Download WHAT IS ARTIFICIAL INTELLIGENCE? Cognitive simulation

Prehistory of AI
WHAT IS ARTIFICIAL INTELLIGENCE?
The idea of artifical man (“second creation”): automata and androids
Cognitive simulation
(
Thinking rationally / human-like)a
Antiquity and Middle Ages
Construction of “intelligent” systems
(
Modern Age: Empirism, Rationalism, Materialism; Robots
Acting rationally / human-like)
The idea of the universal computing engine
Investigation of principles of “information” processing
Leibniz, Pascal, . . . , Babbage
Strict formalization
Turing, Zuse, v. Neumann
Exemplary implementations
The universal language idea
“Thinking as computation” — Logic and functional simulation
Thinking is also / only computation.
(“also” = weak vs. “only” = strong AI hypothesis)
Calculation procedures: Al-Chorezmi (Al-Khwarizmi), Fibonacci, Riese
Interdisciplinary character of AI: Computer science, neuroscience, psychology,
Hobbes, Leibniz
philosophy, linguistics, control theory, economics,. . .
Frege, Boole, Peirce, Hilbert, Gödel,. . .
a
Wittgenstein I: Tractatus and the “Linguistic Turn”; “ideal language” program
cf. Russell/Norvig
c G. Görz, FAU Erlangen-Nürnberg, Inf. 8
5
c G. Görz, FAU Erlangen-Nürnberg, Inf. 8
6
Phases of the Development of AI
The Origin of AI in the 20th Century
Dartmouth conference 1956
Modern formal logic
Foundational phase: “power-based approach” — heuristic search
Turing and the Turing Machine
Knowledge representation: “knowledge-based approach”
McCulloch and Pitts — Neural Networks: analog, topological computation
Knowledge-based systems, broad applications, “fifth generation”
Dartmouth conference 1956 — “GOFAI” McCarthy, Minsky, Newell, Simon,
Neuronal networks connectionism
et al.: discrete, algebraic computation
Multi-agent systems
Vannevar Bush — MEMEX: hypermedia
“Nouvelle AI”: robotics without central representation, subsumption architecture
c G. Görz, FAU Erlangen-Nürnberg, Inf. 8
7
c G. Görz, FAU Erlangen-Nürnberg, Inf. 8
8
Agent Types
Rational “AGENTS”
Reflex agent
“For each possible percept sequence, an ideal rational agent should do whatever
– rule-based
action is expected to maximize its performance measure, on the basis of the
– e.g. factorial agent
evidence provided by the percept sequence and whatever built-in knowledge the
Memory-based agents
agent has.” (Russell/Norvig)
– rule-based plus learned rules
sensors
– e.g., factorial agent that remembers previous solutions
percepts
?
environment
Goal-based agents
agent
actions
– planner
– e.g. route planner
effectors
Utility-based agents
– maximize continuous-valued goals
Agent = architecture + program
– e.g. planner: tradeoff between time and cost
c G. Görz, FAU Erlangen-Nürnberg, Inf. 8
Agent
9
c G. Görz, FAU Erlangen-Nürnberg, Inf. 8
Sensors
10
Sensors
State
What action I
should do now
What my actions do
Conditionaction rules
Agent
Effectors
c G. Görz, FAU Erlangen-Nürnberg, Inf. 8
What the world
is like now
11
c G. Görz, FAU Erlangen-Nürnberg, Inf. 8
What action I
should do now
Environment
Conditionaction rules
How the world evolves
Environment
What the world
is like now
Effectors
12
Sensors
Sensors
State
State
What my actions do
What it will be like
if I do action A
What action I
should do now
Goals
Agent
What the world
is like now
What my actions do
What it will be like
if I do action A
Utility
How happy I will be
in such a state
What action I
should do now
Agent
Effectors
c G. Görz, FAU Erlangen-Nürnberg, Inf. 8
How the world evolves
13
Environment
What the world
is like now
Environment
How the world evolves
Effectors
c G. Görz, FAU Erlangen-Nürnberg, Inf. 8
14
Fundamental Subfields of AI
Questions to Ask
Given the goal to implement rational action in a complex environment, as
in each part. . .
exemplified by the conception of the rational agent, AI is an interdisciplinary,
cross-cutting discipline.
What is the subject?
There is nothing like a big unifying theory of AI, but instead an inventory of methods
Which functionality do we want to achieve?
which can be associated with fundamental subfields:
Which phenomena do we have to deal with?
Problem solving and heuristic search
What are typical examples?
Knowledge representation and inference
Planning
Which subproblems are there?
Learning
What are the underlying theories? Which approach will be chosen?
Perception
Which methods and techniques are used; what are their properties?
Natural language processing and image interpretation
Which extensions and which interactions are there?
Agent architectures
c G. Görz, FAU Erlangen-Nürnberg, Inf. 8
15
c G. Görz, FAU Erlangen-Nürnberg, Inf. 8
16