This article delves into the complexities of the Game of Life, a grid-based automaton that has long fascinated the AI community. Researchers at Swarthmore College and Los Alamos National Laboratory recently explored how neural networks struggle to learn the intricate rules of the Game of Life. They found that, while larger neural networks can eventually achieve near-perfect accuracy, the cost of training and running them is significantly higher. This investigation sheds light on the challenges faced by neural networks when dealing with symbolic rules, suggesting that more sophisticated systems may require even larger models. Ultimately, the findings call for further research into more efficient learning algorithms to overcome these limitations.
https://bdtechtalks.com/2020/09/16/deep-learning-game-of-life/