Legal Aspects and Society
Legal Challenges and Solutions:
Privacy by Design Transparency and
Trustworthy AI Bias Analytics: Machine Learning
and Minority Protection
Legal Challenges and Solutions in Big Data Technologies: Privacy by Design
(Prof. Müller-Mall)
- Motivation
- Big Data and AI technologies play an increasingly and important role in both individual lives and society
- However, the omnipresence of these technologies entails numerous risks with regard to data protection and data security
- Developing technology can therefore not be separated from a further development and investigation of legal framework conditions
- Aim
- Conceptual elaboration of the Privacy by Design-approach, which is among others anchored in the European General Data Protection Regulation
- Identification of eminent tensions between data protection principles and the demands of Big Data/AI technologies
- Inventory and legal evaluation of currently available privacy enhancing technologies
- Modelling of an enhanced Privacy by Design-approach tailored to Big Data/AI requirements
Transparency as a Fundamental Principle of Trustworthy AI: Regulatory Framework and Challenges (Prof. Lauber-Rönsberg)
- Ethics Guidelines for Trustworthy AI:
- Transparency is a key requirement that AI systems should meet in order to be deemed trustworthy.
- Implementation: Humans need to be aware that they are interacting with an AI system; transparency of the data, the system’s capabilities and AI business models; explainable AI.
- Legal Framework:
- Data Protection Laws: Scope and efficiency of information obligations? Specific obligations for automated decision- making-systems about the “logic involved”?
- Fair Trading and Consumer Protection Laws: Data-driven marketing tools are shaping decision-making architectures, thus influencing consumers’ decisions. Legal Regulation of subliminal marketing practices?
Bias Analytics: Machine Learning and Minority Protection (Jun.-Prof. Martin Potthast)
This project studies the analysis and the management of bias in sociotechnic systems that employ machine learning as a tool.
Case Study 1: Vandalism on Wikidata
The damage control system of Wikidata is biased against anonymous editors and newcomers.
Case Study 2: The Direct Answer Dilemma
- Conversational search interfaces are “narrow”
- Users are impatient
- There is only room for one or two search results
- Questions need to be answered directly
- Direct answers
- A direct answer is not necessarily correct
- Users still presume correctness
- This presents a strong bias against diversity
- The dilemma: fast answers vs. accurate ones
