← Parminces
Shane Legg, PhD Dissertation (2008)

Machine Super
Intelligence

A mathematical theory of intelligence: from defining it formally, to building the universal AIXI agent, to measuring the limits of computation. The dissertation that co-founded DeepMind.

7
Chapters
15+
Simulations
50+
Quizzes
Part I: What Is Intelligence?
Chapter 1

Nature and Measurement of Intelligence

Theories, definitions, and tests of intelligence for humans, animals, and machines.

Part II: Universal Artificial Intelligence
Chapter 2

Universal Artificial Intelligence

Inductive inference, Solomonoff prior, Kolmogorov complexity, and the AIXI agent.

Chapter 3

Taxonomy of Environments

Passive and active environments, MDPs, POMDPs, ergodicity, and self-optimising agents.

Chapter 4

Universal Intelligence Measure

A formal mathematical definition of machine intelligence and its properties.

Part III: Limits and Practice
Chapter 5

Limits of Computational Agents

Prediction complexity, hard sequences, and the Godel incompleteness barrier.

Chapter 6

TD Learning Without a Learning Rate

Deriving an automatic learning rate for temporal difference methods from first principles.

Part IV: The Big Picture
Chapter 7

Discussion: Machine Superintelligence

Are superintelligent machines possible? How might they be built? Is it a good idea?