Explore the groundbreaking advancements in self-learning systems through this insightful lecture by Demis Hassabis, Co-Founder and CEO of Google DeepMind. Delve into DeepMind's recent breakthroughs and their implications for scientific and philosophical discovery. Learn about the evolution of AI systems, from chess engines to AlphaZero and AlphaStar, and their applications beyond games. Discover how these technologies are being utilized to tackle real-world challenges, such as improving wind power and protein folding. Gain valuable insights into the architecture, training methods, and intrinsic motivations behind cutting-edge AI systems. Understand the key challenges that remain in the field and the potential future developments in artificial intelligence.
Overview
Syllabus
Intro
Building a general purpose learning system
The lineage of AlphaZero
Chess and Al: a long and storied history
Anatomy of a world champion chess engine
AlphaZero (AlphaGoZero)
AlphaZero versus Stockfish 8
Mobility vs Materiality
Game Changer
Game objectives
AlphaStar architecture
AlphaStar league-population based training
Intrinsic motivations & diversity of strategies
AlphaStar training
Many interesting issues
Many key chalenges remain
Beyond Games
Improving wind power
Desiderata
What is protein folding?
Molecular machines
Training the NN via supervised learning
Structure optimization
Taught by
MITCBMM
