AI Fundamentals

Practice questions

An office worker uses an AI tool to draft a reply to a citizen. The draft states a wrong fee, and the worker sends it unchecked. Who is responsible for the error?

  1. The AI tool, because it wrote the figure
  2. The citizen, for believing the letter
  3. Nobody, because AI errors are nobody's fault
  4. The office that sent the letter, which should have checked the fee against the official source
Answer and explanation

Answer: D

AI cannot be held accountable; the people and organisations using it are. The figure should have been checked against the official source before the letter went out.

A spam filter lets 9 scam emails into the inbox and moves 4 genuine emails to the spam folder. How many false negatives and false positives are there?

  1. 4 false negatives and 9 false positives
  2. 9 false negatives and 4 false positives
  3. 13 false positives and no false negatives
  4. 9 false negatives and no false positives
Answer and explanation

Answer: B

Scams let through are cases where the filter wrongly said 'not spam': 9 false negatives. Genuine emails marked spam are cases where it wrongly said 'spam': 4 false positives.

A realistic video shared in a group shows an official announcing that an exam is postponed. What should a candidate do?

  1. Believe it because the face and voice look real
  2. Forward it quickly so others know
  3. Check the official website or notice of the exam body before believing or sharing it
  4. Ask a chatbot whether the video is real and accept its answer
Answer and explanation

Answer: C

Deepfakes can copy faces and voices, and a chatbot can hallucinate, so neither appearance nor an AI's opinion settles it. Only the official source confirms the news.

After watching a few videos of one kind, a student finds her video app shows almost nothing else. What explains this?

  1. The recommendation system predicts from her past choices and keeps showing more of the same - a filter bubble
  2. The app has general intelligence and has decided what she should study
  3. A fraud filter has blocked all other videos
  4. Computer vision has stopped working
Answer and explanation

Answer: A

Recommendation systems learn from past choices and similar users, so they can narrow what a person sees. Searching for other topics deliberately widens it again.

A hospital uses AI to screen eye scans. Which practice is correct?

  1. Patients whose scans are not flagged are told they are healthy without any check
  2. The AI's flag is treated as the final diagnosis
  3. The AI prescribes treatment for flagged patients
  4. Flagged scans are reviewed first by a doctor, who examines the patient and decides
Answer and explanation

Answer: D

In screening, AI only marks images that may show disease so a doctor can look at them first. Diagnosis and treatment stay with the doctor, and the usual medical checks continue for everyone.

A farmer speaks a question in Marathi about a photo of a leaf, and the app replies aloud in Marathi. Which combination of AI is used?

  1. Fraud detection, recommendation and robotics
  2. Speech recognition, computer vision, NLP and speech synthesis
  3. Computer vision only
  4. An expert system and reinforcement learning only
Answer and explanation

Answer: B

Speech recognition turns the spoken question into text, computer vision examines the leaf photo, NLP forms the answer in Marathi, and speech synthesis reads it aloud.

Why are GPUs well suited to training deep neural networks?

  1. They store more photographs than any other device
  2. They remove the need for training data
  3. They perform many simple calculations at the same time
  4. They make the network explain its decisions
Answer and explanation

Answer: C

Training a deep network needs enormous numbers of simple calculations, and GPUs, first built for graphics, do many of them in parallel. They do not replace data or make a model explainable.

During the training of a neural network, what is changed step by step?

  1. The labels of the training examples
  2. The weights on the connections
  3. The test data kept aside for checking
  4. The number of pixels in each photo
Answer and explanation

Answer: B

Training compares the network's answers with the labels and nudges the weights to reduce the error. The labels and test data must stay fixed, or the check would be meaningless.

A neuron has inputs 1, 0 and 1 with weights 0.3, 0.6 and 0.2, and fires only if the weighted sum is above 0.6. What happens?

  1. The sum is 1.1, and it fires
  2. The sum is 0.8, and it fires
  3. The sum is 0.5, and it fires
  4. The sum is 0.5, and it does not fire
Answer and explanation

Answer: D

Weighted sum = 1 × 0.3 + 0 × 0.6 + 1 × 0.2 = 0.3 + 0 + 0.2 = 0.5. Since 0.5 is not above 0.6, the neuron does not fire. The weight 0.6 does not count because its input is 0.

An office must be able to point to the exact reason for every rejected application, and its rules are clear and rarely change. Which approach suits it best?

  1. A rules-based system whose rules can be read and explained
  2. A deep neural network trained on past decisions
  3. A chatbot that writes a new decision each time
  4. Unsupervised learning that groups applicants
Answer and explanation

Answer: A

Rules-based systems are easy to explain because the rule behind each decision can be read. Clear, stable rules and the need to explain every decision point to them; learning systems are often hard to explain.

Which statement is FALSE?

  1. All deep learning is machine learning
  2. All machine learning is AI
  3. All machine learning is deep learning
  4. Some AI uses hand-written rules instead of learning
Answer and explanation

Answer: C

Deep learning is only one part of machine learning; many machine-learning methods do not use deep neural networks. The other statements are true: DL lies inside ML, ML lies inside AI, and expert systems are AI built from hand-written rules.

A company trains the same learning algorithm on two different datasets - one of Hindi emails, one of English emails. What results?

  1. One model shared by both
  2. No model, because the algorithm is the model
  3. The two datasets merge automatically
  4. Two different models, because a model depends on the data it was trained on
Answer and explanation

Answer: D

The algorithm is the learning method; the model is what that method produces from particular data. Different data give different models.

How many years passed between the Dartmouth workshop that named AI and Deep Blue's win over Kasparov?

  1. 6 years
  2. 41 years
  3. 47 years
  4. 51 years
Answer and explanation

Answer: B

1997 − 1956 = 41 years. (Turing's 1950 paper to Deep Blue is 47 years, and Turing's paper to Dartmouth is 6 years.)

Arrange in time order: (1) Deep Blue beats Kasparov, (2) the term 'artificial intelligence' is coined for the Dartmouth workshop, (3) Turing proposes the imitation game.

  1. 2, 3, 1
  2. 1, 3, 2
  3. 3, 2, 1
  4. 3, 1, 2
Answer and explanation

Answer: C

Turing's paper is 1950 (3), the Dartmouth workshop is 1956 (2) and Deep Blue's win is 1997 (1). Order: 3, 2, 1.

A student says: 'Because a chess computer can beat a world champion, it is general AI.' What is wrong with this?

  1. A chess computer is narrow AI: excellent at chess but unable to do unrelated tasks
  2. Nothing; beating a champion proves general intelligence
  3. Chess computers are not AI at all
  4. General AI is the name for game-playing programs
Answer and explanation

Answer: A

General AI would handle any intellectual task a human can. A chess program excels only at chess and cannot, say, translate or drive, so it is narrow AI.

A fraud filter flags 60 payments: 51 are real fraud and 9 are genuine. It also lets 6 real frauds through. How many real frauds were there in total?

  1. 51
  2. 57
  3. 60
  4. 66
Answer and explanation

Answer: B

Real frauds = frauds caught + frauds missed = 51 + 6 = 57. The 9 genuine payments flagged are false positives and are not counted as fraud.

A recruitment model trained on a company's past hiring records keeps ranking applicants from one background lower. The most likely cause is:

  1. Hallucination in a chatbot
  2. Too many hidden layers
  3. Bias in the training data, which reflected past unfair decisions
  4. A false negative in the fraud filter
Answer and explanation

Answer: C

A model learns the patterns in its data. If past decisions were unfair, the model repeats that unfairness - this is bias. Checking results across groups and human review are needed.

A clerk wants an AI tool to improve the wording of a letter to a citizen. Which is the right practice?

  1. Remove the citizen's name, Aadhaar number and other personal details before pasting the text
  2. Paste the full letter with all details so the tool has context
  3. Paste the office password as well so the tool can send the letter
  4. Paste the letter and send the AI's version without reading it
Answer and explanation

Answer: A

Personal data and passwords must never go into AI tools, which may store what is typed. Removing personal details protects privacy, and the clerk still reads and checks the result before sending.

You receive a voice message that sounds exactly like your uncle, urgently asking you to send money by UPI. What is the safest response?

  1. Send the money at once because the voice matches
  2. Forward the message to family groups to warn them
  3. Reply asking him to send a photo as proof
  4. Call your uncle back on the number you already know before doing anything
Answer and explanation

Answer: D

Voices can be cloned with AI, so a matching voice proves nothing. Verifying through a separate, known channel is the safeguard; a photo could also be faked, and forwarding spreads the scam.

A video app recommends videos to a student who has never searched for them. It most likely works by:

  1. Choosing videos completely at random
  2. Showing only videos uploaded that same day
  3. Predicting from what she watched and what people with similar viewing habits chose
  4. Asking a teacher to pick videos for each user
Answer and explanation

Answer: C

Recommendation systems learn from a user's past choices and from users with similar choices. A side effect is the filter bubble: she may keep seeing more of the same kind of video.

Privacy · Terms · Refunds · Delivery