Home Module 1 - Introduction to AI Module 2 - Problem-Solving & Search Module 3 - Uncertainty in AI Module 4 - Games & CSP Module 5 - AI in Practice & Ethics Important Questions MCQ Practice Glossary
Exam Preparation

Important Questions Bank

300+ exam-style questions organized by module and difficulty - short answer (2–3 marks), long answer (8–10 marks), and interview-style conceptual questions. Patterned after university semester exams, competitive exams, and technical interviews.

Module 1
Module 2
Module 3
Module 4
Module 5

Short Answer Questions (2–3 Marks)

  1. 2MDefine Artificial Intelligence according to the "acting rationally" view.
  2. 2MWhat is the Turing Test, and which definition of AI does it correspond to?
  3. 3MList the four classical definitions of AI along the thought/behavior and human/rational axes.
  4. 2MWhat caused the first AI winter (1974–1980)?
  5. 2MDifferentiate between Narrow AI and General AI with one example each.
  6. 3MDefine agent, percept, and percept sequence.
  7. 2MWhat is the difference between agent function and agent program?
  8. 3MList and briefly describe the five types of agent architectures.
  9. 2MDefine PEAS and give the PEAS description for a vacuum-cleaning robot.
  10. 3MList the six properties used to classify task environments.
  11. 2MDefine rationality. Is rationality the same as omniscience?
  12. 2MWhat are the four factors that determine an agent's rationality?
  13. 2MList the four steps of the problem-solving process.
  14. 3MDefine the five components of a formally specified search problem.
  15. 2MDifferentiate between a state space and a search tree.
  16. 2MWhat data does a "node" store in a search algorithm, besides the state itself?
  17. 2MDefine frontier and explored set.
  18. 2MWhat data structure does BFS use for its frontier? What about DFS?
  19. 3MState the time and space complexity of BFS in terms of branching factor b and depth d.
  20. 2MIs BFS always optimal? Justify your answer.
  21. 3MState the time and space complexity of DFS in terms of b and maximum depth m.
  22. 2MWhy is DFS not complete in infinite state spaces?
  23. 2MWhat data structure does Uniform Cost Search use for its frontier?
  24. 2MWhen does UCS reduce to BFS?
  25. 3MWhy must UCS perform its goal test at node expansion rather than node generation?

Long Answer Questions (8–10 Marks)

  1. 10MExplain the four classical definitions of Artificial Intelligence with examples for each, and justify why the "acting rationally" approach is preferred in modern AI engineering.
  2. 10MTrace the historical evolution of AI from 1950 to the present, explaining the causes and consequences of both AI winters.
  3. 8MDefine an intelligent agent. Explain, with diagrams, the five types of agent architectures (simple reflex, model-based reflex, goal-based, utility-based, learning), giving an example application for each.
  4. 8MExplain the PEAS framework and the six environment properties. Illustrate each property with a real-world example distinct from those used in the course material.
  5. 10MDefine a search problem formally. Using the Romania travel problem (or an equivalent example), identify and explain all five components of the formal definition.
  6. 10MExplain the working of Breadth-First Search with a step-by-step trace on a sample graph. Derive its time and space complexity and discuss when it is and isn't optimal.
  7. 10MExplain the working of Depth-First Search with a step-by-step trace on a sample graph. Compare its completeness, optimality, time, and space complexity against BFS.
  8. 10MExplain Uniform Cost Search with a worked numerical example showing frontier evolution. Prove why UCS is always optimal regardless of step-cost variation.
  9. 8MCompare BFS, DFS, and UCS in a single table across completeness, optimality, time complexity, space complexity, and the data structure used for the frontier.
  10. 8MExplain Iterative Deepening DFS. How does it combine the advantages of BFS and DFS while avoiding their respective drawbacks?

Short Answer Questions (2–3 Marks)

  1. 2MDefine heuristic function h(n).
  2. 2MWhat evaluation function does Greedy Best-First Search use?
  3. 2MWhy is Greedy Best-First Search not optimal?
  4. 2MWhat evaluation function does A* Search use?
  5. 3MDefine admissible heuristic and consistent heuristic.
  6. 2MState the relationship between admissibility and consistency.
  7. 2MWhat is meant by "relaxing" a problem to design a heuristic?
  8. 2MDefine heuristic dominance. Why is a dominant heuristic preferred?
  9. 2MWhat is the space complexity of Hill Climbing, and why?
  10. 3MList the three classic problems that trap Hill Climbing.
  11. 2MWhat is Random-Restart Hill Climbing?
  12. 3MState the acceptance-probability formula used in Simulated Annealing.
  13. 2MWhat happens to Simulated Annealing's behavior as temperature T → 0?
  14. 3MDefine chromosome, gene, and fitness function in the context of Genetic Algorithms.
  15. 2MList the main steps of one Genetic Algorithm generation cycle.
  16. 3MWrite the velocity update formula for Particle Swarm Optimization and explain each term.
  17. 2MDifferentiate pbest and gbest in PSO.
  18. 2MWhat is a conditional (contingency) plan, and when is it needed?
  19. 2MDifferentiate OR nodes and AND nodes in an AND-OR search tree.
  20. 2MDefine belief state.
  21. 3MWhat are the two steps involved in updating a belief state after an action and a percept?
  22. 2MWhat is a conformant (sensorless) plan?
  23. 2MDifferentiate offline search from online search.
  24. 3MWhat is LRTA*, and what table does it maintain?
  25. 2MDefine the exploration-exploitation trade-off.

Long Answer Questions (8–10 Marks)

  1. 10MExplain Greedy Best-First Search and A* Search with worked examples on the Romania map problem. Show why Greedy fails to find the optimal path while A* succeeds.
  2. 10MDerive and prove that A* search is optimal when used with an admissible heuristic in Tree-Search mode.
  3. 8MExplain how heuristic functions are systematically designed using problem relaxation. Illustrate with the 8-puzzle's Misplaced Tiles and Manhattan Distance heuristics, and explain heuristic dominance.
  4. 10MExplain Hill Climbing and Simulated Annealing in detail. Compare how each handles the local-optimum problem, with a diagram of a landscape showing local vs global maxima.
  5. 10MExplain the complete working cycle of a Genetic Algorithm (selection, crossover, mutation) with a worked numerical/bitstring example.
  6. 8MExplain Particle Swarm Optimization's velocity/position update equations and compare PSO with Genetic Algorithms in terms of inspiration, representation, and suitability.
  7. 10MExplain search with non-deterministic actions using AND-OR search trees. Illustrate with the erratic vacuum-world example and the AND-OR-SEARCH pseudocode.
  8. 8MExplain search in partially observable environments using belief states. Differentiate sensorless (conformant) search from partially observable search with examples.
  9. 10MExplain online search agents and the LRTA* algorithm. Discuss the exploration-exploitation trade-off and why offline algorithms like A* cannot be used directly in unknown environments.
  10. 8MConstruct a comparison table of Greedy Best-First Search, A*, Hill Climbing, and Simulated Annealing across completeness, optimality, and typical use cases.

Short Answer Questions (2–3 Marks)

  1. 3MList four sources of uncertainty in AI systems.
  2. 2MWhat is the qualification problem in logic-based reasoning?
  3. 2MDefine a Bayesian Network.
  4. 2MState the chain rule for Bayesian networks.
  5. 3MDifferentiate diagnostic, causal, and intercausal reasoning in Bayesian networks.
  6. 2MWhat is "explaining away"? Give an example.
  7. 2MDifferentiate a fork, chain, and collider structure in a Bayesian network.
  8. 2MWhat is Maximum Likelihood Estimation used for in Bayesian network parameter learning?
  9. 2MWhy is Laplace smoothing used when learning CPTs?
  10. 3MDifferentiate exact inference and approximate inference in Bayesian networks.
  11. 2MWhat problem does Likelihood Weighting solve compared to Rejection Sampling?
  12. 2MWrite the formula: Decision Theory = ? + ?
  13. 2MDefine the Maximum Expected Utility (MEU) principle.
  14. 3MList three axioms of von Neumann–Morgenstern utility theory.
  15. 2MWhat does a concave utility curve indicate about risk attitude?
  16. 2MDefine a Markov Decision Process using its 5-tuple.
  17. 2MState the Markov property in one sentence.
  18. 2MDifferentiate a policy from a plan.
  19. 2MWhat role does the discount factor γ play in an MDP?
  20. 3MWrite the Bellman optimality equation used in Value Iteration.
  21. 2MWhat are the two main steps of Policy Iteration?
  22. 2MWhy does Policy Iteration converge in a finite number of steps?
  23. 2MState the per-iteration time complexity of Value Iteration in terms of |S| and |A|.

Long Answer Questions (8–10 Marks)

  1. 8MDiscuss the major sources of uncertainty in AI and explain why probability theory, rather than pure logic, is used to handle it.
  2. 10MExplain Bayesian Networks: their representation (DAG + CPTs), the chain-rule factorization, and the three canonical connection patterns (chain, fork, collider), with the Burglary-Alarm example.
  3. 10MExplain parameter learning and exact/approximate inference methods in Bayesian networks, comparing Enumeration, Variable Elimination, Rejection Sampling, and Likelihood Weighting.
  4. 10MExplain Decision Theory and the Maximum Expected Utility principle with a worked numerical example. State and explain the von Neumann–Morgenstern axioms.
  5. 8MDefine a Markov Decision Process formally. Explain the Markov property and the role of the discount factor with a grid-world example.
  6. 10MDerive the Value Iteration algorithm from the Bellman optimality equation, and trace its working on a simple grid-world MDP.
  7. 10MExplain the Policy Iteration algorithm (Policy Evaluation + Policy Improvement). Compare it with Value Iteration in terms of convergence guarantees and computational cost per iteration.
  8. 8MCompare and contrast Value Iteration and Policy Iteration for solving MDPs, and discuss scenarios where each would be preferred.

Short Answer Questions (2–3 Marks)

  1. 2MDefine a Nash Equilibrium.
  2. 3MClassify games along the dimensions: player count, sum type, information, determinism.
  3. 2MWhat is a zero-sum game?
  4. 2MWrite the Minimax recursive formula.
  5. 2MState the time and space complexity of Minimax in terms of b and m.
  6. 2MDefine alpha and beta in Alpha-Beta Pruning.
  7. 2MWhat is the pruning condition for a MIN node in Alpha-Beta Pruning?
  8. 2MWhat is the best-case time complexity of Alpha-Beta Pruning, and under what condition is it achieved?
  9. 3MList the four phases of one Monte Carlo Tree Search iteration.
  10. 2MWrite the UCB1 formula used in MCTS node selection.
  11. 2MWhy doesn't standard Alpha-Beta Pruning directly apply to chance nodes?
  12. 2MWhat is a CHANCE node, and which algorithm uses it?
  13. 2MDefine an information set in an imperfect-information game.
  14. 2MDefine a Constraint Satisfaction Problem using its 3-tuple.
  15. 3MDifferentiate unary, binary, and higher-order (n-ary) constraints with an example each.
  16. 2MWhat is a constraint graph?
  17. 2MDefine the Minimum Remaining Values (MRV) heuristic.
  18. 2MDefine the Least Constraining Value (LCV) heuristic.
  19. 2MDifferentiate Forward Checking from Arc Consistency (AC-3).
  20. 2MState the time and space complexity of Backtracking Search.
  21. 2MDescribe the Min-Conflicts algorithm in 2-3 sentences.
  22. 2MIs Min-Conflicts complete? Justify briefly.

Long Answer Questions (8–10 Marks)

  1. 8MExplain game theory fundamentals and Nash Equilibrium using the Prisoner's Dilemma payoff matrix. Discuss why mutual defection is the equilibrium despite being collectively suboptimal.
  2. 10MExplain the Minimax algorithm with a complete worked example on a small game tree, including pseudocode. Discuss its time/space complexity and practical limitations.
  3. 10MExplain Alpha-Beta Pruning with a worked example showing exactly which branches get pruned and why. Discuss the impact of move ordering on pruning efficiency.
  4. 10MExplain the four phases of Monte Carlo Tree Search (Selection, Expansion, Simulation, Backpropagation) with the UCB1 formula, and discuss why MCTS outperforms Minimax for games like Go.
  5. 8MExplain Expectimax search for stochastic games with a worked example involving chance nodes. Discuss why standard Alpha-Beta pruning needs modification for such games.
  6. 8MDiscuss partially observable (imperfect-information) games, the role of mixed strategies, and why bluffing is rational in game-theoretic terms, using Poker as an example.
  7. 10MDefine Constraint Satisfaction Problems formally. Explain map coloring as a CSP, covering variables, domains, constraints, and the constraint graph.
  8. 10MExplain Backtracking Search for CSPs with pseudocode. Discuss the MRV, Degree, and LCV heuristics, and explain Forward Checking and Arc Consistency as constraint propagation techniques.
  9. 10MExplain the Min-Conflicts local search algorithm for CSPs with a worked N-Queens example. Compare it with Backtracking Search in terms of completeness and practical performance on large problems.
  10. 8MCompare Minimax, Alpha-Beta Pruning, and Monte Carlo Tree Search across completeness, optimality, evaluation function requirements, and suitability for different game types.

Short Answer Questions (2–3 Marks)

  1. 3MList three applications of AI in financial modeling.
  2. 2MWhat is algorithmic trading?
  3. 2MDefine churn prediction in the context of AI-driven marketing.
  4. 2MHow does Maximum Expected Utility apply to a bank's loan-approval decision?
  5. 2MHow can Bayesian networks be applied in medical diagnosis?
  6. 2MWhy is treatment sequencing naturally modeled as an MDP?
  7. 2MWhy does human-in-the-loop oversight remain standard in clinical AI deployment?
  8. 2MDefine predictive policing.
  9. 2MDifferentiate place-based and person-based predictive policing.
  10. 3MExplain the feedback-loop problem in predictive policing.
  11. 2MWhat is a proxy variable, and how can it cause bias even in a "race-blind" model?
  12. 3MList the six core pillars of AI ethics discussed in this module.
  13. 3MList and briefly define five sources of bias in AI systems.
  14. 2MDifferentiate demographic parity and equalized odds as fairness metrics.
  15. 2MWhy is human oversight considered an ethical safeguard, not just an engineering best practice?

Long Answer Questions (8–10 Marks)

  1. 8MDiscuss the applications of AI in finance and marketing, explaining how decision theory and probabilistic models underpin credit scoring, fraud detection, and risk management.
  2. 10MDiscuss the applications of AI in healthcare across diagnosis, treatment planning, and data analytics, explicitly connecting each application to a technique from Modules 1–4 (e.g., Bayesian networks, MDPs).
  3. 10MDiscuss predictive policing and crime analysis in AI-based law enforcement. Explain the feedback-loop and proxy-bias problems in detail, with proposed safeguards.
  4. 10MExplain the six core pillars of AI ethics (fairness, transparency, accountability, privacy, safety, human oversight) with a real-world concern for each.
  5. 8MExplain the five sources of bias in AI systems (historical, representation, measurement, aggregation, evaluation) with an example of each.
  6. 10MDiscuss why fairness metrics like demographic parity and equalized odds can be mathematically incompatible, and what this implies for ethical AI deployment practice.
  7. 8MPropose a complete ethical decision-making framework for an AI project, covering stakeholder identification, bias auditing, explainability, human oversight, accountability, and post-deployment monitoring.