Reasoning of a cognitive agent - collective and ethical aspect
We are interested here in modelling different aspects of the reasoning of cognitive agents, that is, agents that explicitly store some knowledge that they can manipulate to maintain their adequacy and derive new conclusions, predictions or decisions. We consider symbolic approaches, where explicit knowledge is represented in some kind of logical form, which can vary from propositional logic to non-monotonic ones or classical first order predicate logic.
Using this knowledge, a cognitive agent can perform different kinds of reasoning which can be turned towards maintaining or improving its representation of the world, adapting its beliefs to external influences or assessing different alternatives in a decision process. The forms of reasoning we study here include different forms of inferences across the deductive, abductive or inductive spectrum. Knowledge representation and inference mechanisms are at the core of this works. In exploring this large domain, we will particularly study two aspects of an agent's reasoning, namely the collective and ethical aspects.
First, we focus on the collective aspect and thus consider a society of communicating cognitive agents. Its interactions with other agents can influence an agent on multiple levels and we will focus on purely communicative interactions in our work, staying on an epistemic level where agents only affect each other by the exchange of information. Given that, we will consider two cases depending on the kind of knowledge and interactions involved. In the first case, ''collective reasoning'', knowledge is distributed but ultimately consistent : the agents interact to share information in order to derive some conclusions from the sum of all the knowledge in the system. In the second case, focused on influence mechanisms, we consider beliefs rather than certain knowledge. The beliefs of the agents in the system are typically not consistent with one another and agents will interact by means of influence mechanisms. Rather than a concerted effort to reach a conclusion, we will study the evolution of the system resulting from these mutual influences. More precisely, collective reasoning study how a cognitive agent can interact with other agents in order to benefit from their additional knowledge and enrich its local reasoning. Considering the whole society, collective reasoning is seen as a collaboration between the agents to achieve some reasoning task together, typically in the forms of local reasoning and information exchanges ensuring that the knowledge of the whole society is taken into account. Considering a different stance on this collective aspect, We then investigate beliefs and consider influence mechanisms within the society of agents. We are then concerned with modeling such influence processes in a formal and rational way and determining how these processes and the structure of the relations between the agents affect the evolution of their beliefs. We present our model for representing and studying the evolution of beliefs under influence : belief revision games.
The second important aspect studied in this work belongs to the fields of computational ethics. We investigate the modelisation of ethical reasoning, identifying the key components that are needed to explicitly represent ethical principles. We present a general framework for ethical reasoning based on a representation of action and change in some version of discrete event calculus integrating some model of causality, before focusing on the core problem of representing ethical principles and assessing ethical permissibility of a given action, presenting formalisation of different ethical principles in a unified way.
Defence : 01/05/2021 - 14h - https://zoom.us/j/94130113373?pwd=Y2FwdHcxakJrWDZUTEcrV2hWaEI3dz09
Jury members :
Jérôme Lang, Université Paris Dauphine [rapporteur]
Emiliano Lorini, Université Paul-Sabatier, Toulouse [rapporteur]
Alessandra Russo, Imperial college, London [rapporteur]
Thomas Agotnes, University of Bergen
Aurélie Beynier, Sorbonne Université, Paris
Torsten Schaub, University of Postdam
Laurent Simon, Université de Bordeaux
Current position : - Sorbonne Université