Transcript Lecture 1 - Computer Science Department, Technion

Introduction to Artificial
Intelligence
236501
Ruth Bergman
Fall 2002
Kismet
What is an Artificial
Intelligence?
Genghis
EduSpeak
ALVINN
See-Threepio
Definitions of AI
• “[The automation of] activities that we associate with human
thinking…” [Bellman 1978]
• “The study of how to make computers do things at which, at the
moment, people are better [Rich and Knight, 1991]
• “The study of mental faculties through the use of computational
models” [Charniak and McDermott, 1985]
• “The branch of computer science that is concerned with the
automation of intelligent behavior” [Luger and Stubblefield,
1993]
Note: - think vs. act
- human vs. rational
Cognitive Modelling
• Construct programs that think like humans.
• How do humans think? Introspection or
psychological experiments.
• Why imitate human thought?
– To solve problems.
– To learn about human cognitive processes.
• Cognitive Science: the interdisciplinary field
that brings together AI and psychological
experimental techniques to try to understand
workings of the human mind.
Turing Test
How can we evaluate intelligence?
• Turing [1950] a machine can be deemed intelligent
when its responses to interrogation by a human are
indistinguishable from those of a human being.
interrogator
human
machine
Turing Test
Requires solving hard problems:
– natural language processing
– knowledge representation
– automated reasoning
– machine learning
Is the test valid?
– Too strict: is human intelligence the only form of intelligence
– Too lax: is giving the appearance of intelligence sufficient
Rational Thought:
Logic
• The logicist school of thought in AI uses
formal logic techniques to automatically solve
problems.
• Successes in fields that lend themselves to
formal description, such as mathematics.
• Obstacles to this approach:
– Real-world problems are difficult to formalize
– Computation time proves a barrier for realistic
problems in practice.
The rational agent
• The agent is assumed to exist in an
environment in which it perceives and acts
• The agent is rational if it acts to attain its
goals given its belief about the environment
• Motivation:
– Not all thought is rational, e.g. reflex
– A correct action may not exist using inference
– Addresses the cognitive skills required by the
Turing Test
– Avoids the complex issue of behaving humanly
which is impossible to measure.
A Brief History of AI
• 1960’s: basic techniques, microworlds
Logic, Lisp, Games (checkers)
• 1970’s: Knowledge-based systems
Expert systems (DENDRAL, MYCIN)
• 1980’s: AI in industry, machine learning.
Industrial expert systems, Lisp Machine, industrial
vision
• 1990’s: Rational agent, multi-agent systems,
formal methods
Decision making, whole agent approach
AI Fields of Study
• Core AI:
– Problem solving games
– Knowledge representation medical diagnosis
– Reasoning
theorem proving
– Learning
neural networks
• Vision
• Natural Language
• Robotics
face recognition
speech recognition
mobile robots
The Intelligent Agent
• Anything that perceives its
environment through sensors and
acts on its environment through
effectors.
• performance measure: goals
• autonomy
sensor
effector
Polly, a vision-based artificial agent (1994)
Structure of Intelligent
Agents
Sensors
Agent
Effectors
Environemnt
agent program
Agent Research
Agent type
Percepts
Actions
Goals
Environment
characteristics
Mobile robot
Obstacle
ahead
Move ahead,
turn
Do not collide
Dynamic,
continuous,
inaccessible
social
humanoid
robot
Pixels of
varying
intensity,
color, speech
Display happy, Engage
sad, angry…
humans
face
Dynamic,
continuous,
accessible
Personal
assistant
text
Sort, inform
user
Dynamic,
discrete,
accessible
Correct
categorization
The agent program
• The agent program maps from percepts to
actions
• The agent uses knowledge representation to
model the environment
• The agent must reason with knowledge and
make decisions
• The agent may learn to improve its world
model and decision making ability.
Thus all AI techniques are part of the intelligent
agent.
Components of the
agent program
Sensors
goals
Agent
Representation
of the world
Inference &
Action selection
Effectors
Environemnt
World
model
Components of the
Polly program
Sensors: b/w video camera
Effectors: wheels, voice synthesizer
World Model: map of MIT AI Lab
Representation of the world: location in map, tour group
Goals: conduct tour
Decision making: avoid obstacles, wander, follow corridor, tour
Sensors
goals
Agent
Rep
Act?
Effectors
Environemnt
World
model