AI Techniques

AI techniques, or artificial intelligence techniques, refer to the methods, algorithms,

and approaches used to create and apply artificial intelligence systems. These
techniques enable machines to mimic human-like intelligence, learn from data, make
decisions, solve problems

There are three important AI techniques:-

• Search Algorithms
• Knowledge Representation
• Abstraction

1. Search Algorithms: This technique provides a way of solving problems for

no more direct approach is available as well as a framework into which any
direct techniques can be embedded.

Search algorithms in AI are methods used to systematically explore and navigate

problem spaces to find solutions. These algorithms aim to find a sequence of actions
that lead from an initial state to a goal state.

• Types of Search Algorithms:

o Breadth-First Search (BFS): Explores all possible nodes at the
current depth before moving to the next level.
o Depth-First Search (DFS): Explores as far as possible along one
branch before backtracking.
o A Search:* Considers both the cost of reaching a node and the
estimated cost from the node to the goal.
o Greedy Best-First Search: Expands the node that is closest to the
goal based on a heuristic function.
• Applications: Search algorithms are utilized in various domains, including
route planning in GPS systems, game playing (like chess or Go), scheduling
problems, and problem-solving tasks in robotics.

2. Knowledge Representation: This technique provides a way of solving

complex problems by exploiting the structures of the objects that are involved.

Knowledge representation in AI involves structuring and organizing knowledge in a

way that machines can understand and use it to solve complex problems or reason
intelligently.

• Techniques of Knowledge Representation:

o Semantic Networks: Representing knowledge using nodes (concepts)
and links (relationships).
o Frames: Structuring knowledge using a collection of attributes
associated with an entity.
o Logic-Based Representation: Expressing knowledge using formal
logic, such as predicate logic or propositional logic.
 • Applications: Knowledge representation is critical in expert systems, natural
language processing, database systems, reasoning systems, and semantic web
technologies.

3. Abstraction: This technique provides a way of separating important features

and variations from the many important ones. It hides the details of something.

Abstraction in AI involves simplifying complex systems or problems by focusing on

essential details while ignoring unnecessary specifics. It helps in managing
complexity and understanding problems at different levels of detail.

• Levels of Abstraction:
o Data Abstraction: Hiding unnecessary details and exposing only
relevant data or information.
o Procedural Abstraction: Representing a sequence of actions as a
single procedure or function.
o Hierarchical Abstraction: Organizing information into levels of
importance or generality.
Level of Model
AI Models:

It refers to the mathematical or computational representation that AI uses to learn

patterns from input data and make predictions or perform tasks.

AI models or artificial intelligence models are programs that detect specific patterns
using a collection of data sets. It is an illustration of a system that can receive data
inputs and draw conclusions or conduct actions depending on those conclusions. Once
trained, an AI model can be used to make future predictions or act on data that was
not previously observed. AI models can be used for a variety of activities, from image
and video recognition to natural language processing (NLP), anomaly detection,
recommender systems, predictive modeling and forecasting, and robotics and control
systems.

Level of AI Model:

AI, categorizing models into different levels helps understand their complexity,
sophistication, and the tasks they can accomplish. Here's an overview of the first and
second-level models:

First-Level Models:

Definition: First-level models in AI refer to basic or foundational models that serve

as initial building blocks for more complex systems. These models are relatively
simple, focusing on fundamental concepts or basic structures.

A class of computer programs that aim to solve problems in a way that might seem
straightforward for a computer but is challenging for humans. These programs fall
into the first class of AI tasks but may not align with the typical human definition of
an AI task.

• Characteristics:
o Simple Representations: They involve straightforward representations or
concepts.
o Limited Complexity: These models have fewer layers or components
compared to higher-level models.
o Basic Functionality: They perform elementary tasks or have limited
capabilities.
• Examples:
1. Perceptrons: The basic unit of a neural network, handling simple binary
classification problems.
2. Linear Regression: Models the relationship between two variables using a
linear equation.
3. Basic Decision Trees: Simple tree structures for classification or regression
tasks.
Second-level models

In AI are like the advanced superheroes of artificial intelligence, going beyond the
basics to handle more complex tasks. Imagine them as smarter and more capable
versions. One type focuses on mimicking human behavior and thinking. These
programs tackle tricky tasks like understanding human-like reasoning, answering
questions from texts, and even simulating specific behaviors, like paranoia, for
psychological testing. They aim to make computers more transparent, meaning they
can explain their thinking in a way that people can understand and trust. Another
type of advanced AI taps into what we know about how humans think. It tries to copy
human reasoning, sometimes down to the details of individual neurons. Early versions
had some struggles, though, as simple attempts to mimic our intelligence faced
theoretical roadblocks, and the super parallel nature of human brains posed
challenges for regular computers, leading to the exploration of special parallel
computing models.

3. Evolution of Models:

The evolution of AI models is like a superhero upgrade. From basic models, we moved
to advanced, second-level ones. They handle complex tasks like mimicking human
behavior and reasoning, even explaining their thinking.

This evolution faced challenges, like early models struggling and the need for special
parallel computing. It's a journey from AI's baby steps to sophisticated problem-solving
superheroes.
Criteria for Success
Success in Artificial Intelligence (AI) is measured through various benchmarks and
evaluations. One prominent criterion is the Turing Test, proposed by Alan Turing,

Turing Test:

• Purpose: A benchmark proposed by Alan Turing to determine if a machine exhibits

intelligent behavior akin to humans.
• Method: Involves human interrogators interacting with both a human and a
machine through text-based communication without knowing which is which.
Success is achieved if the machine can convincingly impersonate a human.

It tests a machine's ability to mimic human-like behavior in interactions. Evaluating

AI performance poses challenges due to the complexity of human-like knowledge
required, especially in nuanced conversations and reasoning. For specific tasks,
measurements can be precise, such as evaluating chess program ratings or assessing
time efficiency gains in technical configurations. Assessments might focus on the
efficiency or cost-effectiveness of AI systems compared to human equivalents in
technical domains or simply checking if the program's performance in everyday tasks,
like paraphrasing, matches human capability. When AI aims to simulate human
behavior, success is judged by how closely the program's behavior mirrors human
abilities, often assessed through experiments and protocol analyses. Despite the
elusive definition of general intelligence for machines, establishing clear criteria
within specific domains remains crucial for assessing AI performance.
State Space Search:

State space search is a fundamental concept in Artificial Intelligence (AI) that

involves navigating through a set of possible states from an initial state to a goal state
by employing various search algorithms. It's like exploring a maze to find the exit,
where each decision leads to a new state until the desired outcome is achieved.

Key Components:
1. State:

• Represents a particular configuration or situation.

• Examples could be positions in a maze, board configurations in a game, or any
definable situation in a problem-solving scenario.

2. Initial State:

• The starting point of the search.

3. Goal State:

• The desired or target state to be reached.

4. Operators/Actions:

• Actions or operations that can be applied to move from one state to another.
• These actions determine how the state transitions occur.

5. Transition Model:

• Defines the result of applying actions to states, determining the state reached after
acting in a given state.
Water Jug Problem

Problem Definition:

In the water jug problem we are given two jugs, a 4-liter one and a 3-liter one. Neither
has any measuring markers on it. There is a pump that can be used to fill the jugs with
water. How can you get exactly 2 liters of water into a 4-liter jug?

Defining Problem in State Space Approach:

The state space is a set of ordered prearranged pairs given the number of gallons of
water in both the chugs at anytime. i.e. (x, y) where x=0,1,2,3,4 and y=0,1,2,3.

“x” represents the number of gallons of water in the 4-gallon jug.

“y” represents the number of gallons of water in the 3-gallon jug.

The start state is (0, 0) and the goal state is (2,n) for any value of n but n is limited to
holding form 0 to 3 because it’s a 3-liter gallon.

Assumptions:

• We can fill the jugs fully.

• We can empty the jugs fully.
• We can pour/swap the contents of jugs fully.
• We can pour/swap the contents of jugs partially.:
•
•
One Solution for the Water Jug Problem:
Production System: A Simple Explanation

A production system is like a smart assistant that helps computers make decisions and solve
problems efficiently. It's a structured framework that guides the process of finding solutions,
especially when a computer needs to search through information. Imagine it as a set of
organized rules and processes for smart problem-solving.

Components of a Production System:

1. Rules:

A set of rules each consisting of a left side that determines the applicability of the rule and a
right side that describes the operation to be performed if the rule is applied.

2. Knowledge/Databases:

- Knowledge or databases are like the computer's memory. They store important
information needed for solving problems.

- These databases contain facts collected during the problem-solving process.

- It can store both permanent and temporary information that is relevant to the current
task.

3. Control Strategy:

- The control strategy is the plan that tells the computer the order in which to use the rules
and how to handle conflicts.

- It dictates the sequence in which rules are checked against the facts in the database.

- When there are multiple rules or conflicting information, the control strategy guides the
computer on what to do next.

4. Rule Applier:

- The rule applier is like the action taker. It executes the rules based on the control
strategy.

- It applies the rules to the knowledge/database to make decisions and find solutions.
Characteristics

Monotonic Production System:

A monotonic production system is like a set of rules that always sticks to its decisions. Once
it makes a conclusion based on the available facts and rules, it won't change its mind, even if
new information comes in. It's like having a rulebook that, once followed, doesn't go back
and revise its decisions. While this simplicity can be straightforward, it might not capture
changes in the real world where situations evolve and new facts might require
reconsideration.

Non-Monotonic Production System:

In contrast, a non-monotonic production system is more flexible. It's willing to update its
conclusions when faced with new information. Imagine it as a system that doesn't mind
revisiting its initial decisions, adjusting them based on the evolving situation. This
adaptability makes it more suitable for handling dynamic scenarios where things can change
over time.

Commutative Production System:

A commutative production system is like a set of rules that doesn't care about the order in
which you apply them. It's as if you have a bag of rules, and no matter in which sequence
you pick and apply them, you'll get the same result. This property simplifies the system,
making it less dependent on the specific order of rule application.

Partially Commutative Production System:

On the other hand, a partially commutative production system allows some flexibility in the
order of rule application but not complete freedom. It's like having certain rules that can be
applied in any order, while others may have dependencies and need a specific sequence.
This middle ground provides a balance between flexibility and structure, accommodating
various problem-solving scenarios.
TWO PATH PROBLEM

The "two-path problem" in AI refers to a situation where an intelligent system or agent faces
a decision-making challenge due to having two equally promising or indistinguishable paths
to choose from. This scenario often occurs in search and decision-making processes where
multiple options appear to be equally viable based on the available information or criteria.

In the context of search algorithms, when exploring potential solutions or paths, the system
may encounter two or more paths that seem equally optimal or have identical evaluation
scores. This creates ambiguity for the system, making it difficult to determine which path to
prioritize.

The two-path problem can introduce challenges in decision-making systems, especially when
the consequences of choosing one path over another are unclear or when there's a need to
break the tie and proceed along a single path.

When an AI faces the two-path problem, where it has to choose between two equally good
options, it uses extra rules or tricks to make a decision. Imagine it's like having two equally
delicious ice cream flavors, and you can't decide which one to pick. To help you choose, you
might use a coin flip (randomness), think about which one is healthier (considering
secondary objectives), or remember a rule like "always pick the one on the left" (utilizing
specific rules). These extra tricks or criteria help the AI break the tie and decide on one path
over the other.
 CHAPTER - KNOWLEDGE REPRESENTATION ISSUES

Knowledge representation in Artificial Intelligence refers to the process of designing

a system to store and organize information in a way that a computer can utilize it to
solve complex problems. It involves capturing knowledge about the world and
making it available for reasoning and decision-making by an AI system.

Imagine you want a computer to understand facts about the world, relationships
between objects, and how things work. Knowledge representation is about finding
ways to structure this information in a format that a computer can understand and
use effectively.
Semantic Networks: These represent relationships between concepts using nodes and links.
For example, in a semantic network, you might have nodes for "cat," "has," and "tail," with
links connecting them to represent the relationship "cat has tail."

Frames: Frames are structures that contain information about a certain object or concept,
including its attributes and relationships. If you have a frame for a car, it might include
information about its color, model, and manufacturer.
The frame problem

The frame problem in artificial intelligence refers to the challenge of determining

and updating relevant information in a changing environment. It arises when a
system, such as a computer or robot, needs to understand what aspects of its
knowledge should be revised or considered when faced with new situations or
changes.
Essentially, the frame problem involves teaching machines to focus on the most
essential details, avoiding unnecessary computations or updates, and efficiently
adapting to evolving circumstances.
It is a crucial issue in AI, as solving it enhances the system's ability to navigate
dynamic environments and make informed decisions without being overwhelmed by
irrelevant information.

Advantages:
1. Enhanced Efficiency: AI systems process data quickly, boosting overall task
efficiency.
2. Precision and Accuracy: AI performs tasks with high precision, minimizing errors.
3. 24/7 Availability: AI operates continuously, providing round-the-clock services.

Disadvantages:
1. Job Displacement: Automation by AI may lead to job loss in certain industries.
2. Ethical Concerns: AI systems can introduce biases and privacy issues.
3. Lack of Creativity: AI lacks human intuition and creativity in problem-solving.
Semantic network

A semantic network is a structured representation of knowledge that uses nodes

(representing concepts) and links (representing relationships) to express
relationships between different concepts. In simpler terms, it's like creating a visual
map of ideas and how they relate to each other.

In a semantic network, nodes could represent objects or ideas, and the links
between them indicate how these objects or ideas are connected. For example, if we
have nodes for "cat" and "has," the link between them could represent the concept
"cat has a tail." This visual representation helps organize and understand the
relationships between different pieces of information.

Semantic networks are commonly used in artificial intelligence for knowledge

representation because they provide a clear and intuitive way to express complex
relationships and ideas. They are like diagrams that help computers understand how
various concepts in the world are connected to each other.

diagram

Advantage:
1. Intuitive Representation:
Semantic networks provide an intuitive and visual way to represent complex
relationships between concepts, aiding in understanding.

Disadvantage:
1. Scalability Challenges:
As the amount of information grows, managing and scaling semantic networks can
become complex and challenging.
 CHAPTER 2
SYMBOLIC REASONING AND UNCERTAINITY

Symbolic reasoning is a way of teaching computers to understand and solve

problems by using symbols. Imagine each piece of information as a symbol, like a
word or an icon. Now, think of logical rules as guidelines for the computer, telling it
how to manipulate these symbols to find solutions. It's almost like teaching a
computer to speak a language of symbols. This approach makes it simpler for
computers to think through complex problems and come up with conclusions by
following the logical rules associated with those symbols. It's a bit like giving
computers their own language to process and solve tasks more effectively.

UNCERTAINITY:
MONOTONIC REASONING

Monotonic reasoning in AI is like a computer making a decision and sticking to it, no

matter what. Once it forms a conclusion based on the available information, it
doesn't change that decision, even if you provide more details later. It's a rigid
approach where the initial inference remains fixed. For instance, if you tell it all birds
can fly, and it concludes penguins can fly, even if you later say penguins can't, the
computer won't revise its first decision. Monotonic reasoning is simple but doesn't
adapt to new information, making it less flexible in handling changes or updates.

NON - MONOTONIC REASONING:

Non-monotonic reasoning in artificial intelligence is an approach where a computer's

conclusions are not set in stone and can be revised with new information. Unlike
monotonic reasoning, which sticks to initial decisions, non-monotonic reasoning
allows for adjustments.
For example, if we initially assume that all birds can fly and conclude penguins can
fly, a non-monotonic system is open to changing its conclusion if we later provide
information that penguins cannot fly. It's a way for computers to be flexible,
adapting their understanding as more details become available, making them better
suited for handling evolving situations and dynamic information.
LOGIC FOR NON MONOTONIC REASONING :
Non-monotonic reasoning in artificial intelligence often involves using specific logics
designed to handle situations where conclusions may need to be revised or adjusted
based on new information. One such logic is Default Logic. Here's a brief
explanation:

Default Logic for non-monotonic reasoning:

Default Logic:
Default Logic is a logical framework used in non-monotonic reasoning, a field of
artificial intelligence concerned with reasoning under uncertainty and the need to
revise conclusions when new information arises. In Default Logic, there are default
rules that express general assumptions or defaults. These defaults capture what is
typically true but can be overridden or revised by specific information.

Example:
Consider the default assumption that "birds can fly." In Default Logic, this is a default
rule representing a generalization. Now, if we encounter a specific bird, like a
penguin, which cannot fly, Default Logic allows us to override the default
assumption. The system revises its conclusion based on the specific case,
acknowledging that not all birds adhere to the default rule.

Defeasible Reasoning:
A key feature of Default Logic is defeasible reasoning. This means that conclusions
reached based on default rules are open to revision or defeat. If new information
contradicts or provides exceptions to the default assumption, the system can adapt
and revise its conclusions.

Significance in AI:
Default Logic is particularly significant in AI because it provides a formalism for
representing and handling common-sense reasoning. It allows AI systems to make
assumptions, recognizing that these assumptions might not always hold true. The
ability to revise conclusions in the face of new information enhances the adaptability
and flexibility of AI systems, making them more capable of handling real-world
complexities.
IMPLEMENTATION ISSUES OF NON MONO TONIC REASONING:
Implementing non-monotonic reasoning in artificial intelligence poses several
challenges and issues that need careful consideration.

Implementation issues related to non-monotonic reasoning in artificial

intelligence:

1. Handling Changes in Information:

- Challenge: Making sure the system can adjust its conclusions when new
information contradicts or modifies default assumptions.
- Example- If the initial assumption is "birds can fly," the system should adapt when
encountering a specific bird, like a penguin, which cannot fly.
2. Dealing with Complexity:
- Challenge: Managing the computational complexity of handling various rules and
exceptions without slowing down the system.
- Example: Ensuring that the reasoning process remains efficient even when
dealing with a large amount of information.
3. Representing Knowledge Clearly:
- Challenge: Designing a clear and expressive way to represent knowledge that
includes default assumptions and specific cases.
- Example: Creating a system that can easily understand and work with rules like
"by default, fish can swim."
4. Resolving Conflicts:
- Challenge: Addressing inconsistencies that may arise when default assumptions
conflict with specific instances.
- Example: Deciding what to do when the assumption "all animals breathe air"
clashes with the specific case of a fish, which breathes underwater.
5. Adapting and Learning:
- Challenge: Allowing the system to learn from experience and update its
assumptions based on feedback.
- Example: Modifying the default assumption about a dog being aggressive if the
system learns about a friendly dog through interactions.
ARGUMENTING A PROBLEM SOLVER

Arguing for a problem solver in AI is like showing off a superhero for specific tasks.
First, clearly state the problem it tackles. Then, explain how the problem solver
works, like the special powers it uses (algorithms and methods). Show examples of it
in action, like saving the day with real results. Highlight that it can handle bigger
challenges and adapts to different situations (like a superhero facing different
villains). Talk about how efficiently it uses resources, making it practical. If it learns
and improves over time, that's a plus. Share stories of it solving real-world problems.
Compare it to other options, proving it's the superhero you need. Mention any
weaknesses honestly, and be aware of any ethical concerns. Essentially, present it as
the reliable superhero solution to the problem at hand.