Research Publications

The field of General Game Playing (GGP) usually yearns to create agents that are able to win any game as efficiently as possible. Utility functions are then easy to find: winning the game, or achieving a better score, nets you better results. This resulted in game playing becoming a testbed for artificial intelligence. New techniques are often tested on games such as Chess, Go or Checkers where the environments are easy to formally define, while the task is still being tied closely to human intelligence. Superhuman-level has been achieved in a great number of games past the 2010s. The first notable instance was Chess world champion Garry Kasparov losing to DeepBlue, a chess playing computer, in 1997 during a six-game match. Researchers became interested in games where heuristics were harder to encode, such as Go. Only in 2016 did AlphaGo become the first computer Go program to beat a 9-dan professional player, Lee Sedol. It did so using novel techniques, such as Deep Neural Networks (DNN) trained using self-play to evaluate positions and Monte-Carlo Tree Search (MCTS) to search the game tree, which are now staples of superhuman-level game playing agents. Such level of play was only achieved in agents playing specific games. In such instances, an optimized game representation and evaluation functions can be built by leveraging domain-specific knowledge. General Game Playing (GGP) aims to generalize game playing agents so that they can pick up any game. In this more general setting, domain-specific knowledge can only be used scarcely, leading to more polyvalent yet weaker a...

Automatically generating novel and interesting games is a complex task. Challenges include representing game rules in a computationally workable form, searching through the large space of potential games under most such representations, and accurately evaluating the originality and quality of previously unseen games. Prior work in automated game generation has largely focused on relatively restricted rule representations and relied on domain-specific heuristics. In this work, we explore the generation of novel games in the comparatively expansive Ludii game description language, which encodes the rules of over 1000 board games in a variety of styles and modes of play. We draw inspiration from recent advances in large language models and evolutionary computation in order to train a model that intelligently mutates and recombines games and mechanics expressed as code. We demonstrate both quantitatively and qualitatively that our approach is capable of generating new and interesting games, including in regions of the potential rules space not covered by existing games in the Ludii dataset.

Imperfect-information games present significant challenges for General Game Playing (GGP) agents. Traditional models have limitations that hinder their applicability in this domain. These models require agents to construct and maintain estimates about the game state, a process that is often game-specific and can unintentionally introduce domain-specific knowledge. Furthermore, this specificity undermines the core goal of GGP to generate domain-independent strategies. To overcome these challenges, we propose the Belief Stochastic Game model. This novel framework shifts the responsibility of state estimation from the agent to the game model itself, allowing agents to focus solely on strategy development. This externalisation of the state estimation process is enabled by the exploitation of the common structures found in many imperfect-information games. The new model facilitates the development of more general agents that can adapt to a wide range of games.

Étienne de Flacourt's Histoire de la grande isle Madagascar is renowned for its detailed descriptions of Madagascar's flora, fauna, and local culture. The posthumously published second edition (1661) includes the earliest known record of a four-row mancala game, accompanied by a drawing and a detailed explanation using local terminology. These rules, resembling the modern game of Bao (known as Katra-be in Madagascar), suggest a long history of four-row mancala games, despite a lack of archaeological evidence. As governor of a French trading post in Fort Dauphin, de Flacourt's work endures, offering a glimpse into 17th-century Madagascar.

Games, as a subject of study, span diverse disciplines, often in isolation. In computer science and mathematics, games stand as testbeds for pioneering methodologies, shaping state-of-the-art techniques in economics, engineering, and AI. Simultaneously, archaeologists, historians, and anthropologists explore the motivations underlying human play, dissecting its social implications on both individual and societal levels. Games have not only been instrumental in pedagogical development but are gaining recognition as integral components of humanity's intangible cultural heritage. However, this recognition is tinged with a poignant realization: much of the world's game heritage has been lost, overshadowed by colonialism, imperialism, and commercialization. Against this backdrop, the paper introduces the GameTable Network, an EU-funded COST Action overcoming disciplinary divides. The network's research integrates AI techniques, archaeology, and gaming research, into a cohesive framework. In doing so, the network aspires to rekindle the essence of games not only as mere entertainment but also as repositories of cultural richness and historical narratives. The aim is not only to uncover the mysteries of the past but to acknowledge and rectify the gaps hindering a holistic understanding of games. Our objectives span from reconstructing historical games to advancing AI methodologies, from fostering educational programs to embracing community involvement in digital initiatives. Key challenges include innovative approaches for studying, reconstructing, and preserving heritage games, reconstructing missing rules in incomplete games, simulating play at a human level, applying AI to study historical games, developing novel methods to identify unrecognized games measuring their evolution across space and time, developing pedagogical tools, and leveraging GameAI in a culturally diverse way, all crucially contributing to the study of heritage games. Through this interdisciplinary lens, our aspiration is to enrich scholarly dialogue within the Eastern Mediterranean, Europe and beyond, advancing our collective understanding of the historical and cultural dimensions embedded in games through AI-driven research.

The inaugural in-person meeting for the "GameTable" COST Action’s Working Group 1(WG1) on Search, Planning, Learning,and Explainability took place at the Leiden Institute of Advanced Computer Science (LIACS) on January 31st,2024. The primary aims of this meeting were to facilitate talks and discussions on, and connect researchers interested in, three core research goals: (1) human-like game-playing AI, (2) imperfect-information games within a general game playing context, and (3) explainable search and reinforcement learning in games. This report provides a summary of the discussions and talks that took place during the meeting.

Games with imperfect information represent a major challenge for General Game Playing (GGP) agents. To address this, we propose to extend Ludii, a state-of-the-art GGP system, by improving the game description to include hidden information, adopting a formalism and modelling closer to reinforcement learning, and adapting the search algorithms to make better use of the game description. These improvements will make it possible to simulate and analyse a wide range of games with imperfect information.

The GameTable COST Action kickoff, focusing on “Computational Techniques for Tabletop Games Heritage,” took place at Leiden University in the Pieter de la Court Building from January 29th to 30th, 2024. This event aimed to convene researchers from diverse backgrounds involved in the Action, offering an opportunity to present an overview of the key research areas, share concrete case studies, and facilitate discussions and idea exchanges across fields that may not typically intersect, thereby enhancing the organization of the Action. This report provides a summary of the organization and discussions of the event, and future plans for GameTable.