Artificial Intelligence is a broad but exam-friendly topic because most questions ask definitions, search algorithms, heuristic reasoning, game playing, learning types, NLP pipeline, planning and robotics components. A strong answer links each subtopic to intelligent-agent behavior.
Engineering Definitions
Artificial Intelligence
Standard definition: The field of building systems that perform tasks requiring intelligence such as reasoning, learning, perception and decision-making.
Exam meaning: Reasoning, learning, perception र decision-making गर्ने intelligent systems बनाउने field।
Intelligent agent
Standard definition: An entity that perceives its environment through sensors and acts through actuators to achieve goals.
Exam meaning: Sensor बाट environment बुझेर actuator बाट goal-oriented action गर्ने system।
Heuristic
Standard definition: A problem-specific estimate or rule that guides search toward promising states.
Exam meaning: Search लाई goal नजिक लैजान प्रयोग हुने estimate/rule।
Machine learning
Standard definition: A method where systems improve performance from data rather than explicit programming only.
Exam meaning: Data बाट pattern सिकेर performance improve गर्ने method।
Concept Teaching
AI can be understood as search plus knowledge plus learning plus action. Search explores possible states, reasoning uses knowledge, learning improves from data, planning chooses action sequences, NLP/vision perceive language/images, and robotics connects perception to physical action.
Search Algorithms
Search explores a state space from initial state to goal.
| Algorithm | Strategy | Exam point |
|---|---|---|
| BFS | Shallowest first using queue | Complete; optimal for equal cost |
| DFS | Deep first using stack/recursion | Low memory, may get stuck |
| Uniform cost | Lowest path cost first | Optimal with positive costs |
| Greedy best-first | Uses heuristic h(n) | Fast but not always optimal |
| A* | Uses f(n)=g(n)+h(n) | Optimal if heuristic is admissible |
Game Playing and Reasoning
Games model adversarial search.
- Minimax assumes optimal opponent.
- Alpha-beta pruning removes branches that cannot affect decision.
- Knowledge representation uses logic, semantic networks, frames or rules.
- Inference derives new facts from known facts.
- Uncertainty can be handled using probability and Bayesian reasoning.
Machine Learning Types
Learning type depends on feedback.
| Type | Feedback | Example |
|---|---|---|
| Supervised | Labeled data | Classification/regression |
| Unsupervised | No labels | Clustering/dimensionality reduction |
| Reinforcement | Reward signal | Agent learns policy |
| Semi-supervised | Small labels + many unlabeled | Practical data-scarce tasks |
NLP, Vision and Robotics
These areas connect AI to real-world perception/action.
- NLP pipeline includes tokenization, morphology, parsing, semantics and pragmatics.
- Computer vision extracts features, detects objects and interprets scenes.
- Planning generates action sequence from initial state to goal.
- Robotics combines perception, localization, planning, control and actuation.
- Feedback loop is essential for autonomous systems.
Engineering Mechanism
- Agent senses environment.
- State representation captures relevant facts.
- Search/planning evaluates possible actions.
- Reasoning or model predicts consequences.
- Learning improves model/policy from data.
- Actuator performs selected action.
- Feedback updates future decisions.
Diagrams / Models To Draw
- Draw intelligent agent loop: sensor, agent, actuator, environment.
- Draw search tree with BFS/DFS frontier.
- Draw minimax tree with alpha-beta cut.
- Draw ML pipeline: data, training, model, prediction.
- Draw robotics perception-planning-control loop.
Formulas, Algorithms and Rules
- A* evaluation: f(n)=g(n)+h(n).
- Admissible heuristic never overestimates true cost.
- Minimax chooses max of minimum opponent outcomes.
- Precision = TP/(TP+FP); Recall = TP/(TP+FN).
| Area | Core task | Exam distinction |
|---|---|---|
| Search | Find path/solution | Blind vs informed |
| Learning | Infer model from data | Supervised/unsupervised/RL |
| NLP | Process language | Syntax vs semantics |
| Vision | Interpret images | Features/classification/detection |
| Planning | Action sequence | State-space reasoning |
| Robotics | Embodied AI | Sensors and actuators |
Exam Point
- A* optimality depends on admissible heuristic.
- BFS uses queue; DFS uses stack.
- Minimax is for adversarial games.
- Differentiate classification and clustering.
- NLP and computer vision are perception tasks.
Worked Example
For route finding, g(n) is distance already traveled and h(n) can be straight-line distance to goal. A* selects node with lowest g+h. If h never overestimates real remaining distance, A* can find optimal path.
Subjective Answer Pattern
- Define AI and intelligent agent.
- Explain search algorithms.
- Discuss heuristic/A* and game playing.
- Explain learning types.
- Add NLP, vision, planning and robotics examples.
- Conclude with applications and limitations.
Common Engineering Mistakes
- Saying every heuristic is optimal.
- Confusing BFS and DFS data structures.
- Calling clustering supervised.
- Ignoring sensors/actuators in robotics.
- Forgetting opponent model in minimax.
MCQ Revision
- A* uses which function?
- Which search uses queue?
- Which ML type uses labels?
- Minimax is used in what?
- What does admissible heuristic mean?
- NLP processes what kind of data?
Final Summary
- AI builds agents that perceive, reason, learn and act.
- Search and heuristics solve state-space problems.
- Machine learning extracts patterns from data.
- NLP and vision handle perception.
- Planning and robotics convert intelligence into action.