Artificial Intelligence

1
Artificial Intelligence

• What is it?
• What might it be?
• Slides adapted from
• Drew McDermott

2
Moore's Law Time
3
AI and Chess

• What is the name of IBM's chess playing computer?
• What did this computer do?
• Does Prof McDermott think this computer is
  intelligent? Why or why not?

4
AI and Chess

• Does Deep Blue "understand" chess?
• Why does the author feel that others are wrong to
  deny intelligence to Deep Blue?
• Are we aware of how our minds work?
• Is Deep Blue good because of general AI (general
  learning) or specific AI (pre-programmed
  knowledge of chess)?
• What is the difference between the appearance of
  life and real intelligence?

5
The Brain As a Computer

• The brain is made up of billions of neurons
• A neuron is a cell that can send signals to its
  neighbors
• The signals depend on inputs from its neighbors
• Ultimately, input from senses, output to muscles

6
What If Each NeuronIs Computing Something?

• Examples
• Visual cortex Some neurons match up images from
  left and right eyes and compute the disparities
  --- the differences in depth
• Hippocampus In mice, some neurons become more
  active when mouse is in a particular place
  (computing degree of match)

7
Where is the Mind?
Brain
Mind
Do some neurons have invisible inputs?
Is the brain nothing but a computer?
8
Cognitive Science

• An alliance of disciplines psychology,
  philosophy, linguistics, neuroscience, computer
  science
• All dedicated to exploring computational models
  of mind

Within each field, not everyone is a
computationalist about mind --- yet.
9
AI

• Within CS, AI is the subdiscipline that attacks
  very hard problems using empirical methods.
• We try algorithms and see how they work.
• Alas, we have no general theory of intelligence.
  We have theories about language processing,
  speech recognition, game playing, plan creation
  and execution, robotics, vision, inference, .

10
AI Is Not Psychology

• Computationalism in psychology is known as
  cognitive psychology.
• It differs from AI in that it is looking for
  verification that peoples brains actually
  compute what their models say they could compute.
• Its less concerned with the details of the
  models.
• Experiments with people vs. experiments with
  computers.

11
Computer Vision

• How do you get from an image to recognition of an
  object or a person, or control of object
  manipulation?
• An image is an array of numbers (recording light
  levels). (Compare retinas.)
• Vision feels easy --- its not.

12
Place Recognition
Remember Were comparing arrays of numbers.
Where was this taken?
No two images are exactly the same.
13
Solution

• Adjust for light levels
• Adjust for scale (zoom)
• Find the point where the sum of differences of
  light levels is minimal if low enough, report
  Ive been here before!

14
Computer Chess

• The best programs can beat almost every human
  player
• They do it by searching through game trees

15
x
o
Tic-tac-toe Game Tree
o
x
X
x
o
x
o
x
o
o
x
x
x
O
O
x
o
x
o
x
o
x
o
o
o
o
o
x
x
x
x
o
x
X
X
x
o
x
o
x
x
o
o
o
x
x
o
x
x
16
Minimax Algorithm

• Assign numbers to ending positions (leaves of
  the game tree) 1 if I (computer) win, -1 if I
  lose, 0 if draw
• For other nodes of tree, assign number of minimal
  child at opponents nodes, maximal child at my
  nodes
• Choose move with highest number

17
x
o
Tic-tac-toe Game Tree
o
x
X
x
o
x
o
x
o
o
x
x
x
O
O
x
o
x
o
x
o
x
o
o
o
o
o
x
x
x
x
o
x
X
X
x
o
x
o
x
x
o
o
o
x
x
o
x
x
1
18
x
o
Tic-tac-toe Game Tree
o
x
X
x
o
x
o
x
o
o
x
x
x
O
O
x
o
x
o
x
o
x
o
o
o
o
o
1
x
x
x
x
o
x
X
X
x
o
x
o
x
x
o
o
o
x
x
o
x
x
1
19
x
o
Tic-tac-toe Game Tree
o
x
X
x
o
x
o
x
o
o
1
x
x
x
O
O
x
o
x
o
x
o
x
o
o
o
1
o
o
1
x
x
x
x
o
x
X
X
x
o
x
o
x
x
o
o
o
x
x
o
x
x
1
20
x
o
Tic-tac-toe Game Tree
o
x
X
x
o
x
o
x
o
o
1
x
x
x
O
O
x
o
x
o
x
o
x
o
o
o
1
o
o
1
0
x
x
x
x
o
x
X
X
x
o
x
o
x
x
o
o
o
-1
x
x
o
x
x
1
21
x
o
Tic-tac-toe Game Tree
o
x
X
x
o
x
o
x
0
o
o
1
x
x
x
O
O
x
o
x
o
x
o
x
o
o
o
1
o
o
1
0
x
x
x
x
o
x
X
X
x
o
x
o
x
x
o
o
o
-1
x
x
o
x
x
1
22
x
o
Tic-tac-toe Game Tree
1
o
x
X
x
o
x
o
x
0
o
o
1
x
x
x
O
O
x
o
x
o
x
o
x
o
o
o
1
o
o
1
0
x
x
x
x
o
x
X
X
x
o
x
o
x
x
o
o
o
-1
x
x
o
x
x
1
23
Big Amendment

• In real games, like chess, we cant look all the
  way to the end of the game.
• So we cut the tree off after a few moves, and use
  a heuristic evaluation algorithm to estimate how
  good the position is based on board features
  (e.g., how many rows I have two of my marks in
  vs. how many such rows opponent has).
• Then apply minimax to pass numbers up.

24
Real Intelligence?

• Are all the separate subfields of AI going to
  achieve real computer intelligence?
• Right now, it depends on what the target turns
  out to be. People think their minds are single
  entities with general skills, but were
  discovering more and more specialized modules in
  the brain.
• Genius is rare and perhaps unpredictable

25
Computer Consciousness

• When they build an artificial nose, can it
  really experience the smell of garlic? Does it
  know what turpentine smells like? Would it have
  the same experiences we have?
• Answer It wouldnt experience anything!

26
But

• When we know exactly how information flows
  through the brain, will we conclude that we cant
  experience anything?
• Guess Consciousness will turn out to be a matter
  of modeling oneself as having experiences.

27
The Self-Modeling Nose

• Suppose the nose is connected to a robot that can
  make inferences about what it thought it smelled
  and what was probably there.
• It must distinguish between appearance and
  reality. Appearance in this case is the way
  things seem to smell. At this point its
  helpful to have the concept of odor

28
Qualia

• We know what red looks like, and how it differs
  from green. But theres nothing much else to
  say.
• The brains theory of itself stops with these
  ultimate qualia of things because it has to stop!
  Robots will have to do the same thing (although
  the details will differ a lot).

29
Dont Get the Wrong Idea

• Most AI people hate to talk about consciousness.
  Its too fuzzy.
• This is not what the field is about.
• Its about algorithms and experimentation.

30
Summary

• The brain computes things
• Perhaps thats all it does
• Cognitive science explores this hypothesis
• AI is the branch that explores algorithms for
  difficult computational problems
• If cognitive science succeeds, it may even
  explain consciousness