AI has grown beyond human knowledge, says Google's DeepMind unit
a year ago
- #AI
- #DeepMind
- #ReinforcementLearning
- Google's DeepMind researchers argue that AI needs experiential learning to advance beyond current limitations.
- Current AI models rely too much on static, human-generated data, which restricts their ability to develop new capabilities.
- The proposed 'streams' approach builds on reinforcement learning, allowing AI to interact with the world and learn from continuous experiences.
- AI agents in streams would have long-term goals, memory, and adaptability, unlike current models that focus on short interactions.
- This shift could lead to AI surpassing human intelligence, unlocking unprecedented capabilities in various domains.
- Potential risks include job displacement and reduced human oversight over autonomous AI agents.
- Experiential data is expected to surpass human-generated data in scale and quality, driving AI advancements.