Debating Ethics: Using Natural Language Datasets to Support Human and AI debate

Artificial intelligence (AI) algorithms, including machine learning and deep learning techniques, have been applied with success to a plethora of natural language datasets and tasks, including argumentative text. The goal of this PhD project is to develop novel natural language datasets and AI algorithms that can support transparent reasoning in the form of philosophical debates on ethical and moral issues in society.

A key area of research in argumentation draws on philosophical work that focuses on the identification and specification of schemes and critical questions. The argument schemes and critical questions provide generic templates that characterise stereotypical forms of arguments and challenges that can be posed to these arguments. As an example, the variables in an argument scheme for action (i.e. “In circumstances S, we should do action A to achieve goal G which will promote value V”) can be instantiated with natural language texts to yield an argument X for a specific course of action. In addition, critical questions associated with a given scheme characterise stereotypical ways to challenge any given argument generated by instantiating the scheme; this is done through the construction of arguments that themselves are yielded by instantiation of a scheme. For example, an argument instantiating the expert opinion scheme might challenge the assumption that the circumstances are indeed as stated in an argument for a given action. The instantiation of schemes and use of critical questions thus yield arguments and counter-arguments that can be organised into an argumentation graph which can be used to determine winning arguments.

This PhD project will explore how debates about ethical/moral issues can help define semi-automated approaches for the identification and extraction of selected argument schemes from natural language texts. This will lead to the creation of a new corpus that can be used to automatically mine arguments from texts. The argument schemes and argument graphs obtained from text can then be utilised to support dialogical exchanges between humans and AI systems, supporting transparent and rational reasoning. This has the potential to guide both humans and AI systems in constructing and challenging arguments related to controversial ethical and moral issues in society.

A selection of relevant references: