AI Ethics Under Scrutiny

AI ethics has come under increasing criticism. Critics argue that while companies publish guidelines for AI development and use, the AI systems themselves often remain problematic. Researchers from the University of Bayreuth, Saarland University and the CISPA Helmholtz Center for Information Security demonstrate in a new publication that debates on AI ethics are frequently shaped by a fundamental misunderstanding.

“The term AI ethics is used to cover a wide range of activities that pursue different objectives. The question should therefore not be ‘Is AI ethics useful or not?’ but rather: Which type of AI ethics is useful for  which purpose?” says Dr. Astrid Schomäcker, Research Associate at the Philosophy, Computer Science and Artificial Intelligence research group at the University of Bayreuth and lead author of the publication.

Alongside Schomäcker, Bayreuth researchers Luca Deck, Dr. Timo Speith and Prof. Dr. Lena Kästner also contributed to the publication. All four are members of the recently established AI Responsibility Centre at the University of Bayreuth.

In their publication, the researchers identify three fundamental objectives of AI ethics: generating knowledge about the moral, social, and normative aspects of AI, communicating that knowledge, and ultimately applying it in practice to develop and deploy AI systems responsibly. The publication also identifies four key domains in which AI ethics operates: research, education, the practical development of AI systems, and governance mechanisms such as ethics boards, guidelines and legislation.

“In this publication, we do not defend AI ethics from any single perspective. Rather, we provide an overview of the various ways in which AI ethics can be valuable, while keeping existing problems firmly in view. This enables us to identify appropriate ethical approaches for different purposes and to combine them in the most beneficial way possible,” says Timo Speith.

One of the publication’s central findings is that criticism of particular instruments should not automatically call the entire field of AI ethics into question. For example, it may be justified to argue that certain ethical guidelines have only a limited influence on the actual development of AI systems. However, this does not mean that research, education or other governance mechanisms are equally ineffective.

Above all, AI ethics must not be reduced solely to legislation or replaced by market mechanisms. Research and public debate are necessary in order to determine which rules and values should govern the development and use of AI in the first place. Moreover, ethics committees and academic assessments can often respond more quickly to new technological developments than legislative processes.

“AI ethics is neither a panacea nor meaningless. Its value depends on the objective being pursued and the instrument being employed. Responsible AI development requires different forms of AI ethics that complement one another,” says Prof. Dr. Lena Kästner, Chair of Philosophy, Computer Science and Artificial Intelligence at the University of Bayreuth.

The publication was funded by the Volkswagen Foundation, the German Research Foundation (DFG), and the Bavarian Research Institute for Digital Transformation (bidt).