Artificial Intelligence practice questions (Jahziah exam)

Area 01 · Algorithms, computability & complexity

A programmer writes an algorithm to compute the greatest common divisor (GCD) of two positive integers by repeatedly replacing the larger number with the remainder of dividing it by the smaller number until the remainder is zero. Which statement best identifies this numerical algorithm and explains its correctness?

Pick an answer to check yourself.

1. AIt is the Euclidean algorithm; it is correct because gcd(a, b) = gcd(b, a mod b) for b ≠ 0, so replacing the pair preserves the GCD while shrinking the numbers until the remainder is zero and the other number is the GCD.
2. BIt is the Sieve of Eratosthenes; it is correct because marking multiples of each prime eliminates all composite numbers and leaves the GCD.
3. CIt is a linear search over all integers from 1 to the smaller number, keeping the largest common divisor found; it terminates because the smaller number is finite.
4. DIt is a sorting algorithm that orders the divisors and returns the largest shared divisor, so its correctness follows from the sorted order.

Area 02 · Programming

A program reads a numeric score and must classify it using this conditional structure: if score >= 90: grade = "A" elif score >= 75: grade = "B" elif score >= 60: grade = "C" else: grade = "F" If score has the value 75, what is the value of grade after the structure executes?

Pick an answer to check yourself.

1. A"A"
2. B"B"
3. C"C"
4. D"F"

Area 03 · Applied computer science

A DBMS uses a page-based storage manager. Each page is 8 KB and carries a 256-byte header. Records are fixed length at 128 bytes. What is the maximum number of whole records that fit on one page?

Pick an answer to check yourself.

1. A60 records
2. B64 records
3. C62 records
4. D61 records

Area 04 · Networks

A packet is sent from a host on network A to a server on network B across two routers. As the packet traverses the path, which statement correctly describes the behavior of the source and destination IP addresses and MAC addresses?

Pick an answer to check yourself.

1. AThe source and destination IP addresses change at each router, while the MAC addresses remain constant.
2. BThe source and destination IP addresses remain constant, while the source and destination MAC addresses change at each router.
3. CBoth IP addresses and MAC addresses remain constant along the entire path.
4. DBoth IP addresses and MAC addresses change at each router.

Area 05 · Software engineering

A component that transfers money between accounts has this contract: - Precondition: source and destination accounts are distinct and both exist. - Postcondition: the source balance decreases by the amount, the destination balance increases by the amount, and the total of the two balances is unchanged. A caller invokes the component with the same account as source and destination. The component performs no operation and returns normally. Which analysis is most accurate?

Pick an answer to check yourself.

1. AThe component violated the postcondition because it did not change the balances as specified, even though the precondition was violated.
2. BThe component's behavior is acceptable under the contract: since the precondition was violated, it is not bound by the postcondition, and doing nothing is a valid choice.
3. CThe contract is inconsistent because if source and destination are the same, the postcondition would require the balance to both decrease and increase by the amount, which is impossible.
4. DThe component should have thrown an exception to signal the precondition violation, so returning normally is a contract violation.

Area 06 · Fundamentals of AI

Two systems are compared. System 1 answers every question by consulting a large lookup table that maps each stored question to a stored human answer; for any question not in the table it replies "I don't know." System 2 answers by composing new sentences from rules and it handles unseen questions, though some of its answers are wrong. A researcher argues that System 1 is 'more intelligent' because every answer it gives matches a real human's. Analysing what the Turing test can and cannot distinguish, which evaluation is correct?

Pick an answer to check yourself.

1. ASystem 1 is more intelligent, because matching real human answers on every stored question is stronger evidence of intelligent behaviour than System 2's approximated answers.
2. BSystem 1 is more intelligent, because it always produces a human answer whereas System 2 sometimes produces incorrect answers.
3. CSystem 2 better exhibits the target of the test, because it generates responses to unseen questions from rules, whereas System 1's lookup table reproduces stored human answers and cannot respond to unanticipated interrogation.
4. DThe two systems are equivalent under the test, because the test compares only the number of correct answers and both can be correct on stored questions.

Area 07 · Machine learning

A multi-layer feed-forward network uses sigmoid activations in all hidden layers and is trained with backpropagation on inputs normalised to small values near zero. The first hidden layer receives input activations that are concentrated near zero. Which statement best analyses the consequence for the first-layer weight updates?

Pick an answer to check yourself.

1. AWeight updates in the first layer are large because the sigmoid derivative is maximal at zero input.
2. BWeight updates in the first layer are small because the sigmoid derivative is near its maximum but the deep chain of sigmoid derivatives is small.
3. CWeight updates in the first layer are zero regardless of input, because sigmoid saturates for all inputs.
4. DFirst-layer weights do not receive gradients at all, because backpropagation stops at the last hidden layer.

Area 08 · Ethics, privacy & security

A hiring model is trained on ten years of past decisions in which one group was historically hired at a lower rate. The model reproduces that gap. Which intervention most directly targets the bias present in the training data?

Pick an answer to check yourself.

1. ARemove the group-membership attribute from the features and retrain.
2. BCollect more historical data of the same kind to improve the model's accuracy.
3. CReweigh or resample the training examples so that historically disadvantaged outcomes contribute proportionally, then retrain and audit group-level error rates.
4. DIncrease the model's complexity until training accuracy reaches 100%.