Summary

  • Scientists David Silver and Richard Sutton claim AI is at the “threshold” of an “era of experience” where AI systems will be less reliant on human-provided data for improvement.
  • Self-improvement will come from the AI agents accessing data from, and interacting with, the real world, says the paper.
  • Although it looks forward and conceptual, the paper claims real-world implications for enterprises building with and for future AI systems and agents.
  • Famous essay by Sutton in 2019 stated that long-term progress in AI comes from leveraging large-scale computation and general-purpose search/learning methods over incorporating complex human-derived domain knowledge.
  • Silver is a senior scientist at DeepMind and was behind AlphaGo, AlphaZero, and AlphaStar, all breakthroughs in deep reinforcement learning.
  • The “era of experience” adapts these concepts and states that progress relying on human-provided data is slowing, requiring a new approach.
  • AI agents will have their own “stream of experience” to enable long-term goals and adapting to new behavioural patterns, and act autonomously.

By Ben Dickson

Original Article