Hebbian Automata Reinforcement Learning Improviser (HARLI) demonstrates its 'wave' strategy. HARLI starts with randomly initialised weights and updates them according to an evolved Hebbian learning policy, eventually reaching a high-scoring wave generation strategy. The waves travel at c, the maximum speed in Life-like CA, and yield a reward of about 40 to 50. -> https://github.com/rivesunder/harli_learning, (Davis 2021)