Textbook Notes - AIMA, R&N
Chapter One
CSC447 - Spring 2010
What is AI?
– Our definition
gain better
understanding of intelligence with the goal of building machines better able to
serve humankind.
– Other definitions
4 quadrants
– upper - thinking; lower -
acting
– leftmost - simulate humans;
rightmost - strive for rationality
results in 4
definitions
– acting humanly :: Turing
Test approach
Need: NLP, KR&R, ML, CV, robotics
Problems: TT based on deception; courtroom in Veterans Stadium
– thinking humanly ::
cognitive modeling approach
Problem: early days of AI there was confusion between computer
models & actual human processes
– thinking rationally :: Laws
of thought approach
Aristotle, George Boole, logicists
Problems: not easy to formalize incomplete knowledge;
solve in principle ≠ solve in practice
– acting rationally ::
rational agent approach
r.a. = one that acts so as to achieve best outcome (or best
expected
outcome)
Needs: same skills as TT
2 advantages over others
more general than laws of thought approach
more amenable to scientific development
R&N claim: standard of rationality is mathematically well defined and
completely general
AIMA concentrates on general principles of r.a.'s and on components for
constructing them
Limitation: achieving perfect rationality in complicated environments is
not feasible
Problems:
Philosophically, since humans set principles of rationality, this is a
subset of thinking/acting humanly
Sheldon Principle - watch some episodes of Big Bang Theory
Justification of scientific induction - knotty, unsolved problem in
philosophy of science
Waning influence of logical positivism, e.g.,
here
– Advantages of our definition
can use any of
the other 4 approaches without being boxed in by philosophical
commitments
have an
identifiable, actionable and justifiable goal
have a clear
research agenda
no need to
prove anything; proof of the pudding is in the eating
notice how many
of the items mentioned on pp. 28-29 fall into this category
– Alternative to the Turing Test
the Touring Car
Test - does it get us from point A to point B?
we set the
goal; if the goal is achieved we move on; no need to engage in
endless philosophical arguments over intelligence, rationality, etc.
The Foundations of AI
Philosophy
– Questions
Formal rules
®
valid conclusions?
Physical brain
®
mind?
Whence
knowledge?
Knowledge
® action?
– Milestones & bumps in the road
Aristotle -
syllogisms
Ramon Lull -
can reason with mechanical artifact
Thomas Hobbes -
reasoning like numerical computation; artificial animal
Calculators -
da Vinci, Schickard, Pascal, Leibniz
Schools of
philosophy - rationalism, dualism, materialism, empiricism, logical
positivism
Individuals -
Descartes, Bacon, Locke, Hume, Wittgenstein, Carnap, Hempel
Carnap - first
theory of mind as computational process
Justification
of induction - unsolved problem
Aristotle -
connection between goals and knowledge of outcome
implemented by Newell & Simon in GPS (General Problem Solver)
Note: GPS
®
STRIPS, bypassed by R. Brooks