PathNet is a novel learning algorithm designed to support transfer, continual, and multitask learning in artificial neural networks. It uses a population of pathways (views) through the network, each of which determines a subset of parameters to be updated during training. During learning, a tournament selection genetic algorithm evolves these pathways, selecting the best-performing ones for replication and mutation. PathNet demonstrates successful transfer learning by fixing the parameters of a pathway learned on task A and re-evolving a new population of pathways for task B, achieving faster learning compared to learning from scratch or fine-tuning. The algorithm has been tested on various supervised and reinforcement learning tasks, including binary MNIST, CIFAR, SVHN, Atari games, and Labyrinth games, showing positive transfer in all cases. PathNet also improves the robustness of a parallel asynchronous reinforcement learning algorithm (A3C) to hyperparameter choices. The framework of PathNet is inspired by Darwinian Neurodynamics and aims to address the challenges of artificial general intelligence, particularly in terms of parameter reuse and catastrophic forgetting.PathNet is a novel learning algorithm designed to support transfer, continual, and multitask learning in artificial neural networks. It uses a population of pathways (views) through the network, each of which determines a subset of parameters to be updated during training. During learning, a tournament selection genetic algorithm evolves these pathways, selecting the best-performing ones for replication and mutation. PathNet demonstrates successful transfer learning by fixing the parameters of a pathway learned on task A and re-evolving a new population of pathways for task B, achieving faster learning compared to learning from scratch or fine-tuning. The algorithm has been tested on various supervised and reinforcement learning tasks, including binary MNIST, CIFAR, SVHN, Atari games, and Labyrinth games, showing positive transfer in all cases. PathNet also improves the robustness of a parallel asynchronous reinforcement learning algorithm (A3C) to hyperparameter choices. The framework of PathNet is inspired by Darwinian Neurodynamics and aims to address the challenges of artificial general intelligence, particularly in terms of parameter reuse and catastrophic forgetting.