Diverse Agents for Ad-Hoc Cooperation in Hanabi

Canaan, Rodrigo; Togelius, Julian; Nealen, Andy; Menzel, Stefan

doi:10.1109/cig.2019.8847944

Cited by 27 publications

(15 citation statements)

References 24 publications

(42 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Liang et al showed that agents who use conversational implications and choose behaviors that emphasize the information given to the turn, and agents who emphasize narrowing the card's potential, are perceived as "human" who tend to feel human (Liang et al, 2019). Canaan et al suggested using a simulation of the play between AI, that the diversity of the player could be correspondent, if 2 parameters of "risk aversion" and "degree of communication" were set appropriately (Canaan et al, 2019). This study verifies whether AI agents with different tendencies of "risk aversion" can be implemented and actually adapted to human diversity.…”

Section: Background Of Hanabi Study: a Unique Testbed For Analyzing Human Cooperationmentioning

confidence: 99%

See 1 more Smart Citation

Emergence of Cooperative Impression With Self-Estimation, Thinking Time, and Concordance of Risk Sensitivity in Playing Hanabi

et al. 2021

View full text Add to dashboard Cite

The authors evaluate the extent to which a user’s impression of an AI agent can be improved by giving the agent the ability of self-estimation, thinking time, and coordination of risk tendency. The authors modified the algorithm of an AI agent in the cooperative game Hanabi to have all of these traits, and investigated the change in the user’s impression by playing with the user. The authors used a self-estimation task to evaluate the effect that the ability to read the intention of a user had on an impression. The authors also show thinking time of an agent influences impression for an agent. The authors also investigated the relationship between the concordance of the risk-taking tendencies of players and agents, the player’s impression of agents, and the game experience. The results of the self-estimation task experiment showed that the more accurate the estimation of the agent’s self, the more likely it is that the partner will perceive humanity, affinity, intelligence, and communication skills in the agent. The authors also found that an agent that changes the length of thinking time according to the priority of action gives the impression that it is smarter than an agent with a normal thinking time when the player notices the difference in thinking time or an agent that randomly changes the thinking time. The result of the experiment regarding concordance of the risk-taking tendency shows that influence player’s impression toward agents. These results suggest that game agent designers can improve the player’s disposition toward an agent and the game experience by adjusting the agent’s self-estimation level, thinking time, and risk-taking tendency according to the player’s personality and inner state during the game.

show abstract

Section: Background Of Hanabi Study: a Unique Testbed For Analyzing Human Cooperationmentioning

confidence: 99%

“…The conventional deterministic strategy is the algorithm proposed by Osawa (Osawa, 2015), which acts with the following priorities. This method is also widely used as a reference in other Hanabi studies (Canaan et al, 2019). Each process is conducted according to Figure 2B.…”

Section: Implementation Of Base Algorithmmentioning

confidence: 99%

Emergence of Cooperative Impression With Self-Estimation, Thinking Time, and Concordance of Risk Sensitivity in Playing Hanabi

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Next to improving the quality of an agent, we are interested in the optimization of diverse agents with unique play-styles. The Multi-dimensional Archive of Phenotypic Elites (MAP-Elites) [9]) is a quality-diversity algorithm that has recently been used to create a diverse set of game-playing agents [22] and game-play elements [23].…”

Section: Map-elitesmentioning

confidence: 99%

Generating Diverse and Competitive Play-Styles for Strategy Games

Peŕez-Liebana¹,

Guerrero-Romero²,

Dockhorn³

et al. 2021

Preprint

View full text Add to dashboard Cite

Designing agents that are able to achieve different play-styles while maintaining a competitive level of play is a difficult task, especially for games for which the research community has not found super-human performance yet, like strategy games. These require the AI to deal with large action spaces, long-term planning and partial observability, among other well-known factors that make decision-making a hard problem. On top of this, achieving distinct play-styles using a general algorithm without reducing playing strength is not trivial. In this paper, we propose Portfolio Monte Carlo Tree Search with Progressive Unpruning for playing a turn-based strategy game (Tribes) and show how it can be parameterized so a qualitydiversity algorithm (MAP-Elites) is used to achieve different playstyles while keeping a competitive level of play. Our results show that this algorithm is capable of achieving these goals even for an extensive collection of game levels beyond those used for training.

show abstract

“…Lastly, there is recent work on ad-hoc team play in Hanabi, in which agents get evaluated against a pool of different teammates (Canaan et al 2019;Walton-Rivers et al 2017).…”

Section: Related Workmentioning

confidence: 99%

Improving Policies via Search in Cooperative Partially Observable Games

Lerer

Foerster

et al. 2020

AAAI

View full text Add to dashboard Cite

Recent superhuman results in games have largely been achieved in a variety of zero-sum settings, such as Go and Poker, in which agents need to compete against others. However, just like humans, real-world AI systems have to coordinate and communicate with other agents in cooperative partially observable environments as well. These settings commonly require participants to both interpret the actions of others and to act in a way that is informative when being interpreted. Those abilities are typically summarized as theory of mind and are seen as crucial for social interactions. In this paper we propose two different search techniques that can be applied to improve an arbitrary agreed-upon policy in a cooperative partially observable game. The first one, single-agent search, effectively converts the problem into a single agent setting by making all but one of the agents play according to the agreed-upon policy. In contrast, in multi-agent search all agents carry out the same common-knowledge search procedure whenever doing so is computationally feasible, and fall back to playing according to the agreed-upon policy otherwise. We prove that these search procedures are theoretically guaranteed to at least maintain the original performance of the agreed-upon policy (up to a bounded approximation error). In the benchmark challenge problem of Hanabi, our search technique greatly improves the performance of every agent we tested and when applied to a policy trained using RL achieves a new state-of-the-art score of 24.61 / 25 in the game, compared to a previous-best of 24.08 / 25.

show abstract

Diverse Agents for Ad-Hoc Cooperation in Hanabi

Cited by 27 publications

References 24 publications

Emergence of Cooperative Impression With Self-Estimation, Thinking Time, and Concordance of Risk Sensitivity in Playing Hanabi

Emergence of Cooperative Impression With Self-Estimation, Thinking Time, and Concordance of Risk Sensitivity in Playing Hanabi

Generating Diverse and Competitive Play-Styles for Strategy Games

Improving Policies via Search in Cooperative Partially Observable Games

Contact Info

Product

Resources

About