Great Ideas in Computer Science

1
Great Ideas in Computer Science

• Artificial Intelligence
• COMP 41 Apr 19 and Apr 24

2
What is Artificial Intelligence?

• Artificial man-made
• What is Intelligence?

3
The Turing Test

• 1950 Turing proposed a test for determining if a
  machine is intelligent
• A machine is intelligent if it can behave in an
  manner indistinguishable from a human

4
The Turing Test
The Tao of Computing by Henry Walker
5
Intelligence?
6
Intelligence?
7
Intelligence?
8
Intelligence?
Conversations taken from Hamlet on the
Holodeck by Janet H. Murray
9
John Searles Counterargument

• 1980 John Searle proposed a counterargument
  against the Turing Test
• a simulation of intelligence is not necessarily
  intelligent

10
Searles Chinese Room
http//www.unc.edu/prinz/pictures/c-room.gif
11
Analysis The Chinese Room

• The human in the Chinese Room behaves exactly as
  a computer program
• follows instructions
• interprets input data
• creates output data
• The human does not understand Chinese!

12
Real AI

• AI research has been successful in more modest
  goals
• rather than creating intelligence, AI has
  focused on clever problem solving

13
Kinds of AI

• Symbolic
• programs that manipulate symbols according to
  well defined algorithms and programs
• the knowledge in symbolic AI is carefully coded
  by humans
• Connectionist
• programs that simulate the brain has a network of
  simple neurons
• neural networks learn to recognize patterns

14
The Water Jug Problem

• You have
• a 3 gallon bottle (b3),
• a 4 gallon bottle (b4) and
• a hose.
• Put exactly two gallons in the 4 gallon bottle.

15
Represent Problem State
state of b3
state of b4
1 gal in b3
2 gal in b4
16
Determine Possible State Changes

• 1. Fill b3
• 2. Fill b4
• 3. Empty b3
• 4. Empty b4
• 5. Move from b3 to b4 until b4 full
• 6. Move from b3 to b4 until b3 empty
• 7. Move from b4 to b3 until b3 full
• 8. Move from b4 to b3 until b4 empty

17
Search for Solution

• Apply all possible changes to current state
• Remove states already seen
• If any state is the goal state, stop
• Repeat for all new states

18
First Level Search
Rule 1
Rule 2
19
Second Level Search
1
2
2
1
7
6
4
3
20
Second Level Search
1
2
2
1
7
6
4
3
X
X
Weve already seen these
21
A Possible Solution
2
7
3
2
7
22
Water Jug System FSA
finite state automata from Simulation Model
Design and Execuction by Paul A. Fishwick
23
Water Jug System Solution
24
A TicTacToe AI

• Problem State represent TTT board as a list of 9
  characters, which might be X, O or B (blank).The
  characters are indexed from 0 to 8, where the
  index indicates positions as followsFor
  example represents

XBBOXBBBB
25
TTT AI 1

• Write down every possible board and the best move
  to make when that board is encountered.
• The AI simply looks for the current board and
  responds with the specified move.

26
TTT AI 1 - Analysis

• This strategy completely encodes some human
  players strategy it will be exactly as good as
  the player (assuming no mistakes in coding)
• There are 83 512 possible board states to be
  encoded
• This AI will be very fast
• This is tedious for TacTacToe and probably not
  practical for more complex games any other game
  will require a complete recoding of moves

27
TTT AI 2

• Determine all possible moves from current board
• If any move wins the game, assign the move the
  highest possible rating.
• Otherwise, for each possible move, determine
  every possible opponent move. See which is worst
  (by recursively applying this procedure).
  Whatever rating that move has, assign it to the
  current move.
• Select the move with the highest rating.

28
TTT AI 2 - Analysis

• This is an example of an AI algorithm called the
  min-max procedure. We attempt to find the move
  that maximizes our chance of winning, while
  minimizing our opponents chance of winning.
• This takes much more computation time than the
  first AI.
• This is more difficult to program, but easier to
  determine the rules.
• This procedure easily adapts to other games (3D
  TicTacToe, checkers, chess, )

29
TTT AI 2 Analysis (continued)

• From the initial empty board, we will have to
  consider as many as 98765432 9!
  362880 moves to determine opening move
• Computers can do 1 billion operations per second,
  easy to consider 632880 moves in a reasonable
  turn time.

30
TTT AI 3

• We can add heuristic rules that can be applied to
  certain situations this are rules that encode
  knowledge or strategies more like human players
• Example always start with center position

31
The Traveling Salesman Problem

• Given a list of cities and traveling distances
  between cities, determine the shortest route that
  takes a salesman to every city

32
Traveling Salesman AI 1

• Generate every possible route
• Compute the distance of every route
• Select the shortest route
• For n cities, this requires evaluation of n!
  routes.

33
TS AI 1 - Analysis

• For n cities, this requires evaluation of n!
  routes. 10! 3,628,80020!
  2,432,902,008,176,640,000
• This gets too expensive for even a small number
  of cities!

34
TS AI 2

• 1. Arbitrarily select a starting city
• 2. Look at all cities that have not yet been
  visited and select the nearest one
• 3. Repeat step 2 until all cities have been
  visited

35
TS AI 2 - Analysis

• This is a heuristic algorithm
• This solution will produce a short route, but not
  necessarily the shortest route (the maximum
  error can be computed)
• For n cities, this executes in time proportional
  to n2, which is considerable faster than n!

36
TS AI Analysis continued
n n2 n!
5 25 120
10 100 3,628,800
20 400 2 x 1018
100 10,000 10158
1000 1,000,000 102564