Report on the IMKR Lecture “Artificial Intelligence and Speech Law – On Liability for Autonomous Communication Systems” by PD Dr. Oster

On November 6, 2019, for the first lecture in the IMKR series of the winter semester 2019/20, Prof. Rösler welcomed Priv.-Doz. Dr. Jan Oster, LL.M. (Berkeley), Assistant Professor of EU Law and Institutions at Leiden University (Netherlands). Dr. Oster, who is teaching the course “Law of Verbal and Visual Reporting” at the Faculty this semester, spoke on “Artificial Intelligence and Speech Law – On Liability for Autonomous Communication Systems.”

To begin, Dr. Oster referred to recent papers and guidelines on the topic of “Artificial Intelligence” (AI), such as those from the European Parliament (Resolution 2015/2103(INL)), the Council of Europe (Unboxing AI), and a high-level expert group of the European Commission (Ethics Guidelines for Trustworthy AI). In his view, many statements were based on a flawed approach. They failed to recognize that the manifestations of AI are too diverse to draw conclusions for individual cases based on general considerations. Instead, Dr. Oster distinguished between “strong” and “weak” AI. Strong AI (or human-level AI) possesses the ability to solve any cognitive task at least as well as humans. Currently, however, only weak AI (narrow AI) exists. This is used only within a specific and limited scope of application, such as to drive a car or play chess.

Dr. Oster therefore focused on artificial intelligence that generates information and communicates with humans. In this context, Dr. Oster cited the search suggestion feature of internet search engines, chatbots such as Microsoft’s Tay and Xiaoice, as well as social bots that participate in discussions on social networks. Dr. Oster examined tort liability for violations of personality rights caused by communication AI under current law (de lege lata) and proposed law (de lege ferenda). He argued that current legal doctrine is fundamentally well-equipped to address the challenges posed by communication AI. Dr. Oster rejected both the concept of strict liability and the idea of legal capacity for communication AI. In his view, there is no need for this in legal transactions.

Using the Federal Court of Justice (BGH) decision on Google’s search suggestion feature (NJW 2013, 2348) as a basis, Dr. Oster discussed the issues surrounding communication AI. Dr. Oster criticized the BGH’s assessment that Google’s search suggestion feature implied a factual connection between the search term and the suggested term. With its decision, the BGH elevates an algorithmic-quantitative probability calculation to an anthropogenic-qualitative conclusion. Search engines, however, are based not on an editorial but on an algorithmic structuring of information. The BGH reveals an overly pessimistic assessment of users’ media literacy.

Dr. Oster then explained the extent to which a communication AI can “commit” an infringement of personality rights with adequate causation, since it regularly makes autonomous decisions – that is, decisions that are unforeseeable even to the programmer. In this regard, a distinction must be made between infringements of legal interests that are already inherent in the algorithm itself (so-called “algorithm bias”), and those in which the AI has autonomously – and thus unpredictably – made the specific statement that gives rise to the infringement, particularly because it has “learned” the necessary information. While in the first category, a violation of personality rights can generally be affirmed without issue, in the second category it should depend on whether the AI operator has breached a duty of care. In this regard, Dr. Oster advocated for a reversal of the burden of proof in favor of the affected party.

Finally, Dr. Oster addressed data protection. Since personal data can also constitute data that infringes on personality rights, data protection claims – such as the “right to be forgotten” under Article 17 of the GDPR – have become an important tool for protecting personality rights as well. This applies in particular to communication AI with its “data-centric,” rather than “human-centric,” communication.