Connection Game

Connection games, encompassing various rule sets from simple Connect Four-like variations to complex word association puzzles like the New York Times' Connections, are a fertile ground for investigating human and artificial intelligence reasoning. Current research focuses on understanding how humans intuitively assess game fairness and fun, employing models based on limited simulations, and on evaluating the abstract reasoning capabilities of large language models (LLMs) and reinforcement learning algorithms in solving these games. These studies highlight the challenges posed by these games as benchmarks for evaluating different AI architectures, revealing limitations in LLMs' ability to handle associative and abstract knowledge while also demonstrating the potential of logic-based approaches for creating more interpretable AI systems for game playing and prediction.