AI Essentials: Concepts & Techniques
Artificial Intelligence builds machines to perform human-like tasks like learning and reasoning, optimizing performance through rational action in complex environments.
Core Principles
- Rational Agent: Selects actions maximizing expected performance based on percept history and knowledge.
- PEAS Framework: Defines agent's task environment by Performance, Environment, Actuators, Sensors for design.
- Uninformed Search: Explores systematically without domain knowledge (BFS for shallow, DFS for deep goals).
- Informed Search: Uses problem-specific heuristic knowledge to guide search towards goals (A* for optimal paths).
- Knowledge Representation: Stores, organizes, and manipulates knowledge for effective reasoning (e.g., Logic, Frames).
- Learning Paradigms: Agents learn via supervised (labeled), unsupervised (patterns), or reinforcement (rewards) methods.
Action Steps
- 1. Define Agent Task (PEAS): Specify Performance, Environment, Actuators, Sensors for clear AI agent design.
- 2. Choose Search Strategy: Use BFS for shortest path in unweighted graphs; DFS for deep goals with limited memory.
- 3. Implement A* Search: Combine actual cost g(n) and heuristic h(n) for optimal pathfinding in complex spaces.
Formulas
- Bayes' Theorem: $$P(H|E) = \frac{P(E|H) \times P(H)}{P(E)}$$
- Q-learning Update: $$Q(s, a) \leftarrow Q(s, a) + \alpha [r + \gamma \max_{a'} Q(s', a') - Q(s, a)]$$
- Entropy (ID3): $$E(S) = -\sum p_i \log_2 p_i$$
- Simulated Annealing Acceptance (worse move): $$P = e^{\frac{\Delta E}{T}}$$
Key Terms
- Intelligent Agent: An entity that perceives its environment through sensors and acts via actuators to maximize its performance measure.
- Heuristic Function h(n): An estimate of the cost from a given node 'n' to the goal state, guiding informed search algorithms.
- Knowledge Base (KB): A repository of domain-specific facts and IF-THEN rules used by an expert system's inference engine for reasoning.
Pro Tips
- For A* search, ensure heuristic h(n) is admissible (never overestimates true cost) for guaranteed optimal solutions.
- When designing Fuzzy Logic, prioritize clear, human-understandable rule design; subjective rule bases are hard to manage.
- In exam answers, start with a definition, add 5-7 clear points, include a diagram/example, and conclude with advantages/applications.
Pitfalls to Avoid
- Ignoring AI winters (1974-1980, 1987-1993) causes unrealistic expectations, wasting resources on unmet promises.
- Using biased training data leads to discriminatory AI decisions, causing ethical and fairness issues in applications like hiring.
- Relying solely on Hill Climbing risks getting stuck in local maxima, preventing discovery of the global optimum solution.
Myth vs Reality
- Current AI systems possess human-level intelligence and self-awareness.: All current AI is Weak AI (Narrow AI), designed for specific tasks without true understanding or consciousness; Strong AI is still theoretical.
Real World Examples
- Medical Diagnosis: MYCIN diagnoses bacterial infections; INTERNIST/CADUCEUS diagnose internal diseases, providing decision support.
- Self-Driving Vehicles: AI agents use perception (cameras, sensors) and decision-making (agent program) to navigate dynamic, partially observable environments.
Statistics
- AI Term Coined: 1956 by John McCarthy
- Deep Blue beats Kasparov: 1997
- AI Winters: 1974-1980 (1st), 1987-1993 (2nd)
Timeline
- 1950: Alan Turing proposes the Turing Test.
- 1956: Dartmouth Conference: 'Birth of AI'; John McCarthy coins the term.
- 1974-1980: First AI Winter due to limited computing power and unmet promises.
- 1980-1987: Rise of expert systems (e.g., MYCIN, XCON).
- 1987-1993: Second AI Winter.
- 1997: Deep Blue beats chess grandmaster Garry Kasparov.
- 2011 onwards: Machine learning, deep learning, Watson, AlphaGo, self-driving cars emerge.
Quiz
- Which search algorithm guarantees the shortest path in an unweighted graph?: B. Breadth First Search
- What does 'P' stand for in the PEAS framework for AI agents?: C. Performance measure
- What is the primary characteristic of Weak AI (Narrow AI)?: C. Designed for a specific task within a limited range
Related cheat sheets
More like this