Foundations of AI and ML

Foundations of AI

• Artificial Intelligence (AI) and Machine Learning (ML) have rapidly increased in popularity and application in recent years.
• AI and ML technologies have become integral to modern software development and innovation.
• This lecture provides a historical overview, fundamental concepts, techniques, and lexicon related to AI and ML.

A Brief History of AI and ML

• AI and ML are rooted in the early 20th century, marked by the work of Alan Turing.

The Turing Test: The Birth of AI

• In 1950, Alan Turing introduced a test to ascertain if a machine's intelligent behavior was indistinguishable from a human's.
• The Turing Test laid the groundwork for AI.
• Turing's work spurred interest in creating machines capable of mimicking human thought.

The Dartmouth Conference: Defining AI

• In 1956, scientists and mathematicians convened at Dartmouth College to discuss the future of AI.
• This conference is considered the birth of AI as a formal academic discipline.
• John McCarthy, Marvin Minsky, and Claude Shannon defined AI as the science of creating machines capable of performing tasks requiring human intelligence.
• In the 1960s and 1970s, several AI programs demonstrated computers' potential for problem-solving, language understanding, and learning.
• The General Problem Solver (GPS) was an AI program that imitated human problem-solving.
• ELIZA was a natural language processing (NLP) program that simulated human conversations.
• SHRDLU was a program capable of understanding and manipulating virtual objects using natural language commands.

The Rise of Machine Learning

• In the 1980s, researchers concentrated on developing algorithms capable of learning from data, which led to ML.
• Teaching machines to learn from data proved more efficient than explicit programming.
• The ID3 algorithm was developed for creating decision trees, marking a milestone in ML research.
• Researchers developed artificial neural networks, inspired by the human brain, that could recognize data patterns.
• Reinforcement learning involved algorithms learning through environmental interaction and feedback.

Deep Learning and Beyond

• The 21st century saw an increased interest in AI and ML driven by computing power, large datasets, and algorithm design.
• Deep learning, a subset of ML, trains large neural networks.
• This led to progress in image and speech recognition, NLP, and game-playing.
• AI and ML are coding components found in web development, data analysis, robotics, and autonomous vehicles.

Core Concepts in AI

• Central to AI is the concept of an intelligent agent.
• An intelligent agent perceives its environment, processes information, and acts appropriately to achieve goals.
• Agents range from rule-based systems to neural networks capable of learning and adapting.

Rule-Based Systems

• A rule-based system is an AI that uses predefined rules to make decisions or perform tasks.
• The system operates using if-then rules to dictate responses to situations or conditions.
• Human experts typically create rules that represent domain knowledge, enabling automated reasoning or problem-solving.

Components of a Rule-Based System

• Knowledge Base: A structured set of rules using "if-then" statements e.g. if the temperature is above 30°C, turn on the air conditioner.
• Inference Engine: Applies rules from the knowledge base to input, determining actions by matching rules to conditions to infer outcomes.
• Working Memory: Stores facts about the current situation, updated as the system processes data and applies rules.

How Rule-Based Systems Work

• Rule Matching: Checks current rules against data in working memory.
• Rule Firing: If conditions are met, the system execute and performs the action.
• Chaining: Involves forward and backward reasoning.
• Forward Chaining: Applies rules to reach conclusion
• Backward Chaining: Starts with goal to determine the rules that need to be acheived

Examples of Rule-Based Systems

• Medical Diagnosis Systems diagnose diseases for example MYCIN system was designed to locate infections
• Customer Support Chatbots uses a simple support chatbot to provide automated responses based on user inputs for example "If the customer asks, "How do I reset my password?", then respond with, "Go to the password reset page and follow the instructions.".
• Home Automation Rule: If the front door opens and it is after 7 PM, then turn on the porch lights.
• Spam Filters Rule: If the subject line contains the word "free" and the body contains the word "offer", then mark the email as spam.

Advantages of Rule-Based Systems

• Transparency behind the reasoning because rules are clearly defined
• Expert Knowledge Encoding: Input domain knowledge through rules.
• Simplicity: Easy for well-understood problems

Limitations of Rule-Based Systems

• Scalability: complex managing rules
• Lack of Learning: They dont learn from data, not adaptable to situations
• Performance Issues: slow where there are many rules

Decision-Making

• Rule-based systems best solve problems where the decision can be found from set rules.
• Often used in medicine, automations, and customer service
• Machine learning better suited for more complex reasoning.

Problem Solving and Search

• AI aims to solve problems using search algorithms to explore and find an appropriate solution.
• These algorithms are classified under these categories:
• Uninformed search: which explores the solution blindly
• Informed search: Knowledge to guide teh search process

Knowledge Representation and Reasoning

• Encoding information about hte world in a form that an AI can understadn is known as Knowledge representation
• This can be done in many steps: Propisitional logic, first-order logic, semantic networks and ontologies
• Reasoning is process of concluding represented knowledge.
• Use deduction, induction, and abduction to reason about the world and make decisions

Techniques for Reasoning: Deduction, Induction, and Abduction

• Deduction, induction and abduction are techniques used to makes decisio
• These are key elements in AI

Deduction

• Conclusions are derived from general premises, to specific decisions or conclusions.
• This should come out to be logically sound
• E.g General Rule is (IF A than B) Specific Case (A is true) Threfore (B is true).

Induction

• Induction dervives from specific observations
• Patterns or trends derive from examples and forming conclusions
• E.g. Observation; A1, A2, and A#, Conculsion: A is generally true

Abduction

• Reasoning to find the the simplesst most likely explination
• Inference to the best "explanstion"
• Unlike induction conclusion does not mean it will be true, but a reasonable hypothisis based on the evidence
• Observation (A is True), Hypothesis (The best explination for A being true (B).

Comparative Analysis: Deduction, Induction, and Abduction

• Deduction: General Rule → Specific Case → Conclusion (Guaratees known and true premises)
• Induction: Specific obervations leading to generation (Generral from Observations. (Conclusions or more pbobable, not certain)
• Abduction; Observations to hypothesis which could be right

How to Use these processes in AI and Machines Leearning

• Deduction: Is the rule based systems where the systems fallow predefined systems-system
• Induction in Machine learning is fundamental where data can be generalized to make more predictions
• Abductions are important for diagnostic systsms.

Further Insight and Application

• Each techinique serves different types of reasoning and decsion making, depending on type desired outcome and data

Machine learning

• A subset of AI algorithms, which allow a computer to learn from data to make predictions
• There are thee types
• Supervised Learning: Trains the data set on a labelled dataset
• Unsupervised Leaning:Learn from patorns for the unlablled dataset
• Reinforcement Learning: Algorithm gets feedback from the environment

Neural Networks and Deep Learning

• ML is inspired through from human brain, these nertworks are interconnected with neurons
• Deep Learning deals with complex models containing multipple layers neurons
• Deep nerotuons are very important for achieving breakthroughs in aplications-Image Recgnition and Speech Processin

Natural Language Processing

• The branch of ai that focus's on enabling computers to read and generate human language
• Involves speech regontition, translations and machine data
• Relise on MI algiritons and linguisitic knowledge to process data

Decoding Machine learning

- Coders must understand the complexities of machine learning to dive deeper into artifical languages
-   ML is a subset of AI and is used ot prove models and aigoritms
There are many Ml and AI APPS

Types of Machine Learning

• Most common ML technique used where algorithms are trained on labeled sets
• Used in speech and speech processing

Supervised algorithm types

• Linear regression, which is used to see the feature of certain values
• Logistics regession for binary classification where emails classify as scam Random forrest , dection trees that clasify with large dataserts

Unsupervised Leaning Algorithm

The program algorithom will be used to look for trends or outliers

Reinforcement Leaning

Agent will leran make desctions through feeback with enviorement and recieving rewards

What are Neural Networks

In A ai a neural network is a computing model inspired by structure with the human brain , it has interconected nodes and neutrons

• An Algrithoms

Define Algorithm

Algorthom is a set of procedure for solving or performing in Ai or MI , making data and making descions.

Features and Models

Feautes include indiviual characaterisrcs and propertys in MI models , is a way to representaion of real world processing. It is devolped through algothims.