Demis Hassabis: Towards General Artificial Intelligence

Overview

Explore the cutting-edge developments in General Artificial Intelligence with Dr. Demis Hassabis, Co-Founder and CEO of DeepMind, in this illuminating talk. Delve into the world of AI research, including the historic AlphaGo match, and discover its potential impact on science, healthcare, and our understanding of the human mind. Learn about general-purpose learning algorithms, reinforcement learning frameworks, and the evolution of AI from Deep Blue to modern neural networks. Gain insights into the development of memory-enhanced neural systems, 3D environment navigation, and the intricacies of training deep neural networks for complex tasks like mastering the game of Go. Understand the cultural significance of AI achievements and the rapid progress in the field, while contemplating the role of intuition and creativity in artificial intelligence.

Syllabus

Intro
General-Purpose Learning Algorithms
Deep Blue vs. Garry Kasparov
Reinforcement Learning Framework
Grounded Cognition
Deep Reinforcement Learning
DON: A General Atari Player
Adding Memory to Neural Networks
Memory: Neural Turing Machines
3D Environments - Navigation
The History of Go
A repeated board position is not allowed
Evaluation function for Go
Intuition vs Calculation
Training the deep neural networks
Two networks: Policy and Value Nets
Combining Neural Nets with Tree Search and Rollouts
Cutting Down the Search Tree
Evaluating AlphaGo against computers
Google DeepMind Challenge Match: 9-15 March
Note on Compute Power
We won the match 4-1
Game 2 - AlphaGo's Move 37
Cultural Impact of the match
Rate of Progress: -1 rank per month
Intuition and Creativity