Lecture 1

1. Fundamental Definitions

Intelligence: The ability of systems to perform tasks that typically require human or natural intelligence.
Agent: A computer system situated in an environment that is capable of autonomous action within that environment to achieve delegated objectives.
Rational Agent: An agent that acts to achieve the best outcome, or the best expected outcome when uncertainty is involved. It uses knowledge representation and reasoning to reach good decisions.
Multiagent System (MAS): A system consisting of several interacting agents that typically exchange messages through a computer network infrastructure.

2. The Four Categories of AI

The definitions of AI are organized into four main approaches:

Thinking Humanly: The automation of activities we associate with human thinking, such as learning, problem-solving, and decision-making.
Thinking Rationally: The study of mental faculties using computational models, making it possible to perceive, reason, and act.
Acting Humanly: The art of creating machines that perform functions requiring intelligence when performed by people (e.g., the Turing Test).
Acting Rationally: The study of the design of intelligent agents and intelligent behavior in artifacts.

3. Strong AI vs. Weak AI (Comparison)

Strong AI:
- Definition: Machines that can genuinely reason, solve problems, and are conscious and self-aware.
- Classification Map: Maps to "Thinking rationally" and "Acting rationally".
- Current Status: Currently, no intelligent agent of this type has been created due to a bottleneck in brain science.
Weak AI:
- Definition: Machines that only appear intelligent but do not have real intelligence, self-awareness, or the ability to truly reason and solve problems.
- Classification Map: Maps to "Thinking like human beings" (e.g., Watson, AlphaGo) and "Acting like human beings" (e.g., humanoid robots like Atlas).

4. Key Characteristics of Agents (Properties)

For an agent to be considered intelligent, it generally exhibits these four properties:

Autonomy: The agent operates without direct human intervention.
Reactivity: The agent perceives its environment and responds to changes occurring within it.
Proactiveness: The agent exhibits goal-directed behavior, taking the initiative to achieve its delegated objectives rather than just passively reacting to the environment.
Social Ability: The agent is capable of interacting with other agents (and possibly humans) to satisfy its design objectives or common goals.

5. "Consists Of" & System Architecture

An Agent's Interaction Loop consists of:
- Sensors: Used to receive percepts from the environment.
- Effectors: Used to execute actions upon the environment.
A Multiagent System consists of:
- Several interacting agents, which can be software entities, robots, or humans.
- Key Idea: These components work together to solve problems that are beyond the capabilities of a single agent.

Lecture 2

1. Fundamental Definitions

Intelligent Agent: Anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators.
Percept: The agent's perceptual inputs at any given instant.
Percept Sequence: The complete history of everything the agent has perceived.
Agent Function: A mathematical mapping from any given percept sequence to an action.
Task Environment: Includes all relevant external factors and conditions that impact an agent's behavior and ability to perform.
Rational Agent: For each possible percept sequence, it selects an action expected to maximize its performance measure, given the evidence provided by the percept sequence and built-in knowledge. Simply put: it does the right thing.
Performance Measure: An objective criterion used to evaluate the success of an agent's behavior.

2. "Consists Of" & Core Architectures

The Agent Architecture

An agent is fundamentally composed of two parts:

Agent = Architecture + Program.
- The Program: Runs on the physical architecture to produce the agent function.

The 4 Pillars of Rationality Rationality is evaluated based on four components:

Performance measure: How to know the agent succeeded.
Prior knowledge: What the agent knows about the environment beforehand.
Actions: The capabilities the agent can perform.
Percept sequence: What the agent has perceived to date.

The PEAS Framework (Specifying the Task Environment) Used to define the relevant external factors impacting an agent:

P - Performance: The goal context or objective criteria (e.g., safe, fast, distance traveled, battery life).
E - Environment: The physical or virtual surroundings the agent interacts with (e.g., roads, traffic, pedestrians).
A - Actuators: The mechanisms used for interaction/action (e.g., steering wheel, motors, cleaning mechanism).
S - Sensors: How the agent perceives the world (e.g., cameras, GPS, keyboard).

3. Key Comparisons

Table 1: Types of Physical Agents

Agent Type	Sensors	Actuators
Human Agent	Eyes, ears, and other organs.	Hands, legs, mouth, and other body parts.
Robotic Agent	Cameras and infrared range finders.	Various motors.

Table 2: Rationality Distinctions

Concept	Characteristics
Rationality	Maximizes the expected outcome based on what it has perceived. Involves learning from perceived information to avoid repeating mistakes.
Omniscience	Having total actual wisdom. Rational agents are not omniscient because percepts rarely supply all relevant information.
Perfection	Maximizes the actual outcome. Rationality is about doing the best with what you have, not necessarily achieving a flawless actual outcome.

Table 3: Task Environment Characteristics

Characteristic Focus	Type 1	Type 2 (The Contrast)
Observability	Fully Observable: Sensors detect all aspects required to choose an action (perfect information).	Partially Observable: Parts of the environment are inaccessible; agent must make informed guesses.
Certainty	Deterministic: The next state depends only on the current state and the agent's action.	Stochastic: Non-deterministic; aspects are beyond the agent's control.
Time/Planning	Episodic: Current action choice does not depend on previous actions (sporadic).	Sequential: Current choice affects future actions; requires planning ahead.
Stability	Static: The environment doesn't change while the agent deliberates.	Dynamic: The environment changes during deliberation. (Note: Semi-dynamic* means the environment itself doesn't change, but the performance score drops over time)*.
State Space	Discrete: A limited, distinct, clearly defined number of percepts and actions.	Continuous: Features a range of values.
Population	Single Agent: Operating by itself in an environment.	Multiagent: Many agents working together.

Lecture 3

1. Fundamental Definitions & Concepts

Agent Function (with State): An agent program can implement an agent function by maintaining an internal state, mapping from percept histories to actions: $f : P^{*} \to A$ . When including the history of actions taken, it maps as $f : P^{*}, A^{*} \to S \to A$ , where $S$ is the set of states.
Markovian State Space: A state space where each internal state includes all information relevant to decision-making.
Perfect Recall: A state space where each state includes the information about the percepts and actions that led to it.
Perfect Information: A condition that requires Perfect Recall, Full Observability, and Deterministic Actions.
Utility Function: A function that maps a state onto a real number, describing the associated degree of "happiness", "goodness", or "success".

2. "Consists Of" & Core Architectures

A Model-Based Agent consists of:

An internal representation of the world, often called a "model".
Knowledge of "How the world evolves" (e.g., an overtaking car gets closer from behind).
Knowledge of "What my actions do" (e.g., turning the wheel clockwise takes you right).

A Learning Agent consists of:

Performance element: What was previously considered the whole agent; it takes sensor input and outputs actions.
Critic: Evaluates how the agent is doing against a fixed performance standard.
Learning element: Modifies the performance element for the future based on feedback.
Problem generator: Suggests exploring new actions and tries to solve problems differently instead of just optimizing.

3. Key Comparisons

Agents vs. Objects

Feature	Agents	Objects
Autonomy	Embody a stronger notion of autonomy; they decide for themselves whether to perform an action requested by another agent.	Possess a weaker notion of autonomy compared to agents.
Behavior	Capable of flexible, reactive, proactive, and social behavior.	The standard object model has nothing to say about such flexible types of behavior.
Control	A multiagent system is inherently multi-threaded, meaning each agent is assumed to have at least one thread of control.	Typically lack independent, multi-threaded control in standard models.

Four Basic Agent Types (Pros & Cons)

Agent Type	Core Characteristics	Pros	Cons
Simple Reflex	Direct mapping from perceptions to actions using condition-action rules. Has no memory; action depends only on the current percept.	The simplest agent. Fast reaction times, making it well-suited for dynamic environments.	Limited intelligence; ignores percept history; fails in partially observable environments; prone to infinite loops, cannot learn or adapt to new situations.
Goal-Based	Uses knowledge about a goal to guide actions through search and planning. Needs the current state and a goal state to make decisions.	Goal-oriented behavior. Can solve complex problems requiring planning and can flexibly adapt by replanning.	Computationally expensive. Defining usable goals is challenging, and it struggles with incomplete information. Cannot learn or adapt to new situations.
Utility-Based	Uses a utility function to evaluate goals based on factors like speed and safety. Focuses on degrees of happiness or success.	Makes rational decisions that maximize expected utility. Can handle uncertainty via probabilities and manage complex, time-variant preferences.	Defining an accurate utility function is challenging. Calculating expected utility is computationally expensive. Utility functions can be subjective.
Model-based Reflex	Maintains an internal state (model) to track parts of the world not currently visible.	Uses transition and sensor models to predict how the world evolves and how actions affect it.	More complex than simple reflex; requires constant updating of the internal world state.
Learning	Can be applied to any architecture to improve performance over time.	Includes a learning element, critic (evaluator), and problem generator for exploration.	Exploration can be costly in the short term (e.g., fewer tips for a taxi driver while experimenting).

Alternative Architectural Perspectives

Architecture Category	Characteristics
Reactive	Behavior-based architectures, such as simple reflex agents.
Deliberative (Intentional)	Involves thoughtful and planned action, reasoning before acting, and relying on internal knowledge-based models of the world.
Hybrid	Combines elements of both reactive and deliberative architectures.
Learning	Utilizes systems like Reinforcement Learning and Deep Learning.

Lecture 4

1. Fundamental Definitions

Graph: A mathematical structure used to model pairwise relations between distinctive entities, made up of vertices (nodes) connected by edges (links/arcs).
Tree: A directed, connected graph without cycles (acyclic) where any two vertices are connected by exactly one path, and nodes have a single parent.
State Space Graph: A mathematical representation of a search problem where nodes are abstracted world configurations and arcs represent action results.
Search Tree: A "what if" tree of plans and their outcomes where the root is the start state, children are successors, and nodes represent the plans that achieve specific states.
Solution: A sequence of actions (a plan) that transforms the start state into a goal state.

2. "Consists Of" & Core Frameworks

A Formal Problem consists of 5 components:

Initial state: The starting configuration.
Possible set of actions: What the agent can do.
Transition model: A description of what each action does and the resulting consequence.
Goal test: Determines whether a given state is a goal state (or belongs to a set of possible goal states).
Path cost: The numeric cost associated with the sequence of actions.

Solving a Problem formally consists of 4 phases:

Goal formulation: Defining the objective.
Problem formulation: Defining the states and actions.
Search: Finding a solution sequence.
Execution: Carrying out the planned actions.

A Planning Agent consists of/relies on:

Asking "what if" to consider how the world would be.
Making decisions based on the hypothesized consequences of its actions.
A model of how the world evolves in response to those actions.
Formulating a specific goal test.

3. Key Comparisons

Table 1: Levels of Agent Representation

Representation Level	Description	Characteristics & Usage
Atomic	A state is a "black box" with no internal structure.	Used in search, game-playing, and Hidden Markov Models (HMM). Standard for problem-solving agents.
Factored	States are represented as a vector of attribute values (variables or properties).	Example: State = [Location: Room A, SoC: 80%]. Frequently used by planning agents.
Structured	States include complex relationships between objects.	Actions can affect specific variables or relationships. Frequently used by planning agents.

Table 2: Graph vs. Tree

Feature	Graph	Tree
Parenting	Nodes can have multiple parents.	Nodes have a single parent.
Cycles	Can contain loops or cyclic paths.	Acyclic; strictly without cycles.
Relationship	A graph is not necessarily a tree.	Any tree is a graph.
Conversion	Can be turned into a tree by replacing undirected links with two directed links and avoiding loops.	The resulting structure after removing cycles from a graph search.

Table 3: State Space Graph vs. Search Tree

Concept	Characteristics
State Space Graph	In this graph, each state occurs only once. It is rarely built fully in memory because it is usually too large.
Search Tree	Built on demand. A single node in a search tree represents an entire path (plan) mapped out in the state space graph. Because different paths can lead to the same state, there is lots of repeated structure in the tree.

Lecture 5

1. Fundamental Definitions

Search Space: The set of objects among which we search for the solution.
Goal Condition: Characteristics of the object we want to find in the search space.
Solution: A sequence of actions to move from the start state to the goal state.
Optimal Solution: The solution that has the lowest path cost among all possible solutions.
Completeness: Whether an algorithm is guaranteed to find a solution when one exists and correctly report failure when there is none.
Optimality: Whether a strategy finds the optimal solution (lowest path cost).
Time Complexity: How long it takes to find a solution (measured in seconds, states, or actions).
Space Complexity: How much memory is needed to perform the search.

2. "Consists Of" & Core Frameworks

A Search Problem consists of:

A state space: Configurations of the world.
A successor function: World dynamics (with actions and costs).
A start state
A goal test: Characteristics to find.

A Tree Node consists of (Infrastructure for Search Algorithms):

STATE: The state in the state space to which the node corresponds.
PARENT: The node in the search tree that generated this node.
ACTION: The action that was applied to the parent to generate the node.
PATH-COST: The cost ( $g (n)$ ) of the path from the initial state to the node, as indicated by the parent line.

Tree Complexity Parameters consist of:

b: Maximum branching factor.
d: Depth of the optimal solution.
m: Maximum depth of the state space.

3. Key Comparisons

Table 1: State Space Graph vs. Search Tree (Revisited for this Lecture)

Feature	State Space Graph	Search Tree
Usage	Can be used directly by developing a suitable search technique.	Often easier to build first before searching.
Nodes represent	Abstracted world configurations (states).	Plans for reaching states.
Cost representation	Arcs represent actions and costs.	Plans have costs (the sum of action costs).

Table 2: Search Strategy Types

Strategy Type	Characteristics
Uninformed (Blind)	Uses only the information available in the problem definition.
Informed	Uses domain-specific knowledge beyond the problem definition (heuristics, metaheuristics).