Foundations of Artificial Intelligence

Foundations of Artificial Intelligence

• Artificial Intelligence (AI) is the field that studies the synthesis and computational agents that act intelligently, focusing on creating computers and programs that can act intelligently.
• AI is defined as "The theory and development of computer systems able to perform tasks normally requiring human intelligence".
• Examples of human intelligence include: logical reasoning, problem-solving, creativity, and planning.

Narrow vs. General AI

• Narrow AI requires reconfiguration or a new algorithm to solve a different task. Examples include chess, speech recognition, and facial recognition.
• General AI applies intelligent systems to any problem and can understand, learn, and apply knowledge across a wide range of tasks.

Main Research Areas of AI

• Reasoning
• Learning
• Problem Solving
• Perception

Applications of AI

• Robotics (industrial, autonomous, domestic)
• Industrial Automation (intelligent control, safety, security)
• Health (drug design, operating theatre robotics)
• Games (NPC, virtual, augmented reality)
• Other areas (education, agriculture, personal assistance, etc.)

Intelligent Agents

• An agent is anything that perceives its environment through sensors and acts upon that environment through actuators.
• Intelligent agents reason and act autonomously to achieve goals.
• The percept sequence is the complete history of all data the agent received from its sensors that is used in action evaluation.
• An agent is judged by its actions.
• A rational agent acts to achieve the best possible outcome based on its knowledge and goals to maximize performance based on specific criteria.
• Agent rationality includes prior knowledge, actions, percept sequence, and success criteria.
• Observability is when not all necessary information is available for an agent to decide; this is called partially observable.

Stochasticity in AI

• Stochasticity refers to the unpredictability in a system or process.
• An action is deterministic if applying it always leads to the same outcome, like in a chess move.
• An action is stochastic if applying it leads to a different outcome, like rolling a die.
• Discrete environments have finite action choices and states, such as in chess.
• Continuous environments have infinite action choices and states, such as in a game of tennis.
• Adversarial environments have agents competing against each other, like in chess.
• Benign environments have agents trying to accomplish a goal without competition with other agents, like predicting weather.

Types of Agents

• Reflex agent: uses the current percept, assuming the environment is fully observable.
• Model-based reflex agent: keeps track of a part of the world it cannot observe, maintaining an internal state that depends on percept, a model of how the environment evolves, and the effect of applied actions.
• Goal-based agent: plans into the future and selects actions to reach its goals, like GPS navigation.
• Utility-based agent: plans into the future and selects actions to maximise some utility (solution quality), like self-driving cars.
• Learning agent: improves performance over time through experiences and feedback, like self-driving cars.

Search Techniques in AI

• Search is a powerful technique for solving AI problems.
• Problems can be formulated as directed graphs, where a system needs to find the right action sequence to achieve a goal while satisfying constraints.
• Blind Search (uninformed search) explores the solution space without any prior knowledge about the problem or environment, such as Breadth-First Search and Depth-First Search.
• Informed Search uses heuristic functions (e.g., A* Search) to prioritize states, leading to more efficient problem-solving.

Modeling Challenges

• Algorithmic complexity is the measure of computational resources an algorithm needs in relation to its input size, typically represented using Big O notation.
• Time and space complexity are important factors to consider when designing efficient algorithms.

Knowledge and Reasoning

• Knowledge in AI is the data, information, and concepts an AI system uses for understanding and solving problems.
• Reasoning is the process of drawing conclusions or making decisions based on the available knowledge, in knowledge-based agents.
• Knowledge-based agents use a knowledge base to make decisions, reason, and solve problems. They maintain an updated knowledge base and use an inference engine to deduce and update knowledge, choosing the best action.
• A knowledge base is a collection of statements typically expressed in a knowledge representation language.
• Axioms are statements that weren't derived from other statements.
• TELL operations add new statements to the knowledge base.
• ASK operations query the knowledge base.

Logical Inference

• Logical inference is the process of reaching conclusions based on known evidence and reasoning.
• Some desirable properties of inference systems are soundness and completeness.
• Soundness ensures that only entailed sentences are derived, avoiding false conclusions.
• Completeness ensures that all entailed sentences can be derived, guaranteeing that nothing is missed.

Planning in AI

• Planning is devising a strategy to achieve a desired goal by choosing actions to maximize the probability of a successful outcome, or a successful execution.
• Planning is a key aspect of high intelligence.
• Planning and scheduling are related but distinct tasks. Planning identifies the tasks, while scheduling determines the right time for each action.

AI Planning Techniques

• Domain-specific planning techniques are tailored to specific domains.
• Domain-independent planning techniques apply across different domains.
• Satisficing planning aims for a good-enough solution quickly.
• Optimal planning aims to find the best possible solution in terms of cost, required steps, or maximum efficiency.

AI Planning and Scheduling

• Planning involves determining the necessary actions to achieve a goal, while scheduling determines the optimal timing and order of these actions.

AI Planning with Uncertainty

• Contingent planning considers alternative outcomes and conditions, developing plans to manage uncertainty.
• Conformant planning is used when the exact details of the environment are unknown. It focuses on planning irrespective of the specifics of a certain state.

Time, Resources, and Exogenous Events

• Real-world planning often involves sophisticated requirements like temporal constraints, numeric resources, and relationships between numeric properties and time, processes and events.

Probability in AI

• Probability theory is a fundamental tool for handling uncertainty in AI.
• Partial observability is when we handle facts that aren't immediately accessible in the environment or when some information is missing or unknown.
• Stochasticity captures uncertainty about outcomes when actions are taken.
• Bayes Networks (graphical models) are tools for probabilistic reasoning, widely used in prediction, diagnosis, and decision-making tasks.

Machine Learning

• Machine learning (ML) is about discovering models from data (patterns).
• ML commonly uses statistics to predict and classify.
• Supervised learning uses pre-classified data to train a model and make predictions about new data, such as images or medical diagnostics.
• Unsupervised learning discovers structures or hidden relationships within data without predefined classifications, such as customer purchase patterns.
• Reinforcement learning learns by interacting with the environment and observing its effects, leading to maximum reward over time.
• Applications include customer purchase pattern identification, robotics, finance (stock market trading) etc.

Computer Vision in AI

• Computer vision is extracting, analyzing, and interpreting images or videos, converting them into numeric arrays for machine learning to process.
• Applications include image preprocessing, image database querying, etc.
• Thresholding is a method of making decisions or classifying data by comparing it to a threshold value.

Law and Ethics in AI

• Key concepts in AI ethics include fairness, transparency, privacy, accountability, and responsibility.
• Concerns about societal biases and potential discrimination, data privacy, and maintaining responsibility are key considerations in AI.

Natural Language Processing (NLP)

• NLP is a branch of AI focused on enabling computer understanding, interpretation, generation, and response to human language.
• The symbolic approach is based on predefined grammar and rules.
• The statistical approach uses large text datasets and machine learning algorithms to identify patterns.
• Stop words are commonly used words that are typically removed from text during processing to reduce noise.

Problems Solving Using Search in AI

• State Space Search: focuses on identifying possible states and transitions between them to find the optimal path;
• Solution Space Search: focuses on discovering candidate solutions, their possible variations to reach the optimum solution.

Genetic Algorithms in AI

• Genetic algorithms (GAs) used for search, optimization, and machine learning, evolving a population of solutions over multiple iterations, utilizing concepts like fitness function values, probabilities, crossover, and mutation.

Alternative Systems for Knowledge Representation

• Semantic networks represent knowledge graphically using nodes representing concepts and edges defining relationships between them.
• Description logics use a formal language for representing and combining categories of knowledge.

Multi-Agent Systems in AI

• Discuss scenarios where multiple intelligent agents operate in an environment, whether benign, cooperative, or adversarial.

Zero-Sum Games

• Zero-Sum games have participants whose gains are equivalent to losses for others. (e.g., chess)

Stochastic Games

• Discuss the concept of stochastic games, where actions could potentially lead to several states with associated probabilities.
• Incorporates elements of traditional game trees with probabilities to handle uncertainty.

Game Theory

• Game theory studies strategic interactions between players, helping understand cooperative or competitive decision-making.
• Pareto-optimal outcome: Any potential outcome where at least one agent cannot improve their result without making another agent's worse.
• Dominant strategy: A strategy that always produces a better outcome than other strategies, regardless of others' choices.
• Nash equilibrium: A situation where no player can improve their outcome by changing their strategy unilaterally, assuming other players maintain their strategies.

Explore the basics of Artificial Intelligence (AI), including its definition, differences between narrow and general AI, and key research areas. This quiz delves into various applications of AI technology, from robotics to industrial automation. Test your understanding of how AI integrates logic, learning, and problem-solving into intelligent systems.