TY - GEN
T1 - Learning to Deceive in Multi-agent Hidden Role Games
AU - Aitchison, Matthew
AU - Benke, Lyndon
AU - Sweetser, Penny
N1 - Publisher Copyright:
© 2021, Springer Nature Switzerland AG.
PY - 2021
Y1 - 2021
N2 - 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.
AB - 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.
KW - Bayesian belief
KW - Deception
KW - Deep reinforcement learning
KW - Intrinsic motivation
KW - Machine learning
UR - http://www.scopus.com/inward/record.url?scp=85121909505&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-91779-1_5
DO - 10.1007/978-3-030-91779-1_5
M3 - Conference contribution
AN - SCOPUS:85121909505
SN - 9783030917784
T3 - Communications in Computer and Information Science
SP - 55
EP - 75
BT - Deceptive AI - First International Workshop, DeceptECAI 2020 and Second International Workshop, DeceptAI 2021, Proceedings
A2 - Sarkadi, Stefan
A2 - Wright, Benjamin
A2 - Masters, Peta
A2 - McBurney, Peter
PB - Springer Science and Business Media Deutschland GmbH
T2 - 1st 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
Y2 - 19 August 2020 through 19 August 2020
ER -