Author: Jochen L. Leidner, Vassilis Plachouras
Publisher: ACL Anthology
Publication Year: 2017
Summary: The following article features Jochen L. Leidner and Vassilis Plachouras who are Data Scientists at Thomson Reuters, Research & Development located in the United Kingdom. Their article suggests a framework called “ethical by design” to ensure ethical standards on data science projects, particularly natural language processing (NLP) projects. The first principle is “Proactive not reactive”: By planning to do things in an ethical way we avoid having to react remedially to non-ethical situations more often than without a planning approach. The second principle is “ethical as the default setting”: By making a commitment to pursuing originally ethical paths, we create alignment within organizations towards a more streamlined set of options that comply with common values. The third principle is “ethics embedded into the process”: A process firmly inclusive of ethics at all stages and levels is less likely to create accidental harm. The fourth principle is “end-to-end ethics”: Ethics cannot be confined to a stage; it must be an all-encompassing property of a process from basic research over product design to delivery. The fifth principle is “visibility and transparency”: A process that is published can be scrutinized, criticized and ultimately improved by a caring community. The sixth principle is “respect for user values”: Whatever values a research institute, university or company may hold is one thing, being user-centric means to also consider the values of the user. I believe applying this framework can go a long way for ensuring companies and institutions adhere to a higher standard of ethics.