and Freedom of Information quoted Max Planck in his opening speech saying that "Insight must precede application" in order to highlight one of the major challenges in the topic of AI. Ulrich Kelber explained: However, particularly in the case of self-learning systems, this is often not so easy to implement.  It is not rare that in such cases, we are faced with a black box. In this respect, it is true that it is difficult to assess AI-systems from the perspective of data protection law,  but in my view, it is fundamentally wrong to think that data protection and AI exclude each other. AI must not only aim at innovation, but also at transparency and fairness. Data protection makes an important contribution to this objective.

This last point was taken up by Professor Dr. Dieter Kugelmann, who in his function as the current Chairman of the Conference of the independent Federal and State Data Protection Authorities (German abbreviation “DSK”) presented the Hambach Declaration on Artificial Intelligence adopted by that Conference.

Many of the other speakers could contribute much practical experience with current AI applications. First of all, Jens Redmer spoke about the experience at Google. Over the past three years, AI has played a key role in making a decisive impact on the user interface and on language and translation software. However, the process was far from being completed because algorithms would never terminate when doing machine learning.

Dr. Alexander Schellinger reported on pilot applications at the Techniker Krankenkasse (health insurance), which, for example, allow a genome-based decision-support in cancer treatment. He was convinced that AI applications will play an essential role in the health sector in the future, as their strength lies in pattern recognition, which could vitally improve diagnosis. In addition, patients ask for new products, which, as soon as they are marketable, improve the treatment.

However, Jürgen Bönninger from FSD Fahrzeugsystemdaten GmbH revealed that good AI can also do without any personal data. On the basis of a government mandate, the company receives vehicle data from all manufacturers, and from all providers of general vehicle inspections. On this basis, the company developed software by means of AI which, for example, detects defective shock absorbers with high accuracy. Despite the FSD’s motto “We do not want any reference to persons in our data”, manufacturers often simply provide personal data. These “dirty data sets” would then have to be cleaned up first.

Prof. Dr. Sabine Sachweh addressed the AI’s acceptance problems. She explained that the impact of a non-understandable response can undermine confidence in the deep neural networks. For this reason, review processes are urgently necessary. In her view, the successful introduction of AI in Europe could be achieved by ensuring trust, transparency and visibility in the field of AI. In this connection, it is recommended not to take the USA and China as an example, but to strive for a solution of our own European problems. 

In the following contribution, Iris Plöger explained that AI applications are not yet widely used in German companies. Small and medium-sized companies in particular require support in order to achieve concrete improvements. However, the General Data Protection Regulation would remain an impediment to development. It is precisely the principle of a ban with permit reservation that is outdated in the AI-era and no longer tenable.

During the final panel discussion, the different topics were brought together, and ideas how the compatibility of artificial intelligence and data protection could succeed were being discussed. In this connection, it is one important point that in Europe, it is intended to create decentralised models of AI, which, in their function as counter-poles to the central approaches of China and the USA, are of interest in particular for the European economy.

At the end of the day, all the participants could look back at a successful event.  The BfDI summed up by saying: The day has shown how broad in scope and how complex the topic “AI” actually is, and he announced that he would like to continue the exchange in a further symposium in the future.