Skip to Main content Skip to Navigation

Raisonner sur la manipulation dans les systèmes multi-agents : une approche fondée sur les logiques modales

Christopher Leturc 1
1 Equipe MAD - Laboratoire GREYC - UMR6072
GREYC - Groupe de Recherche en Informatique, Image, Automatique et Instrumentation de Caen
Abstract : In recent decades, computer development has shifted from designing individual software to designing intelligent, self-contained software called agents and interacting with others to form multi-agent systems. In such systems, malicious agents sometimes implement complex strategies to induce other agents to make decisions in their favor, without the latter noticing them. We are talking about manipulation strategies. These strategies may in some cases cause problems for the agents which are victims. Such strategies are always hidden from agents and therefore hard to detect. How detect them and better fight them? Firstly, it is necessary to define manipulation. Thus, based on work from computer sciences and social sciences, we define manipulation as the deliberate intention of an agent to instrumentalize a victim while making sure to conceal that intent. We propose to answer this question, a logical system named KBE which expresses manipulation. We prove that KBE is correct and complete, and is able to express strategies based on knowledge and beliefs of agents, like lying or bullshiting. This system can also express notions such as coercion and persuasion. Secondly, since trust is a mechanism to regulate the interactions between agents when agents may be malicious or unreliable, we propose another logical system named TB. This system, proved to be correct and complete, expresses a notion of trust in sincerity which represents the choice of an agent to take the risk of believing another agent for its sincerity. Finally, we propose an algorithmic method to reason with such systems. This method is adapted to the TB system and decides on its satisfiability problem by directly using the constraints of the framework to build a model.
Document type :
Complete list of metadatas

Cited literature [143 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Thursday, February 6, 2020 - 12:19:09 PM
Last modification on : Tuesday, February 11, 2020 - 1:38:21 AM
Long-term archiving on: : Thursday, May 7, 2020 - 1:45:18 PM


Version validated by the jury (STAR)


  • HAL Id : tel-02469022, version 1


Christopher Leturc. Raisonner sur la manipulation dans les systèmes multi-agents : une approche fondée sur les logiques modales. Intelligence artificielle [cs.AI]. Normandie Université, 2019. Français. ⟨NNT : 2019NORMC236⟩. ⟨tel-02469022⟩



Record views


Files downloads