Artificial Intelligence Viva-Voce & University Examination Guide

This guide is specifically curated for university students appearing for BCA, MCA, B.Tech, and B.Sc. Computer Science practical lab examinations and oral viva-voce assessments. University examiners evaluate your conceptual clarity, ability to state concise definitions, understanding of algorithm complexity, and knowledge of classical vs modern AI paradigms.

Rapid-Fire Oral Examination Definitions

Concept / Term	1-Sentence Formal Definition for Viva	Key Formula / Characteristic
Artificial Intelligence	The branch of computer science focused on building computational systems capable of performing tasks that typically require human intelligence, such as visual perception, natural language understanding, reasoning, and learning.	Defined by Stuart Russell as the study of Rational Agents.
Turing Test	An operational test of machine intelligence proposed by Alan Turing in 1950, where a human interrogator must determine which of two hidden conversational partners is a human and which is a computer.	Passes if interrogator cannot reliably tell computer apart after 5 minutes.
PEAS Framework	A formal specification framework used to define an intelligent agent's complete task environment: Performance measure, Environment, Actuators, and Sensors.	PEAS: ( P, E, A, S ).
Heuristic Function	An informed, problem-specific evaluation function ( h(n) ) that estimates the remaining cost from node ( n ) to the nearest goal state.	Admissible if ( 0 le h(n) le h^*(n) ) (never overestimates).
*A Algorithm**	An informed search algorithm that evaluates nodes using ( f(n) = g(n) + h(n) ), guaranteed to find the optimal path if ( h(n) ) is admissible.	( f(n) = g(n) + h(n) ).
Alpha-Beta Pruning	An optimization technique for the minimax game tree search that eliminates branches that cannot possibly influence the final decision, reducing effective branching factor to ( sqrt{b} ).	Prunes whenever ( alpha ge eta ).
Arc Consistency (AC-3)	A constraint propagation algorithm ensuring that for every legal value of variable ( X ), there exists some compatible legal value in neighbor variable ( Y ).	Prunes domains in ( O(c cdot d^3) ) time.
Unification	The algorithmic procedure in First-Order Logic that finds a variable substitution ( heta ) that makes two distinct logical expressions syntactically identical.	( ext{UNIFY}(P, Q) = heta ).
Resolution Refutation	A complete proof-by-contradiction inference rule in Clausal Normal Form that repeatedly resolves complementary literals until the empty clause (( Box )) is derived.	( (A ee B) wedge ( eg B ee C) models (A ee C) ).
Supervised Learning	A machine learning paradigm where the algorithm learns a mapping function from input features ( X ) to ground-truth labeled targets ( y ) provided in the training data.	( mathcal{D} = {(x_i, y_i)}_{i=1}^N ).
Confusion Matrix	A ( 2 imes 2 ) table recording True Positives, True Negatives, False Positives, and False Negatives produced by a classification model.	Basis for Precision, Recall, Specificity, and F1-Score.
Perceptron	The simplest artificial neural network consisting of a single neuron computing a weighted linear sum followed by a step or threshold activation function.	( y = ext{step}(mathbf{w}^T mathbf{x} + b) ).
Backpropagation	The algorithm used to train neural networks by computing the partial derivatives of the loss function with respect to every weight parameter using the multivariable calculus chain rule.	( rac{partial mathcal{L}}{partial mathbf{W}} = oldsymbol{delta} mathbf{a}^T ).
Transformer	A deep neural network architecture introduced in 2017 that discards recurrence in favor of the Self-Attention mechanism, processing all tokens in parallel.	( ext{Attention}(Q, K, V) = ext{softmax}left(rac{QK^T}{sqrt{d_k}} ight)V ).

Standard Viva Diagram Questions

Be prepared to sketch the following diagrams on paper for the viva examiner:

The Intelligent Agent Interaction Cycle: Sensors taking percepts from Environment → Agent Function / Program → Actuators executing actions into Environment.
The Binary Confusion Matrix: Grid with Predicted vs Actual, labeling TP, TN, FP (Type I), and FN (Type II).
The Artificial Neuron: Inputs ( x_i ), multiplied by weights ( w_i ), entering summation circle ( sum w_i x_i + b ), passing through activation block ( sigma(z) ), yielding output ( a ).
The Transformer Self-Attention Head: Input embeddings projecting into Query, Key, and Value vectors, computing scaled dot-product and softmax.

Frequently Asked Questions (FAQ) - Student & Interview Review

Q1. What is the difference between Strong AI and Weak AI?

Weak (Narrow) AI is designed and trained for a specific task (such as playing chess, facial recognition, or web search); all existing AI systems today are Narrow AI. Strong (General) AI (AGI) refers to a hypothetical machine that possesses human-like generalized intelligence, consciousness, and self-awareness, capable of solving any cognitive problem a human can.

Q2. What is the difference between BFS and DFS in terms of memory?

BFS has an exponential space complexity of O(b^d) because it must store all unexpanded frontier nodes at the current depth in a queue. DFS has a linear space complexity of O(b*m) because it stores only the single active path from root to current leaf plus unexplored siblings.

Q3. What is the difference between Forward Chaining and Backward Chaining?

Forward Chaining is data-driven: it starts with known facts in working memory and fires IF-THEN rules forward to deduce new conclusions. Backward Chaining is goal-driven: it starts with a target hypothesis and works backward to see if supporting premises are present in memory.

Q4. What is an Admissible Heuristic in A* Search?

An admissible heuristic is one that never overestimates the true cost to reach the nearest goal state (it is optimistic: h(n) <= h*(n)). An admissible heuristic guarantees that A* tree search will return an optimal, lowest-cost path.