0% found this document useful (0 votes)
2 views41 pages

Chapter 2

The document discusses problem-solving with AI, emphasizing the formulation of real-world problems as search problems and the importance of defining choices and consequences. It categorizes problems into single-agent planning and two-player games, illustrating concepts with examples like the chicken crossing puzzle and game trees. Additionally, it covers historical milestones in AI, search techniques, and real-world applications of AI in planning and decision-making.

Uploaded by

sewnegn238
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
2 views41 pages

Chapter 2

The document discusses problem-solving with AI, emphasizing the formulation of real-world problems as search problems and the importance of defining choices and consequences. It categorizes problems into single-agent planning and two-player games, illustrating concepts with examples like the chicken crossing puzzle and game trees. Additionally, it covers historical milestones in AI, search techniques, and real-world applications of AI in planning and decision-making.

Uploaded by

sewnegn238
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Problem-Solving with AI

Getachew Mamo (Ph.D.)

November 16, 2025

Getachew M. Problem Solving November 16, 2025 1 / 40


Table of contents

1 Search Strategies and Problem-Solving Methods

2 AI Approaches to Solving Real-World Problems

3 Search in Games and Strategic Decision-Making

Getachew M. Problem Solving November 16, 2025 2 / 40


What is AI Problem Solving?

Many real-world problems can be formulated as search problems


Requires defining:
Alternative choices
Their consequences
Goal states
Examples: Navigation, game playing, puzzle solving

Key Insight
The first stage - defining choices and consequences - is often the most
challenging part and requires careful thinking.

Getachew M. Problem Solving November 16, 2025 3 / 40


Real-World Example: Navigation

Getting from point A to B using public


transport
Similar problems solved by:
Self-driving cars
Game-playing AI (chess, etc.)
Delivery route optimization

Figure: Navigation app solving


route finding

Getachew M. Problem Solving November 16, 2025 4 / 40


Multiple Paths to Solution

Path 2
Path 1

Start Goal

Path 3
Path 4

Figure: Multiple paths can lead to the same goal

Different solutions may vary in:


Time efficiency
Cost effectiveness
Resource usage
Reliability
Advanced search algorithms help find optimal paths
Getachew M. Problem Solving November 16, 2025 5 / 40
Two Main Problem Categories

Category 1: Single-Agent Planning


Static environments
Only one ”agent” making decisions
Examples: Navigation, puzzle solving
Focus: Find optimal sequence of actions

Category 2: Two-Player Games


Competitive environments
Two ”agents” competing against each other
Examples: Chess, checkers, Go
Focus: Strategic decision making considering opponent’s moves

Getachew M. Problem Solving November 16, 2025 6 / 40


Toy Problem: Chicken Crossing

Scenario: Robot must transport fox,


chicken, and chicken-feed across river
Constraints:
Fox eats chicken if left alone together
Chicken eats chicken-feed if left alone
together
Robot can prevent eating when
present Figure: Chicken crossing puzzle
Boat fits robot + 2 items setup

Getachew M. Problem Solving November 16, 2025 7 / 40


State Representation

Five movable objects: Robot, Boat, Fox, Chicken, Chicken-feed


Boat always with robot → 4 independent items
Each can be on Near (N) or Far (F) side
Total possible states: 24 = 16 combinations

State Notation
NNNN: All on near side
FFFF: All on far side (goal)
NFNF: Robot near, Fox far, Chicken near, Feed far

Getachew M. Problem Solving November 16, 2025 8 / 40


Valid States Analysis
State Robot Fox Chicken Feed
NNNN Near Near Near Near
NNNF Near Near Near Far
NNFN Near Near Far Near
NFNN Near Far Near Near
NFNF Near Far Near Far
FNFN Far Near Far Near
FNFF Far Near Far Far
FFNF Far Far Near Far
FFFN Far Far Far Near
FFFF Far Far Far Far

Table: Valid states after removing forbidden configurations

Forbidden States
States where fox-chicken or chicken-feed are left alone together: NFFN,
NFFF, FNNF, FNNN, NNFF, FFNN
Getachew M. Problem Solving November 16, 2025 9 / 40
State Transition Diagram

NNNN NNNF NNFN NFNN NFNF

FNFN FNFF FFNF FFFN FFFF

Figure: State transition diagram for chicken crossing

Getachew M. Problem Solving November 16, 2025 10 / 40


Solution Path

NNNN NNNF NNFN NFNN NFNF

FNFN FNFF FFNF FFFN FFFF

Figure: Optimal solution path highlighted

Getachew M. Problem Solving November 16, 2025 11 / 40


Solution

Solution Sequence
NNNN → FFFN → NFNN → FFFF
Take fox and chicken to far side
Bring chicken back to near side
Take chicken and feed to far side

Getachew M. Problem Solving November 16, 2025 12 / 40


Key Terminology
State Space
The set of all possible situations or configurations in a problem
Example: 10 valid states in chicken crossing
Navigation: All reachable locations from starting point

Transitions
Possible moves between states with a single action
Direct moves only (no multi-step sequences)
Can be symmetric or directional

Costs
Measures associated with transitions
Distance, time, resources, risk
Used to find optimal solutions
Getachew M. Problem Solving November 16, 2025 13 / 40
Problem Solving Process

1 Define State Space


Identify all possible configurations
Remove invalid/forbidden states
2 Identify Transitions
Determine possible moves between states
Assign costs if applicable
3 Define Goal State
Clear criteria for solution
4 Find Path
Use search algorithms
Consider optimality criteria

Getachew M. Problem Solving November 16, 2025 14 / 40


Table of contents

1 Search Strategies and Problem-Solving Methods

2 AI Approaches to Solving Real-World Problems

3 Search in Games and Strategic Decision-Making

Getachew M. Problem Solving November 16, 2025 15 / 40


The Origins of AI

AI’s Deep Roots


Artificial Intelligence is arguably as old as computer science itself. The
concept of automatic reasoning and intelligence predates modern
computers.

Early thinkers: Alan Turing, John McCarthy


Fundamental question: Can thinking be automated?
Turing’s insight: Computation = Automated reasoning

Getachew M. Problem Solving November 16, 2025 16 / 40


Alan Turing’s Contributions

Turing Machine (1936): Theoretical


model of computation
Programmable Computers: Same
hardware, different tasks
WWII Codebreaking: Practical
application at Bletchley Park
Turing Test (1950): Operational Theoretical Turing Machine
definition of intelligence

Getachew M. Problem Solving November 16, 2025 17 / 40


Example: Turing Machine Concept

Simple Turing Machine for Addition


Input: Two numbers in unary (e.g., 3 = 111, 2 = 11)
Tape: 111011 (3 + 2 separated by 0)
Rules: Replace 0 with 1, then remove last 1
Output: 11111 (5 in unary)

Read/Write Head
Move Left Move Right
Current Cell
0=3011=3012=3013=3014=3015=3016=3017=301
Tape Cells (Memory Storage)

Getachew M. Problem Solving November 16, 2025 18 / 40


Birth of Modern AI: 1956 Dartmouth Conference

John McCarthy’s Conjecture


“The study is to proceed on the basis of the conjecture that 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.”

Father of AI: Coined term ”Artificial Intelligence”


Key Insight: Intelligence can be broken into mechanical steps
Philosophical Stance: Bypasses Chinese Room objections

Getachew M. Problem Solving November 16, 2025 19 / 40


Games as AI Testbeds

Why games became central to AI:


Well-defined rules and objectives
Controlled, restricted domains
Clear success metrics
Progressive complexity

AI
Game Progression

Getachew M. Problem Solving November 16, 2025 20 / 40


Historical AI Game Milestones

Game Year Significance


Checkers 1950s First board game AI
Chess 1997 Deep Blue beats Kasparov
Go 2016 AlphaGo defeats Lee Sedol
Poker 2017 Libratus beats professionals

Complexity Comparison
Chess: ≈ 1047 possible games
Go: ≈ 10170 possible games
Poker: Imperfect information challenge

Getachew M. Problem Solving November 16, 2025 21 / 40


Search in AI Problems

Search as Fundamental AI Technique


Many AI problems can be framed as search problems:
Finding optimal paths
Game move selection
Planning sequences of actions
Constraint satisfaction

Example
Route Planning:
States: Locations on map
Actions: Move between connected locations
Goal: Find shortest path from A to B

Getachew M. Problem Solving November 16, 2025 22 / 40


Example: Simple Pathfinding Problem

4
A B

1 2

C D E
3 2
Possible paths:
Problem: Find shortest path A→E A-B-D-E: 8
A-C-D-E: 6 ← Optimal

Getachew M. Problem Solving November 16, 2025 23 / 40


Minimax Algorithm

Game Theory Foundation


Two-player zero-sum games
Assumption: Opponent plays optimally
Goal: Maximize minimum guaranteed outcome

Example
Tic-Tac-Toe:
MAX player (X), MIN player (O)
Build game tree of possible moves
Propagate values from terminal states

Getachew M. Problem Solving November 16, 2025 24 / 40


Minimax Example

Max
3 1 4

Min Min Min

3 5 2 8 1 7 4 6

Minimax Calculation:
Min nodes choose minimum of children: Min1 = min(3,5,2) = 2
Min2 = min(8,1) = 1
Min3 = min(7,4,6) = 4
Root (Max) chooses maximum: max(2,1,4) = 4
Getachew M. Problem Solving November 16, 2025 25 / 40
Alpha-Beta Pruning

Optimizing Minimax
Alpha: Best already explored option for MAX
Beta: Best already explored option for MIN
Prune branches that cannot affect final decision

Example
Efficiency Gain:
Without pruning: Search entire game tree
With pruning: Skip irrelevant branches
Can double search depth in same time

Getachew M. Problem Solving November 16, 2025 26 / 40


Alpha-Beta Pruning Example

max(5,4)=5

MAX
α=-
β=+
min(5,6)=5 min(7,4)=4

MIN MIN
α=- α=-
β=+ β=+

PRUNED
5 6 7 5 No need to
4
evaluate

Getachew M. Problem Solving November 16, 2025 27 / 40


Alpha-Beta Pruning Example ...

Explanation:
Blue circles = MAX nodes (try to maximize score)
Red circles = MIN nodes (try to minimize score)
Green squares = Terminal nodes (final scores)
Red dashed line = Pruned branch (not evaluated)
After MIN node A returns 5, and MIN node B finds value 4, the third
branch (B3) is pruned because it cannot give a better value for MAX
After evaluating A1=5, A2=6: MIN(A)=5
After B1=7: MIN will be 7
After B2=4: MIN(B)4, which is worse than MIN(A)=5
B3 can be pruned - won’t affect MAX’s decision

Getachew M. Problem Solving November 16, 2025 28 / 40


From Classic Search to Modern AI

Evolution of Search Techniques


1950s-60s: Basic search algorithms
1970s-80s: Heuristic search, planning
1990s-2000s: Machine learning integration
2010s-present: Deep learning, reinforcement learning

Example
AlphaGo (2016):
Combines Monte Carlo Tree Search with neural networks
Learned from human games and self-play
Defeated world champion Lee Sedol 4-1

Getachew M. Problem Solving November 16, 2025 29 / 40


Real-World Applications

Planning & Scheduling: Game-Based Research:


Logistics and delivery routes Autonomous vehicles
Manufacturing processes Robotics navigation
Project management Cybersecurity
Resource allocation Drug discovery
Fundamental Principle
The same search concepts that power game AI underlie many practical
optimization and decision-making systems.

Getachew M. Problem Solving November 16, 2025 30 / 40


Table of contents

1 Search Strategies and Problem-Solving Methods

2 AI Approaches to Solving Real-World Problems

3 Search in Games and Strategic Decision-Making

Getachew M. Problem Solving November 16, 2025 31 / 40


The Game Scenario

We focus on two-player, perfect-information games.


Examples: Tic-Tac-Toe, Chess, Checkers, Go.
Players: Max (Maxine) and Min (Minnie).
Max aims to maximize the score.
Min aims to minimize the score.

O O
X X
O

The Puzzle: Why is Max (X) in despair? Min is about to get three O’s
on the top row, but Max can easily block it. Is there a deeper reason?

Getachew M. Problem Solving November 16, 2025 32 / 40


What is a Game Tree?

A game tree is a directed graph representing all possible game states.


The root node is the initial state (e.g., empty board).
Child nodes are states reachable by a legal move.
Leaf nodes are terminal states (win, lose, or draw).

Root

Child 1 Child 3

Leaf Child 2 Leaf

Leaf Leaf

Getachew M. Problem Solving November 16, 2025 33 / 40


Assigning Values to Outcomes

We assign numerical values to game outcomes:


Max wins: Value = +1
Min wins: Value = -1
Draw: Value = 0

Key Idea:
At Max’s turn, choose the move that maximizes the value.
At Min’s turn, choose the move that minimizes the value.

Getachew M. Problem Solving November 16, 2025 34 / 40


Sample Game Tree Analysis

1
Min

2 4
Max 3 Max
Max
5 6 7 8 9 10
Min Min Min Min Min Min

11 12 13 14 15 16 17 18
-1 -1 +1 -1 +1 -1 +1 -1

Colors: Orange = Min’s turn, Blue = Max’s turn, Green/Red =


Terminal state (Win/Loss for Max).
Goal: Min minimizes the score, Max maximizes it.
The optimal move for Min at the root (Node 1) is to choose Node 3
or 4, which guarantee a minimum score of -1, avoiding the +1
outcome
Getachew [Link] from Node 2. Solving
Problem November 16, 2025 35 / 40
Getachew M. Problem Solving November 16, 2025 36 / 40
Minimax Value Propagation
Step-by-step evaluation:
Level 2: Min nodes take the minimum of their leaf children
Node 5: min(−1, −1) = −1
Node 6: min(+1, −1) = −1
Node 7: min(+1, −1) = −1
Node 8: min(+1, −1) = −1
Node 9: min(+1, −1) = −1
Node 10: min(+1, −1) = −1
Level 1: Max nodes take the maximum of Min children
Node 2: max(−1, −1) = −1
Node 3: max(−1, −1) = −1
Node 4: max(−1, −1) = −1
Root (Node 1, Min):
min(−1, −1, −1) = −1

Final Result
Game Value = −1 ⇒ Min has advantage
Getachew M. Problem Solving November 16, 2025 36 / 40
The Minimax Algorithm

Minimax Principle
Max nodes: Choose the child with the maximum value.
Min nodes: Choose the child with the minimum value.

Algorithm Pseudocode
1 if maximizingPlayer: value = - for each child of node: value =
max(value, minimax(child, depth-1, False)) return value else:
value = + for each child of node: value = min(value,
minimax(child, depth-1, True)) return value
2

Getachew M. Problem Solving November 16, 2025 37 / 40


The Problem: Massive Game Trees

Game Branching Factor Nodes after 10 moves


Tic-Tac-Toe ≈4 ≈ 1 million
Chess ≈ 35 ≈ 2.7 quadrillion
Go ≈ 250 ≈ 1024

The Combinatorial Explosion


Chess: 3510 ≈ 2.7 × 1015 nodes
Go: 25010 ≈ 9.5 × 1023 nodes
Impossible to search entire trees for complex games!

Getachew M. Problem Solving November 16, 2025 38 / 40


Solutions for Massive Trees

Depth-Limited Search
Stop search at certain depth
Use heuristic evaluation function to estimate position value

Chess Heuristics Example


Material: Queen=9, Rook=5, Bishop=3, Knight=3, Pawn=1
Position: Control center squares
King safety: Keep king protected
Mobility: Number of legal moves

Real-World Success
IBM’s Deep Blue (1997) used depth-limited Minimax with
sophisticated heuristics
Defeated world champion Garry Kasparov
Getachew M. Problem Solving November 16, 2025 39 / 40
Exercise: Analyze the Original Position

O O X

X X

Given:
Max (X) must block top row → places X in top-right
Now it’s Min’s (O) turn
Build game tree from this position
Apply Minimax to find game value
Possible answers: +1 (Max wins), -1 (Min wins), 0 (Draw)

Getachew M. Problem Solving November 16, 2025 40 / 40

You might also like