Learning to Deceive in Multi-agent Hidden Role Games

Matthew Aitchison*, Lyndon Benke, Penny Sweetser

*Corresponding author for this work

    Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

    1 Citation (Scopus)

    Abstract

    Deception is prevalent in human social settings. However, studies into the effect of deception on reinforcement learning algorithms have been limited to simplistic settings, restricting their applicability to complex real-world problems. This paper addresses this by introducing a new mixed competitive-cooperative multi-agent reinforcement learning (MARL) environment, inspired by popular role-based deception games such as Werewolf, Avalon, and Among Us. The environment’s unique challenge lies in the necessity to cooperate with other agents despite not knowing if they are friend or foe. Furthermore, we introduce a model of deception which we call Bayesian belief manipulation (BBM) and demonstrate its effectiveness at deceiving other agents in this environment, while also increasing the deceiving agent’s performance.

    Original languageEnglish
    Title of host publicationDeceptive AI - First International Workshop, DeceptECAI 2020 and Second International Workshop, DeceptAI 2021, Proceedings
    EditorsStefan Sarkadi, Benjamin Wright, Peta Masters, Peter McBurney
    PublisherSpringer Science and Business Media Deutschland GmbH
    Pages55-75
    Number of pages21
    ISBN (Print)9783030917784
    DOIs
    Publication statusPublished - 2021
    Event1st International Workshop on Deceptive AI, DeceptECAI 2020, held in conjunction with 24th European Conference on Artificial Intelligence, ECAI 2020 and 2nd International Workshop on Deceptive AI, DeceptAI 2021, held in conjunction with 30th International Joint Conference on Artificial Intelligence, IJCAI 2021 - Virtual Online
    Duration: 19 Aug 202019 Aug 2020

    Publication series

    NameCommunications in Computer and Information Science
    Volume1296 CCIS
    ISSN (Print)1865-0929
    ISSN (Electronic)1865-0937

    Conference

    Conference1st International Workshop on Deceptive AI, DeceptECAI 2020, held in conjunction with 24th European Conference on Artificial Intelligence, ECAI 2020 and 2nd International Workshop on Deceptive AI, DeceptAI 2021, held in conjunction with 30th International Joint Conference on Artificial Intelligence, IJCAI 2021
    CityVirtual Online
    Period19/08/2019/08/20

    Fingerprint

    Dive into the research topics of 'Learning to Deceive in Multi-agent Hidden Role Games'. Together they form a unique fingerprint.

    Cite this