Game playing has long served as a fundamental benchmark for evaluating Artificial General Intelligence (AGI). While Large Language Models (LLMs) have demonstrated impressive capabilities in general reasoning, their effectiveness in spatial strategic reasoning, which is critical for complex and fully observable board games, remains insufficiently explored. In this work, we adopt Chinese Chess (Xiangqi) as a challenging and rich testbed due to its intricate rules and spatial complexity. To advance LLMs' strategic competence in such environments, we propose a training framework tailored to Xiangqi, built upon a large-scale dataset of five million board-move pairs enhanced with expert annotations and engine evaluations. Building on this foundation, we introduce Xiangqi-R1, a 7B-parameter model trained in multi-stage manner: (1) fine-tuning for legal move prediction to capture basic spatial rules, (2) incorporating strategic annotations to improve decision-making, and (3) applying reinforcement learning via Group Relative Policy Optimization (GRPO) with multi-dimensional reward signals to enhance reasoning stability. Our Experimental results indicate that, despite their size and power, general-purpose LLMs struggle to achieve satisfactory performance in these tasks. Compared to general-purpose LLMs, Xiangqi-R1 greatly advances with an 18% rise in move legality and a 22% boost in analysis accuracy. Our results point to a promising path for creating general strategic intelligence in spatially complex areas.
Deep Reinforcement Learning (DRL) has achieved remarkable success in complex sequential decision-making tasks, such as playing Atari 2600 games and mastering board games. A critical yet underexplored aspect of DRL is the temporal scale of action execution. We propose a novel paradigm that integrates contextual bandits with DRL to adaptively select action durations, enhancing policy flexibility and computational efficiency. Our approach augments a Deep Q-Network (DQN) with a contextual bandit module that learns to choose optimal action repetition rates based on state contexts. Experiments on Atari 2600 games demonstrate significant performance improvements over static duration baselines, highlighting the efficacy of adaptive temporal abstractions in DRL. This paradigm offers a scalable solution for real-time applications like gaming and robotics, where dynamic action durations are critical.
Large language models (LLMs) memorize a vast amount of prior knowledge from the Internet that help them on downstream tasks but also may notoriously sway their outputs towards wrong or biased answers. In this work, we test how the knowledge about popular subjects hurt the accuracy of vision language models (VLMs) on standard, objective visual tasks of counting and identification. We find that state-of-the-art VLMs are strongly biased (e.g, unable to recognize a fourth stripe has been added to a 3-stripe Adidas logo) scoring an average of 17.05% accuracy in counting (e.g., counting stripes in an Adidas-like logo) across 7 diverse domains from animals, logos, chess, board games, optical illusions, to patterned grids. Insert text (e.g., "Adidas") describing the subject name into the counterfactual image further decreases VLM accuracy. The biases in VLMs are so strong that instructing them to double-check their results or rely exclusively on image details to answer improves counting accuracy by only +2 points, on average. Our work presents an interesting failure mode in VLMs and an automated framework for testing VLM biases. Code and data are available at: vlmsarebiased.github.io.
Large Language Models (LLMs) have demonstrated strong performance on tasks with short time frames, but struggle with tasks requiring longer durations. While datasets covering extended-duration tasks, such as software engineering tasks or video games, do exist, there are currently few implementations of complex board games specifically designed for reinforcement learning and LLM evaluation. To address this gap, we propose the 4Hammer reinforcement learning environment, a digital twin simulation of a subset of Warhammer 40,000-a complex, zero-sum board game. Warhammer 40,000 features intricate rules, requiring human players to thoroughly read and understand over 50 pages of detailed natural language rules, grasp the interactions between their game pieces and those of their opponents, and independently track and communicate the evolving game state.
Significant progress has been made in AI for games, including board games, MOBA, and RTS games. However, complex agents are typically developed in an embedded manner, directly accessing game state information, unlike human players who rely on noisy visual data, leading to unfair competition. Developing complex non-embedded agents remains challenging, especially in card-based RTS games with complex features and large state spaces. We propose a non-embedded offline reinforcement learning training strategy using visual inputs to achieve real-time autonomous gameplay in the RTS game Clash Royale. Due to the lack of a object detection dataset for this game, we designed an efficient generative object detection dataset for training. We extract features using state-of-the-art object detection and optical character recognition models. Our method enables real-time image acquisition, perception feature fusion, decision-making, and control on mobile devices, successfully defeating built-in AI opponents. All code is open-sourced at https://github.com/wty-yy/katacr.
Reasoning and strategic behavior in social interactions is a hallmark of intelligence. This form of reasoning is significantly more sophisticated than isolated planning or reasoning tasks in static settings (e.g., math problem solving). In this paper, we present Strategic Planning, Interaction, and Negotiation (SPIN-Bench), a new multi-domain evaluation designed to measure the intelligence of strategic planning and social reasoning. While many existing benchmarks focus on narrow planning or single-agent reasoning, SPIN-Bench combines classical PDDL tasks, competitive board games, cooperative card games, and multi-agent negotiation scenarios in one unified framework. The framework includes both a benchmark as well as an arena to simulate and evaluate the variety of social settings to test reasoning and strategic behavior of AI agents. We formulate the benchmark SPIN-Bench by systematically varying action spaces, state complexity, and the number of interacting agents to simulate a variety of social settings where success depends on not only methodical and step-wise decision making, but also conceptual inference of other (adversarial or cooperative) participants. Our experiments reveal that while contemporary LLMs handle basic fact retrieval and short-range planning reasonably well, they encounter significant performance bottlenecks in tasks requiring deep multi-hop reasoning over large state spaces and socially adept coordination under uncertainty. We envision SPIN-Bench as a catalyst for future research on robust multi-agent planning, social reasoning, and human--AI teaming. Project Website: https://spinbench.github.io/
We introduce ZeroSumEval, a dynamic, competition-based, and evolving evaluation framework for Large Language Models (LLMs) that leverages competitive games. ZeroSumEval encompasses a diverse suite of games, including security challenges (Capture the Flag), classic board games (chess), and knowledge tests (MathQuiz). These games are designed to evaluate a range of capabilities such as strategic reasoning, planning, knowledge application, safety, and adaptability. Building upon recent studies that highlight the effectiveness of game-based evaluations for LLMs, ZeroSumEval enhances these approaches by providing a standardized and extensible framework for easily implementing games and leverages DSPy to provide a better abstraction for LLM player strategies.
The aim of this paper is to discuss the potential of using methods from Reinforcement Learning for Life Cycle Assessment in a circular economy, and to present some new ideas in this direction. To give some context, we explain how Reinforcement Learning was successfully applied in computer chess (and beyond). As computer chess was historically called the "drosophila of AI", we start by describing a method for the board representation called 'rotated bitboards' that can potentially also be applied in the context of sustainability. In the first part of this paper, the concepts of the bitboard-representation and the advantages of (rotated) bitboards in move generation are explained. In order to illustrate those ideas practice, the concrete implementation of the move-generator in FUSc# (a chess engine developed at FU Berlin in C# some years ago) is described. In addition, rotated binary neural networks are discussed briefly. The second part deals with reinforcement learning in computer chess (and beyond). We exemplify the progress that has been made in this field in the last 15-20 years by comparing the "state of the art" from 2002-2008, when FUSc# was developed, with the ground-breaking innovations connected to "AlphaZero". We review some application of the ideas developed in AlphaZero in other domains, e.g. the "other Alphas" like AlphaFold, AlphaTensor, AlphaGeometry and AlphaProof. In the final part of the paper, we discuss the computer-science related challenges that changing the economic paradigm towards (absolute) sustainability poses and in how far what we call 'progressive computer science' needs to contribute. Concrete challenges include the closing of material loops in a circular economy with Life Cycle Assessment in order to optimize for (absolute) sustainability, and we present some new ideas in this direction.
Hex and Counter Wargames are adversarial two-player simulations of real military conflicts requiring complex strategic decision-making. Unlike classical board games, these games feature intricate terrain/unit interactions, unit stacking, large maps of varying sizes, and simultaneous move and combat decisions involving hundreds of units. This paper introduces a novel system designed to address the strategic complexity of Hex and Counter Wargames by integrating cutting-edge advancements in Recurrent Neural Networks with AlphaZero, a reliable modern Reinforcement Learning algorithm. The system utilizes a new Neural Network architecture developed from existing research, incorporating innovative state and action representations tailored to these specific game environments. With minimal training, our solution has shown promising results in typical scenarios, demonstrating the ability to generalize across different terrain and tactical situations. Additionally, we explore the system's potential to scale to larger map sizes. The developed system is openly accessible, facilitating continued research and exploration within this challenging domain.
Advancing planning and reasoning capabilities of Large Language Models (LLMs) is one of the key prerequisites towards unlocking their potential for performing reliably in complex and impactful domains. In this paper, we aim to demonstrate this across board games (Chess, Fischer Random / Chess960, Connect Four, and Hex), and we show that search-based planning can yield significant improvements in LLM game-playing strength. We introduce, compare and contrast two major approaches: In external search, the model guides Monte Carlo Tree Search (MCTS) rollouts and evaluations without calls to an external game engine, and in internal search, the model is trained to generate in-context a linearized tree of search and a resulting final choice. Both build on a language model pre-trained on relevant domain knowledge, reliably capturing the transition and value functions in the respective environments, with minimal hallucinations. We evaluate our LLM search implementations against game-specific state-of-the-art engines, showcasing substantial improvements in strength over the base model, and reaching Grandmaster-level performance in chess while operating closer to the human search budget. Our proposed approach, combining search with domain knowledge, is not specific to board games, hinting at more general future applications.
Ludii is a Java general game system with a considerable number of board games, with an API for developing new agents and a game description language to create new games. To improve versatility and ease development, we provide Python interfaces for agent programming. This allows the use of Python modules to implement general game playing agents. As a means of enabling Python for creating Ludii agents, the interfaces are implemented using different Java libraries: jpy and Py4J. The main goal of this work is to determine which version is faster. To do so, we conducted a performance analysis of two different GGP algorithms, Minimax adapted to GGP and MCTS. The analysis was performed across several combinatorial games with varying depth, branching factor, and ply time. For reproducibility, we provide tutorials and repositories. Our analysis includes predictive models using regression, which suggest that jpy is faster than Py4J, however slower than a native Java Ludii agent, as expected.
Neural scaling laws are observed in a range of domains, to date with no clear understanding of why they occur. Recent theories suggest that loss power laws arise from Zipf's law, a power law observed in domains like natural language. One theory suggests that language scaling laws emerge when Zipf-distributed task quanta are learned in descending order of frequency. In this paper we examine power-law scaling in AlphaZero, a reinforcement learning algorithm, using a theory of language-model scaling. We find that game states in training and inference data scale with Zipf's law, which is known to arise from the tree structure of the environment, and examine the correlation between scaling-law and Zipf's-law exponents. In agreement with quanta scaling theory, we find that agents optimize state loss in descending order of frequency, even though this order scales inversely with modelling complexity. We also find that inverse scaling, the failure of models to improve with size, is correlated with unusual Zipf curves where end-game states are among the most frequent states. We show evidence that larger models shift their focus to these less-important states, sacrificing their understanding of important early-game states.
Monte-Carlo Tree Search (MCTS) typically uses multi-armed bandit (MAB) strategies designed to minimize cumulative regret, such as UCB1, as its selection strategy. However, in the root node of the search tree, it is more sensible to minimize simple regret. Previous work has proposed using Sequential Halving as selection strategy in the root node, as, in theory, it performs better with respect to simple regret. However, Sequential Halving requires a budget of iterations to be predetermined, which is often impractical. This paper proposes an anytime version of the algorithm, which can be halted at any arbitrary time and still return a satisfactory result, while being designed such that it approximates the behavior of Sequential Halving. Empirical results in synthetic MAB problems and ten different board games demonstrate that the algorithm's performance is competitive with Sequential Halving and UCB1 (and their analogues in MCTS).
MuZero has achieved superhuman performance in various games by using a dynamics network to predict environment dynamics for planning, without relying on simulators. However, the latent states learned by the dynamics network make its planning process opaque. This paper aims to demystify MuZero's model by interpreting the learned latent states. We incorporate observation reconstruction and state consistency into MuZero training and conduct an in-depth analysis to evaluate latent states across two board games: 9x9 Go and Outer-Open Gomoku, and three Atari games: Breakout, Ms. Pacman, and Pong. Our findings reveal that while the dynamics network becomes less accurate over longer simulations, MuZero still performs effectively by using planning to correct errors. Our experiments also show that the dynamics network learns better latent states in board games than in Atari games. These insights contribute to a better understanding of MuZero and offer directions for future research to improve the playing performance, robustness, and interpretability of the MuZero algorithm.
Although AlphaZero has achieved superhuman levels in Go, recent research has highlighted its vulnerability in particular situations requiring a more comprehensive understanding of the entire board. To address this challenge, this paper introduces ResTNet, a network that interleaves residual networks and Transformer. Our empirical experiments demonstrate several advantages of using ResTNet. First, it not only improves playing strength but also enhances the ability of global information. Second, it defends against an adversary Go program, called cyclic-adversary, tailor-made for attacking AlphaZero algorithms, significantly reducing the average probability of being attacked rate from 70.44% to 23.91%. Third, it improves the accuracy from 59.15% to 80.01% in correctly recognizing ladder patterns, which are one of the challenging patterns for Go AIs. Finally, ResTNet offers a potential explanation of the decision-making process and can also be applied to other games like Hex. To the best of our knowledge, ResTNet is the first to integrate residual networks and Transformer in the context of AlphaZero for board games, suggesting a promising direction for enhancing AlphaZero's global understanding.
This paper presents novel game-playing AI agents based on the AlphaZero framework, enhanced with Vision Transformer (ViT): AlphaViT, AlphaViD, and AlphaVDA. These agents are designed to play multiple board games of various sizes using a single network with shared weights, thereby overcoming AlphaZero's limitation of fixed-board-size constraints. AlphaViT employs only a transformer encoder, whereas AlphaViD and AlphaVDA incorporate both transformer encoders and decoders. In AlphaViD, the decoder processes outputs from the encoder, whereas AlphaVDA uses a learnable embeddings as the decoder input. The additional decoder layers in AlphaViD and AlphaVDA provide flexibility to adapt to various action spaces and board sizes. Experimental results show that the proposed agents, trained on either individual games or multiple games simultaneously, consistently outperform traditional algorithms such as Minimax and Monte Carlo Tree Search and approach the performance of AlphaZero, despite using a single deep neural network (DNN) with shared weights. In particular, AlphaViT shows strong performance across all tested games. Furthermore, fine-tuning the DNN using pre-trained weights from small-board games accelerates convergence and improves performance, particularly in Gomoku. Interestingly, simultaneous training on multiple games yields performance comparable to, or even surpassing, single-game training. These results indicate the potential of transformer-based architectures to develop more flexible and robust game-playing AI agents that excel in multiple games and dynamic environments.
Defining and measuring decision-making styles, also known as playstyles, is crucial in gaming, where these styles reflect a broad spectrum of individuality and diversity. However, finding a universally applicable measure for these styles poses a challenge. Building on Playstyle Distance, the first unsupervised metric to measure playstyle similarity based on game screens and raw actions, we introduce three enhancements to increase accuracy: multiscale analysis with varied state granularity, a perceptual kernel rooted in psychology, and the utilization of the intersection-over-union method for efficient evaluation. These innovations not only advance measurement precision but also offer insights into human cognition of similarity. Across two racing games and seven Atari games, our techniques significantly improve the precision of zero-shot playstyle classification, achieving an accuracy exceeding 90 percent with fewer than 512 observation-action pairs, which is less than half an episode of these games. Furthermore, our experiments with 2048 and Go demonstrate the potential of discrete playstyle measures in puzzle and board games. We also develop an algorithm for assessing decision-making diversity using these measures. Our findings improve the measurement of end-to-end game analysis and the evolution of artificial intelligence for diverse playstyles.
Reinforcement learning (RL) has progressed substantially over the past decade, with much of this progress being driven by benchmarks. Many benchmarks are focused on video or board games, and a large number of robotics benchmarks lack diversity and real-world applicability. In this paper, we aim to simplify the process of applying reinforcement learning in the 3D simulation league of Robocup, a robotic football competition. To this end, we introduce a Robocup-based RL environment based on the open source rcssserver3d soccer server, simple pre-defined tasks, and integration with a popular RL library, Stable Baselines 3. Our environment enables the creation of high-dimensional continuous control tasks within a robotics football simulation. In each task, an RL agent controls a simulated Nao robot, and can interact with the ball or other agents. We open-source our environment and training code at https://github.com/Michael-Beukman/RobocupGym.
People can evaluate features of problems and their potential solutions well before we can effectively solve them. When considering a game we have never played, for instance, we might infer whether it is likely to be challenging, fair, or fun simply from hearing the game rules, prior to deciding whether to invest time in learning the game or trying to play it well. Many studies of game play have focused on optimality and expertise, characterizing how people and computational models play based on moderate to extensive search and after playing a game dozens (if not thousands or millions) of times. Here, we study how people reason about a range of simple but novel Connect-N style board games. We ask people to judge how fair and how fun the games are from very little experience: just thinking about the game for a minute or so, before they have ever actually played with anyone else, and we propose a resource-limited model that captures their judgments using only a small number of partial game simulations and almost no look-ahead search.
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. A sample of the generated games are available to play online through the Ludii portal.