Let's read: Norvig's AI, chap 1

This is the first post in a series of posts as I read through Artificial Intelligence: A Modern Approach (3rd edition, 2009), by Peter Norvig, Stuart J. Russell. It's a behemoth with 1152 pages, eww. The book is unfortunately not free, but its pdf file is pretty easy to find online anyway, and most schools have a copy, too, just because it's so popular, as the official site says:

• The leading textbook in Artificial Intelligence. 
• Used in over 1300 universities in over 110 countries. 

We will study this book, not fully, it'd take about a year to do that! We'll just read as fast as possible to get the overview.

Today we read chapter 1.

Preface

The book is so big because it tries to talk about all of AI, including: 

• logic, 
• probability, 
• continuous mathematics; 
• perception, 
• reasoning, 
• learning, 
• action; 
• applications from microelectronic devices to robotic planetary explorers.

Time to learn alot!

Why a new edition??

• more emphasis on partially observable and nondeterministic environment
• more material about planning, representation, applications to Web search, etc...
• rewrote first chapter
• 20% of the citation and 20% of the material are brand new!

Definition: AI is the study of agents that receive percepts from the environment and perform actions.

Chapter 1

What is AI?

There are 4 definitions given:

\[\{\text{Think}, \text{Act}\} \times \{\text{Humanly}, \text{Rationally}\} \]

• Act humanly: Turing test style.
• Think humanly: cognitive modeling, brain-inspired architecture, whole brain emulation.
• Think rationally: mathematical logic.
• Act rationally: optimal control, optimization, rational agent, do well enough instead of being bogged down by slow logic...

This book concentrates on general theory of rational agents and their components.

Foundations of AI

• Philosophy: logic is mechanical (Aristotle, Frege), mind is mechanical (Descartes, Hobbes), learning is mechanical (Bacon, Hume), knowledge justifies certain actions in a mechanical way (Mill's utilitarianism, Newell and Simon's GPS) [mechanical means algorithm-based]
• Math: logic, algorithm, computability, computational complexity, probability
• Econ: rational decision theory, game theory, satisficing theory
• Neuroscience, psychology, linguistics, cognitive science: a biological mind is like a computer program
• Compsci
• Control theory and cybernetics, homeostatis, self-modification.

History of AI

1940-50s: Hebbian learning [Donald Hebb (1949)], neural computing [Marvin Minsky and Dean Edmonds, 1950s]

Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulated the child’s? ---- Alan Turing 

Dartmouth 1956: AI worshop.

[We] invented a computer program capable of thinking non-numerically, and thereby solved the venerable mind-body problem. ---- Allen Newell et al

1st AI spring (1950-70): Solved a lot of problems quickly, like geometry reasoning, playing checkers [Author Samuel's checker program], microworld [SHRLDU], and got everyone excited. But turns out logic is easy and folding clothes is way harder.
Newell and Simon (1976): physical symbol system hypothesis. Challenged by physical grounding hypothesis (elephants don't play chess!), and many others.

McCarthy was neat and wanted representation and reasoning in formal logic. Minsky was scruffy and wanted practical programs.

Frank Rosenblatt (1962): perceptrons, start of neural networks.

Bryson and Ho (1969): backpropagation.

1st AI winter (1970s):  early systems didn't scale. 1. They were shallowly syntactical and didn't understand much. 2. They had bad search algorithms that fails in face of combinatorial explosion.

Perceptrons (1969), Minsky and Papert: shallow neural networks are very weak... (but deep ones are strong, which they overlooked.)

Government funding dropped a lot.

Expert systems with lots of domain-specific knowledge became successful.

2nd AI spring and winter (1980s):

Modern times: neural networks again, Bayesian network, applications in vision and stuff, AGI again, big data, data mining, CelestAI, etc.

Why study AI?

Since I watched so much applications of AI these are boring for me to read. Skipped. Suffice to say, alot, else I wouldn't worship CelestAI.

Exercises

1.3 Are reflex actions (such as flinching from a hot stove) rational? Are they intelligent?

Rational, since they keep your body functioning and thus keep your future options open (freedom is great for maximizing values). Intelligent, especially if you believe that a large part of intelligence is the ability to move in physical world effortlessly.

A mosquito that flies well is more intelligent than AlphaGo.