This final set of Artificial Intelligence MCQs covers reasoning under uncertainty and several applied topics. It includes Bayes rule calculations, Bayesian network factorisation and parameter counting, conditional independence and the Markov blanket, fuzzy set operations, STRIPS and partial-order planning, the components of expert systems, and core NLP ideas such as n-grams, lemmatisation and ambiguity. It helps GATE and UGC NET aspirants, university semester candidates and anyone preparing for AI or ML interviews.
How to practise: read each question and tap the option you think is correct. You will see right away whether you got it, along with a short explanation, and your score is tracked at the bottom of the screen. This set has 18 questions.
Topics covered in this set
- Bayes rule and conditional probability
- Bayesian networks and conditional independence
- Fuzzy logic operations and defuzzification
- STRIPS and partial-order planning
- Expert systems and their components
- NLP basics: n-grams, lemmatisation, ambiguity, Naive Bayes
Artificial Intelligence MCQs with answers and explanations
Q1. A disease affects 1% of a population. A test detects it in 90% of sick people and gives a false positive for 5% of healthy people. If a person tests positive, what is the approximate probability that they have the disease?
a) 90%
b) 50%
c) 15.4%
d) 5.3%
Q2. For the Bayesian network below, what is P(A = true, B = true, C = true)?
Network: A -> B -> C (all Boolean) P(A = true) = 0.3 P(B = true | A = true) = 0.8 P(C = true | B = true) = 0.5
a) 0.12
b) 0.24
c) 0.40
d) 0.15
Q3. In a Bayesian network, a Boolean node has three Boolean parents. How many independent probabilities are needed in its conditional probability table?
a) 3
b) 6
c) 4
d) 8
Q4. How many independent numbers are needed to specify the full joint distribution over 5 Boolean random variables with no independence assumptions?
a) 32
b) 31
c) 10
d) 25
Q5. In a Bayesian network, each node is conditionally independent of which set of variables given its parents?
a) Its children
b) All other variables in the network
c) Its non-descendants
d) Its ancestors' children only
Q6. A node's Markov blanket in a Bayesian network consists of:
a) its parents, its children and its children's other parents
b) its parents only
c) all of its ancestors and descendants
d) its children and grandchildren
Q7. In fuzzy logic with standard min/max operators, mu_A(x) = 0.7 and mu_B(x) = 0.4. What is the membership of x in A AND (NOT B)?
a) 0.4
b) 0.3
c) 0.7
d) 0.6
Q8. In a fuzzy inference system, what is the purpose of defuzzification methods such as the centroid method?
a) To convert crisp inputs into membership values
b) To convert the aggregated fuzzy output into a single crisp value
c) To learn membership functions from data
d) To check the consistency of fuzzy rules
Q9. A STRIPS action schema is specified by:
a) a precondition list, an add list and a delete list
b) a utility function and a transition matrix
c) a set of Horn clauses and a goal test
d) a heuristic and a cost function
Q10. How does the STRIPS representation sidestep the frame problem?
a) By listing every fact that stays true after each action
b) By using probabilistic effects
c) By assuming that any literal not mentioned in an action's effects remains unchanged
d) By forbidding actions with preconditions
Q11. The Sussman anomaly in the blocks world demonstrates that:
a) forward state-space search is always optimal
b) achieving subgoals one after another without interleaving can fail to find a plan
c) STRIPS cannot represent the blocks world
d) partial-order planning is incomplete
Q12. What principle characterises partial-order planning?
a) Always add the step that reduces the heuristic most
b) Order every pair of steps as soon as they are added
c) Plan backwards from a single goal literal at a time
d) Least commitment: order steps only when necessary to resolve threats or causal links
Q13. Which two components form the core of a rule-based expert system?
a) Knowledge base and inference engine
b) Sensors and actuators
c) Training set and test set
d) Compiler and linker
Q14. What is the purpose of the explanation facility in an expert system?
a) To acquire new rules automatically from data
b) To translate rules into machine code
c) To show the user how a conclusion was reached and why a question is being asked
d) To store the working memory of facts
Q15. How does lemmatisation differ from stemming in NLP?
a) Lemmatisation simply chops suffixes using fixed rules
b) Lemmatisation uses vocabulary and morphological analysis to return a dictionary form, for example 'better' to 'good'
c) Lemmatisation works only on nouns
d) Lemmatisation always produces shorter strings than stemming
Q16. In a corpus, the word 'the' occurs 10 times and the bigram 'the cat' occurs 3 times. What is the maximum-likelihood bigram estimate P(cat | the)?
a) 0.03
b) 0.10
c) 0.13
d) 0.30
Q17. The sentence 'I saw the man with the telescope' can mean that either the speaker or the man has the telescope. What type of ambiguity is this?
a) Syntactic (structural) ambiguity
b) Lexical ambiguity
c) Phonological ambiguity
d) Anaphoric ambiguity
Q18. What simplifying assumption does a Naive Bayes classifier make?
a) All classes are equally likely
b) Features follow a normal distribution
c) Features are conditionally independent of each other given the class
d) The class is independent of the features
Quick revision notes
- Bayes rule: P(H | E) = P(E | H) P(H) / P(E).
- Bayesian network joint: P(X1, ..., Xn) = product of P(Xi | Parents(Xi)).
- Boolean node with k Boolean parents needs 2^k CPT numbers; full joint of n Boolean variables needs 2^n - 1.
- Markov blanket = parents + children + children's other parents.
- Fuzzy AND = min, OR = max, NOT = 1 - mu; centroid is a common defuzzification method.
- STRIPS action = preconditions + add list + delete list; partial-order planning follows least commitment.
- Expert system = knowledge base + inference engine (+ explanation facility, user interface); MYCIN used backward chaining with certainty factors.
- Bigram MLE: P(w2 | w1) = count(w1 w2) / count(w1).
Frequently asked questions
Why do Bayesian networks need fewer numbers than a full joint table?
They exploit conditional independence so each variable depends only on its parents. If each node has at most k parents, the network needs about n * 2^k numbers instead of 2^n.
How is fuzzy logic different from probability?
Probability measures how likely an uncertain crisp event is, while fuzzy membership measures the degree to which something belongs to a vaguely defined set. A glass can be 0.6 'full' without any randomness involved.
What is the difference between STRIPS and partial-order planning?
STRIPS is a representation for states and actions using preconditions, add lists and delete lists. Partial-order planning is a search strategy that can use such actions while leaving steps unordered until an ordering is required.
Comments
Post a Comment