Organization

Generalise the applicability of the AJ Jiminy architecture for AI ethics. Transform the Autonomous Jiminy moral advisor into an interactive Dialogue Jiminy (DJ) by replacing argumentation as inference with argumentation as dialogue. Create a communication language and a protocol for persuasion dialogues on moral dilemmas. Study strategic aspects of these dialogues for the stakeholder avatars participating in these dialogues.

Create a natural languageinterface between the Dialogue Jiminy and the stakeholders (or the general public). Use machine learning to construct two language modules. First, use NLP to transform the stakeholders’ informal norms into the avatars’ formal rules. Secondly, use natural language generation to synthesise the outputs of AJ (argument extensions) or DJ (formal dialogues) into explanations in plain English of why a particular decision was passed as the moral recommendation to the agent.

Integrate the DJ dialogue system and the natural language interface. Integrate the AJ argumentative system and the natural language interface. Build demonstrators of AJ and DJ. Evaluate these systems using case studies involving autonomous cars and food recommender systems from our INTEGRAUTO and EXPECTATION projects.