AI Fundamentals and Literacy

1. A business leader must choose between a rules-based system and a machine learning model for approving expense claims. Which factor most strongly favours machine learning?

1. A The organization requires every decision to follow an explicitly documented rule
2. B There is no historical data available
3. C The rules are few, stable, and written in policy documents
4. D The approval criteria are complex, change over time, and are best inferred from thousands of historical decisions

2. A business problem can be solved by a set of clearly defined rules that rarely change. Which approach should be selected?

1. A Rule-based automation, because the logic is deterministic and does not require a model
2. B A custom machine learning model trained on historical examples
3. C An AI agent that plans its own sequence of steps
4. D A generative AI application built on a foundation model

3. An executive asks why a generative AI pilot produced confident but factually wrong answers about company policy. Which explanation is correct?

1. A The model had insufficient network bandwidth
2. B The training data was encrypted
3. C The model was not trained on the company's internal policies and generated plausible text without grounding in a verified source
4. D The model was configured with too little memory

4. An executive asks how generative AI differs from the predictive machine learning the company already uses. Which distinction is most accurate?

1. A Predictive models require no data while generative models do
2. B Predictive models estimate an outcome from structured inputs; generative models produce new content such as text, images, or code
3. C Generative AI is simply predictive AI running on faster hardware
4. D Generative AI replaces the need for predictive models entirely

5. A business leader is told a proposed use case needs "unstructured data". Which example qualifies?

1. A A sales table with columns for date, region, and revenue
2. B A spreadsheet of monthly headcount
3. C A relational database of product SKUs
4. D Customer support email text and call recordings

6. Which statement best describes why data quality matters more than model choice for most business AI initiatives?

1. A Models cannot be changed once selected
2. B Model performance is bounded by the quality and representativeness of the data it learns from, and no algorithm compensates for unreliable inputs
3. C Poor data quality only affects training speed
4. D Data quality determines the licensing cost of the model

7. A team proposes AI for a task with a small number of stable, legally mandated rules that must be applied identically every time. What should the strategist recommend?

1. A A predictive model trained on past decisions
2. B A hybrid where the model overrides the rules when confident
3. C A deterministic rules engine, which is auditable and behaves identically on every input
4. D A generative model with a detailed prompt

8. An organization must decide whether to build an AI capability in house, buy a vendor product, or partner. Which combination of factors should be weighed? (Select TWO.)

1. A The number of foundation models the vendor offers
2. B The size of the vendor's engineering organization
3. C The programming languages the internal team prefers
4. D Budget and any regulatory compliance obligations the solution must satisfy
5. E Internal capability and the timeline the business requires

9. What does it mean that a foundation model is "stateless" between requests?

1. A It stores every conversation permanently for training
2. B It cannot be fine-tuned
3. C It retains nothing from previous requests; any conversation history must be resupplied in the prompt
4. D It requires no infrastructure to run

10. A vendor claims their model achieves 96 percent accuracy on a task where 95 percent of cases belong to one class. How should a strategist interpret this?

1. A The model is clearly effective and should be adopted
2. B Accuracy above 95 percent always indicates production readiness
3. C The claim proves the dataset was large
4. D Accuracy is nearly meaningless here; ask for precision, recall, and performance on the minority class

11. Which describes the relationship between artificial intelligence, machine learning, and deep learning?

1. A Machine learning and deep learning are the same thing
2. B Artificial intelligence is a subset of deep learning
3. C They are three independent and unrelated fields
4. D Deep learning is a subset of machine learning, which is a subset of artificial intelligence

12. A proposed use case has no historical examples of the outcome to be predicted, and the outcome cannot be reliably labelled. What should a strategist conclude?

1. A The problem should be solved with generative AI instead
2. B The team should proceed and label the data after deployment
3. C A larger model will compensate for the missing labels
4. D Supervised machine learning is not currently viable; the first investment should be in establishing reliable labelled data or reframing the problem

13. A business wants to estimate next quarter's demand for each product line from several years of sales history. Which class of problem is this?

1. A Content generation
2. B Clustering
3. C Forecasting, a form of supervised learning on time-ordered data
4. D Anomaly detection

14. Which limitation should a strategist expect from a general-purpose language model performing precise multi-step arithmetic?

1. A It is always more accurate than a spreadsheet
2. B It may produce confident but incorrect results, so calculations should be delegated to a tool or code
3. C It refuses all numerical questions
4. D It cannot process numbers at all

15. An executive asks what "training" costs compared with "inference" for a business using a managed foundation model. Which statement is correct?

1. A Training and inference cost the same per request
2. B Inference is free once the model is selected
3. C Training the base model is borne by the provider; the business pays mainly for inference, and optionally for customization
4. D The business pays for base model training on every request

16. Which use case is best served by computer vision rather than natural language processing?

1. A Classifying the sentiment of customer reviews
2. B Detecting damaged packaging on a production line from camera images
3. C Summarizing a legal contract
4. D Routing support tickets by topic

17. What does a model's confidence score represent, and how should a business treat it?

1. A The financial value of the prediction
2. B A measure of how much data the model was trained on
3. C An indication of the model's certainty that should be calibrated and validated, not assumed to equal probability of correctness
4. D A guarantee that the prediction is correct at that percentage

18. Which describes an AI agent as distinct from a single model call?

1. A A model deployed on dedicated hardware
2. B A model with more parameters
3. C A model that has been fine-tuned
4. D A system where the model plans a sequence of steps and invokes tools or APIs to accomplish a goal

19. What is MLOps, in business terms?

1. A A job title for data scientists
2. B A type of machine learning algorithm
3. C A pricing model for AI services
4. D The practices and tooling that make machine learning systems reliably deployable, monitorable, and maintainable in production

20. Which statement about a foundation model's knowledge is accurate?

1. A It includes all publicly available information without exception
2. B It reflects the data available up to its training cutoff and does not include recent or private information unless supplied at inference
3. C It is continuously updated with current events
4. D It automatically learns from each conversation

21. A business wants to recommend products to customers based on the behaviour of similar customers. Which approach describes this?

1. A Optical character recognition
2. B Time series forecasting
3. C A recommendation system using collaborative filtering
4. D Anomaly detection

22. Which factor most determines whether a generative AI output can be used without human review?

1. A The consequence of an error in that specific context, weighed against measured output quality
2. B The speed at which the model responds
3. C The size of the model
4. D Whether the output is text or an image

23. Which describes the practical difference between interpretability and explainability?

1. A They mean the same thing
2. B Explainability requires access to the model's source code
3. C Interpretability applies only to generative models
4. D Interpretability is the degree to which a model's mechanism is inherently understandable; explainability is the ability to describe why a particular output occurred

24. Which capability distinguishes an AI agent from a single-response AI tool?

1. A It can plan a sequence of steps and use tools to act toward a goal
2. B It produces responses more quickly
3. C It is trained on the organization's proprietary data
4. D It uses a model with a larger context window

25. Why do deployed AI solutions require ongoing monitoring?

1. A Model licences expire and require periodic renewal
2. B Model performance degrades as the real-world data drifts away from the training data
3. C Models lose their training data over time
4. D Models consume more compute the longer they run

26. Employees are using unapproved AI tools with company data. Which governance measure addresses this risk?

1. A Publish a transparent classification of AI tools as approved, blocked, or under evaluation
2. B Ignore the practice while the tools remain free of charge
3. C Prohibit all use of AI tools across the organization
4. D Ask each employee to assess the tools they use

27. A business user reports that a generative AI assistant loses track of earlier parts of a long document. Which constraint explains this?

1. A The model has not been fine-tuned on the document
2. B The assistant is using an outdated model version
3. C The context window limits how much text the model can consider at once
4. D The model's temperature setting is too low

28. Which adaptation technique allows a generative AI system to answer from company documents that change weekly?

1. A Fine-tuning the model each week
2. B Increasing the model's temperature
3. C Retrieval Augmented Generation
4. D Continued pre-training on the document corpus

29. Which statement distinguishes artificial intelligence from machine learning?

1. A Artificial intelligence is a subset of machine learning
2. B Machine learning is a subset of artificial intelligence in which systems learn patterns from data rather than following explicit rules
3. C The two terms describe the same technology
4. D Machine learning applies only to generative systems

30. What does it mean for a model to generalize?

1. A It can be applied to any business problem without change
2. B It produces the same output for every input
3. C It requires no training data
4. D It performs well on data it has not seen before, not only on its training data

31. What is a parameter in the context of a machine learning model?

1. A A measure of the model's accuracy
2. B A setting an operator configures before training begins
3. C An input the model receives at prediction time
4. D A value the model learns during training that determines how it transforms input into output

32. Which statement describes training data?

1. A The examples a model learns from, whose quality and representativeness shape what the model can do
2. B The infrastructure on which training runs
3. C The configuration settings used during training
4. D The data a model produces as output

33. What does a confidence score accompanying a model's prediction indicate?

1. A The time the model took to produce the prediction
2. B The proportion of training data supporting the prediction
3. C A guarantee that the prediction is correct above the stated percentage
4. D The model's estimated likelihood that its prediction is correct, which may be poorly calibrated

34. A business problem requires predicting which of three categories a record belongs to. Which solution type applies?

1. A Clustering
2. B Classification
3. C Regression
4. D Anomaly detection

35. A business problem requires predicting a numeric value such as expected demand. Which solution type applies?

1. A Recommendation
2. B Regression
3. C Clustering
4. D Classification

36. An organization must decide whether to adopt an autonomous AI agent for a process. Which consideration is most important?

1. A Whether the consequences of an incorrect autonomous action are acceptable, and what human oversight the process requires
2. B How many tools the agent can call
3. C How quickly the agent responds
4. D Whether the agent uses the newest available model

37. A business process must be evaluated for AI suitability. Which characteristic indicates a poor fit?

1. A The process currently takes staff a long time
2. B The process requires an explanation of each decision that a complex model cannot provide, and no simpler interpretable approach reaches the required accuracy
3. C The process handles a high volume of transactions
4. D The process has years of historical records

38. What does a foundation model's pre-training provide?

1. A A model tuned for one specific business task
2. B A set of prompts for a particular application
3. C A retrieval index over the organization's documents
4. D Broad general capability learned from a very large corpus, which later adaptation refines for specific tasks

39. Which adaptation technique changes a foundation model's weights?

1. A Prompt engineering
2. B Retrieval Augmented Generation
3. C Fine-tuning
4. D Adjusting the temperature setting

40. Why does a generative AI system sometimes produce different answers to the same question?

1. A The model retrieves different documents each time by design
2. B Generation involves sampling, so the same prompt can yield different outputs unless sampling is constrained
3. C The model retrains itself between requests
4. D The model's parameters change with each use

41. Which characteristic indicates that a business problem suits a generative AI solution rather than a traditional model?

1. A The task produces varied natural language or content rather than a single categorical or numeric prediction
2. B The task has extensive labelled historical examples
3. C The task requires a highly explainable decision
4. D The task must produce an identical answer every time

42. An organization must decide whether a process should be automated with an AI agent or kept under human control. Which consideration is decisive?

1. A How many tools the agent would need
2. B Whether an incorrect autonomous action can be detected and reversed before it causes harm
3. C Whether the process is performed frequently
4. D Whether the agent can complete the task faster than a person

43. Which AI solution type predicts a category from labelled historical examples?

1. A Reinforcement learning
2. B Supervised classification
3. C Unsupervised clustering
4. D Generative modelling

44. Why does a generative AI system require different governance from a traditional predictive model?

1. A Its output is open-ended and cannot be exhaustively enumerated in advance, so controls must evaluate content rather than a fixed set of outcomes
2. B It runs on different infrastructure
3. C It produces results more quickly
4. D It requires more training data

45. Which concept describes supplying a generative AI system with an organization's own documents at query time?

1. A Model distillation
2. B Continued pre-training
3. C Retrieval Augmented Generation
4. D Fine-tuning