Time: 2 Hrs | Max. Marks: 60
Section - A
I. Answer any TEN questions of the following. (2x10=20)
1. List the Al tasks. Give alteast two examples for each task.
Perception: Computer Vision, Speech Recognition
Reasoning: Logical Inference, Theorem Proving
Learning: Supervised Learning, Unsupervised Learning
Planning: Robot Path Planning, Automated Scheduling
2. What are PEAS properties of agent?
PEAS stands for: (Performance measure, Environment, Actuators, Sensors.)
Performance Measure: Criteria to evaluate agent success.
Environment: The surroundings in which the agent operates.
Actuators: Mechanisms through which the agent acts.
Sensors: Devices that perceive the environment.
3. What is environment? List the types of environment.
An environment is the external context in which an agent operates. or An environment in AI refers to everything that an agent interacts with or operates within.
Types of environments:
Fully Observable vs. Partially Observable
Deterministic vs. Stochastic
Episodic vs. Sequential
Static vs. Dynamic
Discrete vs. Continuous
Single-agent vs. Multi-agent
4. State the difference between informed and uninformed search.
| Informed Search | Uninformed Search |
|---|---|
| Uses heuristics (domain knowledge) to guide search. | No heuristics; No domain knowledge used. |
| More efficient (targets goal faster). | Less efficient (explores blindly). |
| Lower memory usage (prioritizes promising paths). | Higher memory usage (e.g., BFS stores entire frontier). |
| Examples: *A Search**, Greedy Best-First. | Examples: BFS, DFS, UCS. |
| Can guarantee optimality (e.g., A* with admissible heuristic). | No optimality guarantee (e.g., DFS may find suboptimal paths). |
5. Mention the different heuristic search techniques.
Greedy Best-First Search
A Search*
Hill Climbing
Simulated Annealing
Beam Search
6. What is adversarial search?
Adversarial search involves AI agents competing against each other (e.g., in games like Chess or Tic-Tac-Toe). It uses techniques like Minimax algorithm and Alpha-Beta Pruning to make optimal decisions.
7. What are the issues in knowledge representation?
Representation Accuracy: How well knowledge is captured.
Inferential Adequacy: Ability to derive new knowledge.
Efficiency: Computational feasibility.
Handling Uncertainty: Managing incomplete or noisy data.
8. List the basic logical operators in propositional and predicate logic.
Propositional Logic: AND (∧), OR (∨), NOT (¬), IMPLIES (→), IFF (↔)
Predicate Logic: Universal Quantifier (∀), Existential Quantifier (∃)
9. What is decision tree?
A decision tree is a supervised learning model that splits data into branches based on feature values to make predictions. It consists of:
Nodes: Decision points (features)
Edges: Possible outcomes
Leaves: Final decisions (class labels)
10. What is reinforcement learning?
Reinforcement Learning (RL) is a machine learning paradigm where an agent learns by interacting with an environment, receiving rewards or penalties for actions. Examples include:
Q-Learning
Deep Q-Networks (DQN)
Policy Gradient Methods
11. What is expert system?
An expert system is a computer program designed to simulate the decision-making ability of a human expert. It uses a knowledge base of facts and rules, along with an inference engine, to solve complex problems within a specific domain.
Examples:
MYCIN (for medical diagnosis)
DENDRAL (for chemical analysis)
12. What is knowledge engineering?
Knowledge engineering is the process of creating and maintaining knowledge-based systems, such as expert systems. It involves:
Identifying domain experts
Extracting knowledge from those experts
Encoding that knowledge into a usable format (like rules or ontologies)
Testing and refining the system
Section - B
Answer any SIX of the following, each not exceeding 200 words. (6×5=30)
13. Explain rationality and rational agent.
Rationality refers to the quality of making decisions that maximize expected performance, based on the knowledge an agent has.
A rational agent is one that acts to achieve the best expected outcome, given what it knows. It doesn’t always make perfect decisions but always chooses the action that is most likely to lead to success based on available information.
A rational agent is an entity that makes optimal decisions to achieve its goals based on:
Perception: Input from sensors/environment
Knowledge: Available information and rules
Actions: Possible choices it can make
Rationality means the agent selects actions that maximize its performance measure given:
Its knowledge of the environment
Its available actions
Its percept sequence
14. Write a note on : Local maximum, plateau and ridge.
Local Maximum: A state that is better than its neighbors but not the best overall (global maximum). The algorithm may get stuck here, thinking it’s at the peak.
Plateau: A flat region where neighboring states have the same value. Since there’s no clear direction to move, search algorithms may wander or halt.
Ridge: A region with a narrow path of increasing value. It’s difficult for the algorithm to find the correct path as it needs to make precise moves.
15. Explain mapping between facts and representation.
In AI, facts about the world must be represented in a way that machines can process and reason about. This process is called knowledge representation.
- A fact is a piece of information that is true about the world. It exists independently of how we choose to describe or represent it. Facts are objective truths.
- Representation is the way in which we encode or express facts within a system (e.g., a computer program, a database, a logical formalism). It’s a structured and symbolic way of capturing information about the world.
The mapping between facts and representation is the process of translating objective truths about the world into a formal, symbolic structure that a system can process and reason with. It’s about giving a “voice” to the facts within a computational environment
16. Explain statistical learning methods.
Statistical learning methods use data to build models that can make predictions or decisions under uncertainty. These methods identify patterns and estimate probabilities based on past data.
Common methods:
Linear Regression – Predicts a continuous output based on input variables
Logistic Regression – Used for binary classification
Naive Bayes Classifier – Based on Bayes’ theorem for classification
Support Vector Machines (SVM) – Finds the best boundary to separate classes
Bayesian Networks – Probabilistic graphical models representing dependencies
Applications:
Spam detection
Medical diagnosis
Risk analysis
17. What are the different methods of knowledge acquisition.
Knowledge acquisition refers to the process of gathering, organizing, and structuring knowledge from various sources for use in AI systems.
Methods include:
Interviews with domain experts – Direct extraction of knowledge
Observation – Studying expert behavior in real tasks
Document analysis – Extracting knowledge from books, manuals, reports
Machine learning – Automatically learning from data (supervised, unsupervised)
Crowdsourcing – Collecting knowledge from multiple contributors
Sensor-based acquisition – Using devices to collect environmental data
Section - C
III. Answer any TWO questions of the following. (10x2=20)
18. Explain the different disciplines in foundation of AI.
Artificial Intelligence (AI) draws knowledge from multiple disciplines to create intelligent systems. The key foundational disciplines include:
1. Mathematics
Logic: Propositional, predicate, and fuzzy logic for reasoning.
Probability & Statistics: Bayesian networks, decision theory, and uncertainty handling.
Optimization: Gradient descent, linear programming for machine learning.
2. Computer Science
Algorithms: Search (BFS, DFS, A*), sorting, and optimization techniques.
Data Structures: Trees, graphs, and hash tables for efficient knowledge representation.
Theory of Computation: Turing machines, computability, and complexity (P vs. NP).
3. Cognitive Science & Psychology
Human Cognition: Studies how humans think, learn, and solve problems.
Behavioral Models: Helps in designing natural language processing (NLP) and human-like AI.
4. Neuroscience
Neural Networks: Inspired by the human brain’s structure (deep learning).
Brain-Computer Interfaces (BCI): Enhances AI with biological intelligence.
5. Philosophy
Ethics of AI: Moral implications of autonomous systems.
Consciousness Debate: Can machines truly “think” or just simulate intelligence?
6. Linguistics
Natural Language Processing (NLP): Syntax, semantics, and discourse analysis.
Machine Translation: Google Translate, ChatGPT, etc.
7. Engineering (Robotics & Control Systems)
Robotics: Combines AI with mechanical systems (e.g., self-driving cars).
Control Theory: PID controllers, reinforcement learning in automation.
19. Explain "Generate and Test" algorithm with an example.
The Generate and Test algorithm is a simple problem-solving technique in Artificial Intelligence and computer science. It works by generating potential solutions and then testing each one to see if it satisfies the required conditions or goals.
How It Works (Step-by-Step):
Generate Phase
System produces a potential solution (either randomly or systematically)
Example: In puzzle-solving, generates possible moves
Test Phase
Checks if generated solution meets requirements
If YES → Solution found (algorithm stops)
If NO → Generates next candidate
Key Characteristics:
Brute-force approach (tries all possibilities)
No heuristics used (unlike informed search)
Can be computationally expensive for large problems
Example: Solving the 8-Queens Problem
Problem: Place 8 queens on a chessboard so none attack each other.
Generate and Test Approach:
Generate a random arrangement of 8 queens
Test the arrangement:
Check if any queens share same row, column, or diagonal
If safe → Solution found
If conflict → Generate new arrangement
Visualization of One Test:
Attempt 1: Q . . . . . . . . . Q . . . . . . . . . Q . . . . . . . . . Q . . Q . . . . . . . . . Q . . . . . . . . . Q . . . . . . . . . Q → Fails (diagonal conflict) Attempt 2: . . . Q . . . . . . . . . . Q . . . Q . . . . . Q . . . . . . . . . . . . . . Q . . . . Q . . . . Q . . . . . . . . . . . Q . . → Valid solution!
| Advantages | Disadvantages |
|---|---|
| Simple to implement | Inefficient for large spaces |
| Guaranteed to find solution if exists | No optimization (blind search) |
| Works for any problem with testable solutions | Slow for complex problems |
Real-World Applications:
Puzzle solving (Sudoku, crosswords)
Software testing (generating test cases)
Drug discovery (molecular structure generation)
20. a) Explain the types of expert systems.
Expert systems are AI programs that emulate human expertise in specific domains. They are classified as:
1. Rule-Based Systems
Use “IF-THEN” rules for decision-making.
Example: MYCIN (medical diagnosis).
2. Fuzzy Logic Systems
Handle uncertainty using degrees of truth (0 to 1).
Example: Washing machines adjusting cycles based on dirt level.
3. Frame-Based Systems
Organize knowledge in hierarchical structures (frames).
Example: Tax calculation software.
4. Neural Network-Based Systems
Learn from data patterns (e.g., deep learning in diagnostics).
Example: AI radiology image analysis.
5. Case-Based Reasoning (CBR) Systems
Solve new problems by recalling past similar cases.
Example: Legal advisory systems.
6. Hybrid Systems
Combine multiple approaches (e.g., rules + neural networks).
Example: IBM Watson (NLP + machine learning).
20. b) State the difference between propositional and first order logic.
| Propositional Logic | First-Order Logic (FOL) |
|---|---|
| Basic elements are propositions (P, Q) | Uses predicates (P(x), Q(x,y)) |
| No variables or relations | Uses variables (x,y) and relations |
| Cannot express quantifiers | Uses ∀ (all) and ∃ (exists) |
| Limited to simple true/false | Handles complex relationships |
| Example: P → Q | Example: ∀x(P(x)→Q(x)) |
| No functions or equality | Supports functions and = operator |
| Used in basic logic circuits | Used in advanced AI/KR systems |
| Less expressive power | More expressive and versatile |