Comprehensive AI Notes

Definition: It is the simulation of human intelligence in machines, especially computer systems.
1. AGENT
- An entity that perceives its environment through sensors and acts upon that environment through actuators.
- Example: Roomba.
2. RATIONAL AGENT
- Chooses actions that maximize goal achievement.
- Example: Stock trading bot.
3. PERCEPTION
- Collecting data from the environment.
- Example: Camera and microphone in bots.
4. ACTUATION
- Performing actions.
- Example: Robotic arms/actuators.
5. HEURISTIC
- Rule of thumb to solve problems; often faster but not guaranteed to be optimal.
6. KNOWLEDGE REPRESENTATION
- Encoding information so that AI can reason over it.
- Includes methods to store, organize, and manipulate knowledge for reasoning.
7. INFERENCE
- Deriving new information or reacting correctly from known facts.
8. LEARNING
- Acquiring new patterns and adapting new behavior.
- Example: Face recognition systems.
9. PLANNING
- Selecting actions to achieve a goal.
- Example: Route optimization.
1. PROBLEM SOLVING
- Finding solutions to problems defined within constraints.
- Example: Puzzle solving.

The Turing Test evaluates whether a machine can imitate human intelligence well enough to be indistinguishable from a human in conversation.
Three participants in the classic setup: $3$ participants:
- i. Human interrogator
- ii. Human respondent
- iii. AI respondent
The interrogator does not have pre-information about who is who (no pre-knowledge).

Alan Turing
- Proposed foundational ideas in AI and computation.
John McCarthy
- Co-originated the term AI; helped establish the field; Dartmouth Conference attribution.
Turing Machine and Turing Test as early foundations of AI thought.
Marvin Minsky
- Pioneered AI architectures and cognitive modeling.
Geoffrey Hinton
- Father of Deep Learning; contributed to neural networks and backpropagation insights.
Yann LeCun
- Co-inventor of Convolutional Neural Networks (CNNs); deep learning pioneer.
Andrew Ng
- Online AI education innovator; co-founder of Google Brain and Coursera; promoted applied ML.
Summary: AI evolution spans from symbolic reasoning to statistical learning and deep learning revolutions.

1950s: Birth of AI and Turing influence.
1960s – 1970s: Symbolic reasoning and early AI approaches; first AI winter follows periods of reduced funding and interest.
1980s: Expert systems rise (rule-based systems capturing domain knowledge).
1990s: Statistical AI and game AI gain prominence; shift toward data-driven approaches.
2000s: Machine Learning (ML) and Big Data become central to AI advances.
2010s: Deep Learning revolution accelerates capabilities (large-scale neural nets, GPU acceleration).
2020s: Generative AI emerges (e.g., large language models, advances in creative and autonomous systems).
Quick reference timeline terms (years in broad strokes):
- $1950s$ , $1960s$ , $1970s$ , $1980s$ , $1990s$ , $2000s$ , $2010s$ , $2020s$ .

Searching
- Finding the best way to reach a goal in a problem space (e.g., solving a maze).
Reasoning
- Making logical decisions (e.g., robot playing chess).
Uncertainty and Knowledge Representation
- Handling incomplete or noisy information and storing usable knowledge.
Learning
- Teaching machines to learn from data and improve with experience (e.g., recommendations).
Planning
- Devising a sequence of actions to achieve goals in the presence of uncertainty.
NLP (Natural Language Processing)
- Teaching AI to understand and generate human language (e.g., Siri, GPT).
Perception
- Understanding the world through sensors, cameras, etc.
Representation and Knowledge Representation
- Storing knowledge in a form usable by machines for reasoning and inference.

I. Based on Capabilities
- Narrow AI: AI systems designed for a specific task; cannot perform tasks outside their programmed scope (e.g., facial recognition).
- General AI: Can learn, reason, and apply knowledge across multiple domains at human-like levels.
- Super AI: Hypothetical AI that surpasses human intelligence in all aspects.
II. Based on Functionalities
- i. Reactive Machines: AI that reacts to current input; no memory of past information.
- ii. Limited Memory: AI that uses past data to improve decisions (e.g., some modern systems like chat models rely on recent data).
- iii. Theory of Mind: AI that understands human emotions and social behavior and can replicate it; under construction.
- iv. Self-Aware AI: AI with self-consciousness, self-awareness, and emotions.

In AI, Search refers to the process of navigating a state space to reach a goal state from a given initial state by applying a sequence of actions.
AI operates in environments where the solution isn't known in advance.
Search helps in:
- Decision making
- Planning
- Path finding
- Problem solving
Examples:
- Robots navigating a room
- Game bots computing next moves
- Chatbots planning their next response

Key components:
1) Initial State: Starting condition of the problem.
2) State Space: All possible states the agent can be in.
3) Successor Function: Generates new states from a given state.
4) Search Strategy: Decides which node to expand next.
5) Goal Test: Checks if the goal has been achieved.
6) Solution Path: Sequence of actions to reach the goal, including visited states, actions applied, total path cost, and time taken.
Note: State spaces can be explored with different strategies (e.g., breadth-first, depth-first, heuristic-guided).

State space search is a mathematical model of a problem. It represents:
- All possible configurations (states) that a system/problem can be in.
- How the system can move between these states using actions and a transition function.
Formal components (typical notation):
- $S = {s<em>0, s</em>1, \dots, s_n}$ // Set of states
- $A = {a<em>1, a</em>2, \dots, a_m}$ // Set of actions
- $T(s, a) = s'$ // Transition function: applying action a in state s leads to state s'
- $s_0 \in S$ // Initial state
- $G \subseteq S$ // Goal state(s)
Example体系 (common illustration):
- Actions: $A = {UP, DOWN, LEFT, RIGHT}$
- States: configurations of a puzzle (e.g.,
  the 8-puzzle)
- Transition function moves the blank tile and updates the configuration.
Diagrammatic intuition (not required to reproduce here but typical): initial state → possible successors → further successors → goal state(s).