New AI Learns Without Forgetting Previous Skills

This article discusses a new AI technique called Progressive Neural Networks that can learn new tasks without forgetting previous skills. The system adds new parts that connect to old ones, allowing it to reuse prior knowledge when facing new problems.

Details

The Progressive Neural Networks approach aims to enable AI systems to continuously learn new tasks and skills without forgetting what they already know. Instead of erasing old skills when learning new ones, the system adds new parts that connect to the previous network. This allows the AI to reuse useful features and knowledge from prior tasks when facing new problems. Researchers have tested this on simple video games and 3D mazes, finding that it tends to learn faster and retain earlier skills better than traditional train-and-tweak methods. The key is the side connections that share features across tasks, enabling new skills to build on old ones rather than destroying them. While not perfect for every application, this approach points towards a path for developing AI systems that can learn and improve over time without losing past lessons, potentially leading to smarter and more capable AI assistants.