AI Foundations practice questions

Worked examples for AI Foundations

1. Question 1

   How do artificial intelligence (AI), machine learning (ML), and deep learning (DL) relate to one another?

   1. A. Machine learning is the broadest field, artificial intelligence is a subset of it, and deep learning is a subset of artificial intelligence.

      Inverts the nesting by putting machine learning at the top. Machine learning is one approach *within* the wider AI field, not its parent.

   2. B. The three are separate, non-overlapping fields that solve different classes of problems.

      Treats the three as disjoint disciplines. They are concentric: every deep learning system is a machine learning system, and every machine learning system is an AI system.

   3. C. Artificial intelligence is the broadest field, machine learning is a subset of artificial intelligence, and deep learning is a subset of machine learning.Correct answer

      Matches the standard containment AI ⊃ ML ⊃ DL: AI is the umbrella discipline of machines performing tasks that require human-like intelligence, ML is the subset that learns patterns from data, and DL is the ML subset built on multi-layer neural networks.

   4. D. Deep learning is the broadest field and contains machine learning, which in turn contains artificial intelligence.

      Reverses the hierarchy completely. Deep learning is the narrowest of the three — a specific neural-network technique — not the containing field.

   Explanation

   Oracle describes AI as the broad discipline of building systems that perform tasks normally requiring human intelligence, machine learning as the branch of AI in which systems learn patterns from data instead of following hand-coded rules, and deep learning as the branch of machine learning that uses multi-layer neural networks. The relationship is therefore strictly concentric — AI ⊃ ML ⊃ DL. Answers that place machine learning or deep learning at the top invert that nesting, and treating the three as unrelated fields ignores that each inner term is a specialization of the one outside it.

2. Question 2

   A hospital is deploying an AI system that helps triage incoming patients. Which TWO of the following are recognized responsible/ethical AI principles that should govern the deployment?

   1. A. Transparency — those affected can learn when AI is being used and understand how its decisions are reached.Correct answer

      Transparency and explainability are core responsible-AI principles, and they matter most in high-stakes settings such as healthcare where people must be able to understand and question a decision.

   2. B. Model velocity — pushing new model versions to production as frequently as the team can build them.

      An invented 'principle'. Release cadence is a delivery practice; shipping faster carries no ethical guarantee and in a clinical setting can increase risk.

   3. C. Maximum throughput — processing the largest possible number of cases per second.

      Mistakes a performance objective for an ethical principle. Throughput says nothing about whether outcomes are fair, explainable, or answerable to anyone.

   4. D. Accountability — a clearly identified organization or role answers for the system's outcomes and provides a route to redress.Correct answer

      Accountability is a core responsible-AI principle: humans, not the model, remain answerable for outcomes, and affected people need a path to challenge or appeal a decision.

   Explanation

   Responsible AI rests on principles such as fairness, transparency and explainability, accountability, privacy, and safety — so being able to explain how decisions are reached and having an identified party answer for outcomes both belong on the list. Frequent releases are a delivery cadence with no ethical content, and high case throughput is a performance target that says nothing about whether the system's outcomes are fair, explainable, or answerable.

3. Question 3

   A loan applicant is declined by an automated model and asks which factors drove the decision. Which responsible-AI principle is most directly at stake?

   1. A. Fairness

      Fairness concerns whether the outcome is biased against a group. The applicant's request is about being able to see the reasoning, which is a separate principle.

   2. B. Transparency and explainabilityCorrect answer

      Transparency/explainability is the principle that an AI system's decisions should be understandable and its reasoning communicable to the people affected — exactly what the applicant is asking for.

   3. C. Scalability

      Scalability is an engineering property about handling growing load. It is not a responsible-AI principle at all.

   4. D. Model accuracy

      Treats a quality metric as an ethics principle. Even a highly accurate model leaves the applicant with no insight into why the decision went the way it did.

   Explanation

   When someone affected by an automated decision needs to know how that decision was reached, the principle at issue is transparency and explainability — AI systems should not be opaque to the people they act on. Whether the outcome is biased against a group would be fairness, engineering capacity to handle load is scalability rather than an ethical principle, and predictive accuracy is a performance measure that says nothing about whether reasoning can be explained.

4. Question 4

   In responsible AI, which concern does the principle of *fairness* primarily address?

   1. A. Ensuring a human or organization can be held answerable for the system's outcomes.

      Describes accountability, a different responsible-AI principle. Fairness is about who is affected and how equitably, not about who answers for the result.

   2. B. Ensuring stakeholders can understand how the system reached a decision.

      Describes transparency and explainability. A system can be fully explainable and still produce biased outcomes, so the two principles are distinct.

   3. C. Ensuring the system does not produce biased or discriminatory outcomes for particular groups of people.Correct answer

      Fairness is the responsible-AI principle concerned with avoiding bias — often inherited from unrepresentative training data — that would systematically disadvantage certain groups.

   4. D. Ensuring the model achieves the highest possible accuracy on its test data.

      Mistakes a performance metric for an ethical principle. A highly accurate model can still be unfair if its errors fall disproportionately on one group.

   Explanation

   Fairness in responsible AI means guarding against bias so that the system's outputs do not systematically disadvantage particular groups, a risk that arises when training data reflects historical or unrepresentative patterns. Being able to say who is answerable for an outcome is accountability, being able to understand how a decision was reached is transparency and explainability, and raw predictive accuracy is a performance measure rather than an ethical principle at all.

5. Question 5

   What most fundamentally distinguishes a machine learning approach from a traditional rule-based program?

   1. A. Machine learning derives its decision logic from patterns in data, while a rule-based program follows logic a developer wrote explicitly.Correct answer

      This is the defining distinction: an ML model is trained on examples and infers its own rules, whereas traditional software executes conditions a human enumerated in advance.

   2. B. Machine learning always runs in the cloud, while rule-based programs run only on-premises.

      Confuses deployment location with methodology. Either approach can run anywhere; where code executes says nothing about how its logic was produced.

   3. C. Machine learning produces exact answers, while rule-based programs produce probabilistic ones.

      Backwards. ML models typically output probabilistic or approximate predictions, while deterministic rule-based logic yields the same exact result for the same input.

   4. D. Machine learning requires no data, while rule-based programs require large datasets.

      Inverts the data dependency. Training data is the essential ingredient for ML, whereas a hand-coded rule engine can function without any historical dataset.

   Explanation

   Machine learning is defined by learning: the system is shown examples and infers the patterns that map inputs to outputs, rather than executing decision logic a programmer enumerated by hand. Where the workload is hosted is a deployment concern unrelated to this distinction, ML outputs are generally probabilistic rather than exact while rule engines are deterministic, and it is machine learning — not rule-based software — that depends on substantial training data.

6. Question 6

   A retailer has purchase histories for millions of shoppers but no predefined customer segments. It wants the system to discover natural groupings of similar shoppers on its own. Which task category fits this goal?

   1. A. Classification

      Classification requires a predefined set of labels and labeled training examples. Here no segments exist yet — the groups are what the system must discover.

   2. B. Regression

      Regression predicts a continuous numeric target. The retailer wants groupings of customers, not a numeric estimate for each one.

   3. C. ClusteringCorrect answer

      Clustering is the unsupervised task of grouping unlabeled records so that similar records fall together — exactly what discovering customer segments from raw purchase history requires.

   4. D. Anomaly detection

      Anomaly detection singles out the rare records that do not fit normal patterns. The retailer wants to organize the whole customer base into groups, not isolate a few outliers.

   Explanation

   When the data carries no labels and the objective is to let the algorithm find structure by grouping similar records, the task is clustering — an unsupervised technique. Classification would require the segments to be defined and labeled in advance, regression would produce a continuous number rather than group membership, and anomaly detection targets the unusual minority of records rather than partitioning the entire population.

7. Question 7

   An architecture review board is documenting terminology for a new initiative. One team builds a rules-based scheduling engine with hand-coded logic, a second team trains a gradient-boosted decision-tree model on historical tickets, and a third team trains a multi-layer neural network on raw images. Which statement most accurately describes how artificial intelligence, machine learning, and deep learning relate to these three efforts?

   1. A. Only the neural-network effort qualifies as artificial intelligence, because artificial intelligence is defined by the use of deep neural networks.

      Equates AI exclusively with deep learning. AI is the broad field of systems performing tasks that normally require human intelligence, and it includes both symbolic/rules-based approaches and non-deep machine learning such as decision trees.

   2. B. All three fall under artificial intelligence; the decision-tree and neural-network efforts are both machine learning because they learn from data, and only the neural-network effort is additionally deep learning.Correct answer

      Correct containment: AI ⊃ ML ⊃ DL. AI is the umbrella field (including rules-based systems), machine learning is the subset whose behaviour is learned from data rather than explicitly coded, and deep learning is the subset of ML built on multi-layer neural networks.

   3. C. Machine learning is the broadest of the three fields, artificial intelligence is a specialization of it, and deep learning is a separate discipline that overlaps neither.

      Inverts the nesting and detaches deep learning. Machine learning is a subset of AI, not its parent, and deep learning sits strictly inside machine learning rather than beside it.

   4. D. Artificial intelligence, machine learning, and deep learning are three non-overlapping disciplines, so each team's effort belongs to exactly one of them.

      Treats the three terms as mutually exclusive categories. They are nested concentric sets, so a deep learning system is simultaneously machine learning and artificial intelligence.

   Explanation

   Artificial intelligence is the broad field of building systems that perform tasks normally requiring human intelligence, and it spans explicitly programmed rules as well as learned behaviour. Machine learning is the subset of AI in which a model infers patterns from data instead of following hand-coded logic, and deep learning is the further subset of machine learning that uses multi-layer neural networks to learn representations from raw input. Reading the relationship as an inversion (ML containing AI), as three disjoint disciplines, or as AI meaning only neural networks all contradict this concentric AI ⊃ ML ⊃ DL structure.

8. Question 8

   Which TWO scenarios are best modeled as classification tasks?

   1. A. Predicting tomorrow's temperature in degrees Celsius

      The target is a continuous quantity on a numeric scale, which makes this regression rather than classification.

   2. B. Deciding whether a medical image shows a benign or malignant lesionCorrect answer

      The model assigns each image to one of two predefined labels using labeled training examples — a textbook binary classification problem.

   3. C. Grouping news articles into themes that were not defined in advance

      No labels exist beforehand and the groups must be discovered from the data, which makes this unsupervised clustering, not classification.

   4. D. Tagging an incoming customer review as positive, neutral, or negative sentimentCorrect answer

      Each review is assigned to one of three predefined discrete categories, which is multi-class classification.

   5. E. Estimating how many units of a product will sell next quarter

      A demand forecast produces a continuous numeric quantity, which is regression; mislabeling forecasting as classification is a common category error.

   Explanation

   Classification is the supervised task of assigning an input to one of a fixed set of predefined labels, so diagnosing an image as benign or malignant and tagging a review's sentiment both qualify. Predicting a temperature or a unit sales figure produces a continuous value and is therefore regression, while grouping articles into themes that were never defined in advance has no labels to predict and is clustering.