Transcript ppt

Notes
 Program evaluation – Sept 11-12
 Student submissions
 Mon. Sept 11, 4-5PM
 FA 181
 Comments to committee are anonymous
 Wine and cheese – Sept 15, 1:30-4:00PM
D Goforth - COSC 4117, fall 2006
1
COSC 4117
Artificial Intelligence
Thinking and acting
at least as well
as humanly possible
http://www.cs.laurentian.ca/dgoforth/cosc4117/outline.html
Course outline
 Russell and Norvig – ch 1-12
plus handouts
 Major programming project
 3 assignments
 Final exam
D Goforth - COSC 4117, fall 2006
50%
24%
26%
3
Artificial intelligence
 what an ‘electronic brain’ does…
 methods for solving (partially)
problems that are difficult because
 known algorithms are O(en)
 play chess
 no algorithms are known
 converse in English
 we hardly know where to begin
 fall in love
D Goforth - COSC 4117, fall 2006
4
Definitions of AI




think like a human being
think like a human but better
act like human being – Turing test
be a successful functioning entity
D Goforth - COSC 4117, fall 2006
5
Definitions of AI – human thinking
 think like a human being
 based on philosophy, psychology,
neuroscience,...
 programs to emulate human processes:
learning, reasoning, memory
 testing theories of how humans think
 influence on models of human thinking
 weak and strong AI – mind-body
problem – Searle’s Chinese room
thought experiment
D Goforth - COSC 4117, fall 2006
6
Definitions of AI – think rationally
 think like a human but better
 battle of neats and scruffies
 extension of long tradition of analysing
thought – rationality
 logic as basis for programming – completed
by 1960’s BUT limited usefulness
 extensions to real situations: too little, too
much or inconsistent information
D Goforth - COSC 4117, fall 2006
7
Definitions of AI – act human
 act like a human being
 bird-flight argument
 ‘black box’ emulation – Turing test
 intelligence/thinking plus the ‘hard’
peripherals: perception, language,
motion
D Goforth - COSC 4117, fall 2006
8
Definitions of AI – act rationally
 be a successful functioning entity as
‘agent’ in an ‘environment’





set goals
perceive environment
communicate with other agents
plan
act
 object-oriented model, subsumes all others
 current best definition of AI
D Goforth - COSC 4117, fall 2006
9
50th anniversary
Dartmouth Conference 1956
 McCarthy coined “Artificial Intelligence”
 Slogan that was starting point of discussion:
“Every aspect of learning or any
other feature of intelligence can in
principle be so precisely described
that a machine can be made to
simulate it.”
 Newell & Simon: Logic Theorist
 first noncalculating program – abstract proofs
D Goforth - COSC 4117, fall 2006
10
History of AI - 1
 ‘electronic brain’ – potential of logic
circuitry for automated reasoning
 Minski - neuron models of processing –
father of scruffies – “Society of Mind”
 Newell & Simon - symbol processing –
in contrast to
‘computing’ ( == number crunching)
 McCarthy – neat freak – computing
environments (LISP, time-sharing)
D Goforth - COSC 4117, fall 2006
11
History of AI - 2
 progress fast; predictions faster
 systems demo-ing capabilities in
reasoning, planning, robotics control
conversation
 logics extended to uncertain knowledge,
assumptions, viewpoints
 crash of 1970’s
 ‘scaling up’ – exponential performance
 ‘brittleness’ – importance of knowledge
 failure of neural network ‘perceptrons’
D Goforth - COSC 4117, fall 2006
12
History of AI - 3
 knowledge-based systems
 encoding experience and expertise
 rules and heuristics
 meta-knowledge – knowing how to
know – ontology (semantic net)
 common sense
 triumph of the meta-neats – Cohen’s
analysis, formalizing the science of AI
D Goforth - COSC 4117, fall 2006
13
History of AI - 4
 the agent model – Newell
 influence of Russell and Norvig
 apparent future directions:
 autonomy in bigger and bigger
environments
 *bots of all kinds
 internet as environment – applications
 search engines, language translation,...
D Goforth - COSC 4117, fall 2006
14
History of AI - 5


1997 Chess champion IBM Deep Blue
2005 DARPA Grand challenge
 Sebastian Thrun’s Stanley
‘On Saturday, the Stanford Racing
Team's robotic car, "Stanley,"
drove autonomously across 131.6
miles in the Mojave Desert in six
hours and 53 minutes, finishing
about 11 minutes faster than
Carnegie Mellon's "Sandstorm." Its
average speed was 19.1 mph, versus
Sandstorm's 18.6 mph.’
cnet news, Oct 9, 2005 (news.com.com)
D Goforth - COSC 4117, fall 2006
15
The programming project
 intelligent game player agent to act
autonomously in game environment




perceptions and actions are easy
goal setting is easy
evaluation of performance is easy
focus is on core intelligence: playing well;
not easy
D Goforth - COSC 4117, fall 2006
16
The programming project
 stages:
1.
2.
3.
4.
5.
pick teams and design game structure
program the game
program agent to play game
compete against other agents
write up the results
 work in teams of 1 or 2
D Goforth - COSC 4117, fall 2006
17
The programming project
 this year’s game is
KING’S COURT
 first deadline:
Monday, Sept 25
 pick a team
 design data structure
for the game
D Goforth - COSC 4117, fall 2006
18