Understanding Artificial Intelligence

• Artificial Intelligence (AI) focuses on creating systems capable of performing tasks that require intelligence
• This involves systems that can think and act like humans or in a rational, goal-oriented manner
• AI includes mimicking human thought and behavior, and focusing on logical and optimal decision-making

Four Approaches to Defining and Understanding AI

• The four fundamental approaches are categorized along two dimensions: thinking vs. acting, and human-centered vs. rational-centered
• These frame different perspectives and methodologies for developing AI systems
• Thinking vs. Acting: Focuses on how AI thinks (reasoning) or acts (behavior)
• Human-centered vs. Rational-centered: Emulating human capabilities versus achieving ideal, rational performance

Thinking Humanly

• AI focuses on replicating human thought processes
• This approach relies on psychology and cognitive science to understand how humans think

Thinking Rationally

• AI focuses on reasoning and logic to achieve ideal performance
• Grounded in formal logic, mathematics, and computational theory

Acting Humanly

• AI focuses on creating machines that behave like humans
• Emphasizes mimicking human actions, often evaluated through the Turing Test

Acting Rationally

• AI focuses on creating agents that act in a way to maximize success or achieve rational goals
• Systems are designed to operate effectively based on what they know, without necessarily emulating humans

AI Historical Context and Methods

• AI approaches evolved through contributions from diverse fields like psychology, cognitive science, engineering, logic, and mathematics
• Human-centered approaches often require empirical studies of human behavior, and rationalist approaches combine theoretical methods and engineering to design optimized systems

Thinking Humanly: The cognitive modeling approach

• Understanding how humans think involves studying the mind through introspection, psychological experiments, and brain imaging
• The General Problem Solver (GPS) program was designed to solve problems like humans

Cognitive Science

• Cognitive science studies how the mind works, combining psychology, neuroscience, and AI
• It creates precise and testable theories about the brain, and compares computer and human problem-solving

The Connection Between AI and Cognitive Science

• Building a smart computer and understanding human thought are now known as separate goals
• Computer vision uses ideas from neuroscience to improve, and both AI and cognitive science help us learn more about intelligence

Acting humanly: The Turing Test approach

• The Turing Test, proposed in 1950 by Alan Turing, measures a machine's intelligence
• A machine passes if a human interrogator can't distinguish its responses from a human's through written questions and answers, focusing on linguistic and cognitive skills

Example of Turing Test

• Based on the "Imitation game," it involves a computer, a human responder, and an interrogator who must identify which player is the machine
• Conversation is via keyboard and screen, with the computer permitted to force a wrong identification
• The test result does not depend on each correct answer, but only how closely its responses like a human answer.

Requirements to Pass the Turing Test

• Natural Language Processing (NLP): To understand and respond effectively in human language.
• Knowledge Representation: To store and retrieve information for reasoning and communication.
• Automated Reasoning: To use stored knowledge for problem-solving and drawing conclusions.
• Machine Learning: To adapt to new situations and identify patterns.

Total Turing Test

• It expands the original by introducing physical interaction
• Necessary inclusion of Computer Vision to perceive and interpret visual inputs, and Robotics to manipulate physical objects and interact with the environment
• The Turing Test is a conceptual milestone, inspiring advancements in NLP, machine learning, and robotics

Thinking rationally: The “laws of thought” approach

• Aristotle introduced syllogisms, patterns of reasoning that lead to correct conclusions if premises are true
• These rules were foundational to logic and believed to govern human thought

Challenges in Logic-Based AI

• Two key challenges include informal knowledge, which is often uncertain and hard to express in formal logic, and practical limitations, where complex problems can overwhelm a computer's resources

Acting rationally: The rational agent approach

• An agent is anything that takes action, functioning autonomously, sensing environments, persisting over time, adapting to changes, and setting goals
• A rational agent acts to achieve the best possible outcome, or when uncertainty exists, the best expected outcome
• Rationality extends beyond logical inference; quick actions without reasoning can be more effective in certain situations

Advantages of the Rational-Agent Approach

• It is more general, including various mechanisms beyond logical inference, and scientifically grounded, providing a solid framework for designing agents
• Rational agents, unlike human behavior, are built to handle general scenarios effectively

Challenges of Rationality

• While perfect rationality is an ideal starting point, AI often relies on limited rationality due to high computational demands, focusing on making good decisions within time and resource constraints

Foundations of Artificial Intelligence

• It identifies key areas including philosophy, mathematics, economics, neuroscience, psychology, and more
• AI seeks to formalize reasoning, combines insights from diverse fields, implements theories, and aims to replicate human intelligence

Overview of Philosophy's Contribution to AI

• Philosophy has contributed foundational questions like the nature of reasoning, the mind-body relationship, and the origins of knowledge
• Aristotle developed syllogisms, Ramon Lull proposed early mechanical reasoning, and René Descartes introduced dualism
• Empiricists believed knowledge comes from sensory experience and bridged rationalism and empiricism

Mathematics

• Boole developed Boolean logic and Alan Turing defined the concept of computation with the Turing machine
• Probability theory provided tools for reasoning under uncertainty, and NP-Completeness identified intractable problems

Economics

• Adam Smith introduced decision-making and self-interest ideas, and Utility Theory provided frameworks for decision-making under uncertainty
• Game Theory analyzed strategic interactions, and Operations Research developed tools for sequential decision-making

Neuroscience

• Paul Broca localized brain functions, and Neurons were identified as basic processing units
• EEG and fMRI study brain activity, and John Searle advocated the idea that minds arise from brains

Psychology

• Behaviorism focused on observable behaviors, and Cognitive Psychology treated the mind as an information-processing system
• Cognitive Science merges psychology, neuroscience, and computer modeling to simulate human thought processes

Computer Engineering

• Early Computers laid the foundation of modern computing with programmable machines, and Charles Babbage and Ada Lovelace pioneered algorithms
• AI Contributions include ideas like time-sharing, symbolic programming, and interactive interfaces

Control Theory and Cybernetics

• Control Theory deals with self-regulating systems, and Cybernetics explores analogies between biological and mechanical control systems
• Feedback Systems served as an early model for AI systems capable of learning and self-correction

Linguistics

• Noam Chomsky introduced syntactic structures, and Computational Linguistics developed NLP systems
• Knowledge Representation linked language understanding to reasoning

Neurons

• Neurons consist of a cell body and nucleus, dendrites, and a single long fiber called the axon
• The axon stretches out for a long distance, much longer than the scale in this diagram indicates. Typically, an axon is 1 cm long (100 times the diameter of the cell body), but can reach up to 1 meter
• Signals are propagated from neuron to neuron by a complicated electrochemical reaction, they make short-term changes and also enable long-term changes in the connectivity of neurons
• This is thought to form the basis for learning in the brain and most information processing goes on in the cerebral cortex, the outer layer of the brain

Key Functions and Processes in Neurons

• Neurons communicate via electrical impulses and neurotransmitters at synapses and a neuron connects with between 10,000 and 100,000 other neurons, creating a vast neural network
• Changes in synaptic strength and connectivity underlie learning and memory formation Axons can range from 1cm to 1m in Length
• Signals occur the brain's outer layer . Organizational units contain roughly 20,000 neurons.

THE HISTORY OF ARTIFICIAL INTELLIGENCE

• 1943-1955: Theoretical and practical foundations of AI such as neuroscience, logic, and computation was discovered
• 1956: AI was established. It was focused on simulating human intelligence
• 1952-1969: There was optimism is fostered through success but most of the systems lacked scalability
• 1966-1973: Over-optimism shifted to realism as practical and theoretical limitations emerged
• 1969-1979: Domain expertise demonstrated the power of knowledge-based systems over general-purpose approaches

THE STATE OF THE ART

• AI's versatility and impact across domains were driven by advancements in science, engineering, and mathematics
• Autonomous cars navigate challenging environments
• Automated systems handle entire conversations using speech recognition and dialog management
• NASA's remote agent autonomously plan and execute spacecraft operations
• AI tools translate languages, using statistical models trained on massive datasets

AGENTS AND ENVIRONMENTS defined

• Agent: is an entity that interacts with its environment through sensors and actuators
• Types: human, robotic, and software
• Human Agents: possess eyes, ears skin nose and tongue as their sensors and Hands, legs, and vocal tract as actuators
• Robotic Agents: possess cameras, infrared range finders, proximity sensors, or gyroscopes as their sensors and motors wheels, or robotic arms as actuators
• Software Agents: inputs like Keystrokes, files, or network packets as sensors and screen displays, sending files, and network requests as actuators

Percept and Percept Sequence:

• Percept: agent input
• Percept Sequence: the whole history of all the senses experiences from the start

Agent Function and Behavior

• Agent function: agent's mathematical behavior for a perceived sequence of action
• Agent program: agent's real-world implementation into action

Figure 2.2: Vacuum-Cleaner World

• Environment: has 2 squares A and B, which contains dirt
• Agent: a vacuum cleaner, that can sense if the current square is dirty
• Actions: move left, move right, suck, and do nothing

Good Behavior

• A rational agent makes the best possible decisions in any given situation and maximizes all results
• The best action is determined by the results of a behavior i.e Desirable outcomes with the performance measure and good behavior

Key Components of Rationality Depend on 4 Factors:

• 1. Performance measure: What success looks like (keeping floor clean)
• 2. Prior knowledge: the agent already knows about the starting trip
• 3. Available Actions: move right, move left, move suck up dirt
• 4. Percept sequence: use past data to make predictions

Good behavior-Example

• Vacuum cleaner agent has a task to clean dirt efficiently
• The task environment contains dirt, and squares that contains clean and dirt
• The agent senses its current locations, and detects where its moving

Challenges in Designing Performance Measures

• Designing it with effects
• Defining success
• Must measure real world goals
• Exploiting loopholes

Philosophical and Practical Considerations

• Exists in tradeoff with efficiency, and must balance competing needs
• Involves ethical consideration and economic
• The performance measure is different, specific goals and trade-offs. How Rational Agents Learn and Adopt
• Knowing what must be done through omniscience, and rationality, to get the best available data
• Has low and high autonomy, with preprogramed behavior

The Nature of Environments

• PEAS stands for performance measure environment , actuators and sensors
• It answers for important questions about intelligent agents

The Structure of Agents

• An agent program is the mapping of data that transfers data, with sensors and actuators, forming an agent
• Made up of the architecture and programing together
• Agent Program takes in the current data to direct mapping
• Architecure includes legs, sending, seeing

Why Table Driven Agents are Inefficient?

• They used the look up table to map out all possible pre cert sequence
• Also unrealistic for learning and no guidance

AT's Solution-Efficient Programs

• It is more efficient replace to table driven methods with more effective programs

Agent:

• Is fully defined by the current percept, based on condition action rules, is simple and efficient, but limited, and can get stuck if moving aimlessly

Model Based Reflex Agents

• Handles partiality, and tracks how the world's world resolves the agent turning It increases complexity, since car has data to take

The Structure of Agents are

• Goal based to obtain the specific requirements
• It's much more adaptable and has a model of actions to see if possible results are positive

Levels of Representation in Agents

• 1. Atomic representation : treats each data
• 2. Factories representation: list of elements
• 3. Structure representration