Unit 1 - Practice Quiz

INT428 60 Questions
0 Correct 0 Wrong 60 Left
0/60

1 What is the primary goal of Artificial Intelligence (AI)?

what is AI Easy
A. To design visually appealing user interfaces for software.
B. To create machines that can perform tasks that typically require human intelligence.
C. To increase the storage capacity of digital devices.
D. To build faster and more efficient computer hardware.

2 An AI system that is designed to perform one specific task, such as playing chess or identifying faces, is known as:

Evolution, and types of AI (narrow, general) Easy
A. Sentient AI
B. Artificial General Intelligence (AGI)
C. Narrow AI (or Weak AI)
D. Artificial Superintelligence (ASI)

3 Which of the following is a common application of AI in the healthcare domain?

Applications across domains (business, healthcare, automation, vision, language) Easy
A. Keeping track of employee work schedules.
B. Manufacturing surgical tools.
C. Designing hospital building layouts.
D. Analyzing medical images like X-rays to help detect diseases.

4 What are TensorFlow and PyTorch?

Modern AI Toolkits (TensorFlow, PyTorch) Easy
A. Popular open-source software libraries for machine learning.
B. Cloud storage services.
C. Types of computer processors.
D. Web browsers for AI research.

5 Which concept in Responsible AI deals with ensuring that an AI system's decisions are understandable and explainable to humans?

Responsible AI Easy
A. Scalability
B. Transparency (or Explainability)
C. Performance
D. Availability

6 In the context of an AI problem like solving a puzzle, what does a 'state' in a 'state space' represent?

AI Problem Modeling & Search Concepts: Defining AI problems as State Space and Search Problems Easy
A. The person who is solving the puzzle.
B. A specific configuration or arrangement of the puzzle at a given moment.
C. The time it takes to solve the puzzle.
D. The final solution to the puzzle.

7 The hypothetical concept of an AI that could understand, learn, and apply knowledge across a wide range of tasks at a human level is called:

Evolution, and types of AI (narrow, general) Easy
A. Artificial General Intelligence (AGI)
B. Limited Memory AI
C. Narrow AI
D. Reactive AI

8 Which of the following fields is considered a core foundation for the development of Artificial Intelligence?

Foundations of AI Easy
A. Computer Science
B. Geology
C. World History
D. Marine Biology

9 An AI system that can understand and respond to human language, like a chatbot or a voice assistant, is an application of which AI subfield?

Applications across domains (business, healthcare, automation, vision, language) Easy
A. Robotics
B. Computer Vision
C. Natural Language Processing (NLP)
D. Expert Systems

10 What is typically the first step in a standard AI/machine learning workflow?

Introduction to AI Workflows & Data-Centric Modeling Easy
A. Gathering and preparing the data.
B. Training the algorithm.
C. Evaluating the model's accuracy.
D. Deploying the model to production.

11 What is a major challenge in developing AI systems, especially those based on deep learning?

Challenges in AI problem solving Easy
A. They run too quickly.
B. They require very little computer memory.
C. They often require a very large amount of high-quality, labeled data.
D. They are too easy for anyone to build.

12 The task of automatically grouping similar items together from an unlabeled dataset is known as:

Key AI problems and techniques Easy
A. Clustering
B. Classification
C. Regression
D. Reinforcement Learning

13 Which of the following abilities is a key component of intelligence, both human and artificial?

What is Intelligence Easy
A. The ability to lift heavy objects.
B. The ability to hold one's breath.
C. The ability to change color.
D. The ability to learn from experience.

14 What is the primary focus of a 'data-centric' approach to building AI systems?

Data-Centric Modeling Easy
A. Using the most powerful hardware available.
B. Constantly changing the model's algorithm.
C. Systematically improving the quality of the dataset.
D. Writing the code in multiple programming languages.

15 In a simple search problem like finding the best route between two cities on a map, the 'problem space' consists of:

Characteristics of AI Problem Spaces Easy
A. Only the start city and the end city.
B. The type of car you are driving.
C. All the possible routes and cities that can be visited.
D. The weather conditions.

16 In AI, what does the term 'bias' often refer to?

Challenges in AI problem solving Easy
A. A systematic error where the model produces unfair or prejudiced outcomes.
B. The model's ability to make fair decisions.
C. The physical weight of the computer running the model.
D. A type of computer virus.

17 A key characteristic of many modern AI systems is their ability to improve their performance on a task over time without being explicitly reprogrammed. This is known as:

Characteristics of artificial intelligence Easy
A. Hard-coding
B. Learning
C. Static programming
D. Rebooting

18 What is an example of AI being used in business for automation?

Applications across domains (business, healthcare, automation, vision, language) Easy
A. Using AI to sort and respond to customer support emails automatically.
B. Designing a new company logo by hand.
C. Hiring more employees for manual data entry.
D. Manually creating financial reports in a spreadsheet.

19 The primary function of a machine learning library like PyTorch is to:

Modern AI Toolkits (TensorFlow, PyTorch) Easy
A. Design the physical circuits for an AI chip.
B. Store large video files.
C. Provide a text editor for writing code.
D. Offer pre-built tools and functions to simplify the process of creating AI models.

20 What is the principle of 'Fairness' in Responsible AI?

Responsible AI Easy
A. Ensuring the AI model does not produce systematically biased or discriminatory outcomes against certain groups.
B. Ensuring the AI model is profitable for the company.
C. Ensuring the AI model is written in a popular programming language.
D. Ensuring the AI model runs as fast as possible.

21 A GPS navigation system needs to find the shortest route between two locations in a city. In a state space search model for this problem, what would be the most accurate representation of a state and an action?

AI Problem Modeling & Search Concepts: Defining AI problems as State Space and Search Problems Medium
A. State: A specific vehicle. Action: The speed of the vehicle.
B. State: The destination city. Action: Calculating the total distance.
C. State: The list of all possible routes. Action: Choosing the route with the fewest turns.
D. State: The current geographical intersection or junction. Action: Driving along a road segment to the next intersection.

22 An AI model used for loan approvals is trained on historical data. If the historical data contains biases where a certain demographic was unfairly denied loans, the resulting AI model is likely to perpetuate this bias. This issue is primarily a failure of:

Responsible AI Medium
A. Data privacy
B. Model scalability
C. Algorithmic efficiency
D. Fairness and equity

23 A company develops an AI system that is exceptionally good at translating languages but cannot perform any other task, such as identifying objects in images or composing music. This system is a prime example of:

Evolution, and types of AI (narrow, general) Medium
A. A Theory of Mind AI
B. Artificial Superintelligence (ASI)
C. Artificial General Intelligence (AGI)
D. Artificial Narrow Intelligence (ANI)

24 An AI team has a reasonably good model for detecting manufacturing defects but wants to improve its performance. Instead of focusing on hyperparameter tuning or trying a more complex architecture, the team decides to spend its effort on acquiring more diverse and accurately labeled images of defects. This approach is best described as:

Introduction to AI Workflows & Data-Centric Modeling Medium
A. Model-centric AI
B. Data-centric AI
C. Algorithm-centric AI
D. Hardware-centric AI

25 The Turing Test, proposed by Alan Turing, is a test of a machine's ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human. This test is most closely aligned with which definition of AI?

Foundations of AI Medium
A. Acting humanly
B. Thinking rationally
C. Acting rationally
D. Thinking humanly

26 A researcher is developing a novel neural network architecture and needs to frequently debug and inspect the gradients at each step of the training process. They prefer a framework that builds the computation graph as the code is executed. Which toolkit's core design philosophy is most suitable for this 'define-by-run' approach?

Modern AI Toolkits (TensorFlow, PyTorch) Medium
A. Scikit-learn
B. PyTorch
C. Apache Spark MLlib
D. Early versions of TensorFlow (using static graphs)

27 In the context of solving a Rubik's Cube, the problem space is characterized by a massive number of possible states (over ) but a small, well-defined set of actions (turning a face). This immense growth in the number of states is a classic example of:

Characteristics of AI Problem Spaces Medium
A. A non-deterministic problem
B. A partially observable environment
C. A continuous action space
D. Combinatorial explosion

28 An email service provider wants to build a system that automatically sorts incoming emails into categories like 'Primary,' 'Social,' 'Promotions,' and 'Spam.' This task is best framed as what type of machine learning problem?

Key AI problems and techniques Medium
A. Clustering
B. Reinforcement learning
C. Regression
D. Multi-class classification

29 A hospital uses an AI system to analyze Magnetic Resonance Imaging (MRI) scans to detect and segment tumors. This application is a primary example of AI in the domain of:

Applications across domains (business, healthcare, automation, vision, language) Medium
A. Natural Language Processing (NLP)
B. Predictive Analytics for patient readmission
C. Computer Vision
D. Robotic Process Automation (RPA)

30 An AI agent is designed to play the card game Poker. A key challenge for the agent is that it does not know the cards held by its opponents. This lack of complete information about the game state makes the problem environment:

Challenges in AI problem solving Medium
A. Partially observable
B. Deterministic
C. Fully observable
D. Static

31 When modeling the 8-puzzle problem for an AI search algorithm, which of the following would be the most suitable heuristic function () for an A* search?

AI Problem Modeling & Search Concepts: Defining AI problems as State Space and Search Problems Medium
A. The number of steps taken so far.
B. The number of tiles in the correct row.
C. The Manhattan distance, which is the sum of the distances of each tile from its goal position.
D. A constant value of 1 for every state.

32 A company deploys a complex deep learning model for medical diagnosis. A doctor questions a specific diagnosis, but the developers cannot provide a clear reason for the output, only that it is what the model learned. This situation highlights a critical lack of:

Responsible AI Medium
A. Explainability (or Interpretability)
B. Data privacy
C. System security
D. Computational robustness

33 The development of Bayesian networks, which allow AI systems to reason with uncertainty, is most directly built upon which foundational field?

Foundations of AI Medium
A. Control Theory
B. Computer Engineering
C. Linguistics
D. Mathematics (Probability Theory)

34 If an AI system could not only write a compelling novel but also understand the emotional impact of its writing on a human and discuss its literary themes with a critic, it would be demonstrating capabilities associated with:

Evolution, and types of AI (narrow, general) Medium
A. Artificial General Intelligence (AGI)
B. A standard Reactive Machine
C. A supervised learning model
D. A purely symbolic AI system

35 In a standard supervised machine learning workflow, what is the specific purpose of holding out a test dataset?

Introduction to AI Workflows & Data-Centric Modeling Medium
A. To tune the model's hyperparameters like learning rate or tree depth.
B. To be used for data augmentation and creating synthetic samples.
C. To train the model's primary parameters.
D. To provide a final, unbiased evaluation of the trained model's performance on unseen data.

36 An AI-powered thermostat learns your daily routines and automatically adjusts the temperature for comfort and energy efficiency, adapting over time as your schedule changes. Which key characteristic of AI is most prominently demonstrated?

characteristics of artificial intelligence Medium
A. Complete logical deduction
B. Static knowledge representation
C. Symbolic reasoning
D. Learning and adaptation

37 An AI system analyzes customer service chat logs to automatically categorize complaints into topics like 'Billing Issue,' 'Technical Support,' or 'Product Inquiry.' This is a direct application of which specific AI technique?

Applications across domains (business, healthcare, automation, vision, language) Medium
A. Computer Vision for image segmentation
B. Time-series forecasting
C. Natural Language Processing (NLP) for text classification
D. Reinforcement Learning for robotic control

38 Consider a self-driving car navigating a busy city street. The environment is constantly changing due to other cars and pedestrians, and the car's sensors may have noise or be occluded. This environment is best described as:

Characteristics of AI Problem Spaces Medium
A. Dynamic, continuous, and partially observable.
B. Static, continuous, and deterministic.
C. Dynamic, discrete, and fully observable.
D. Static, discrete, and fully observable.

39 According to the 'Acting Rationally' (Rational Agent) definition of AI, the primary measure of an agent's success is:

What is Intelligence, what is AI Medium
A. Its ability to perform complex symbolic computations.
B. Whether its behavior is indistinguishable from a human's in a conversation.
C. Its ability to achieve the best expected outcome based on its knowledge and the situation.
D. How closely its internal thought processes mimic human cognition.

40 A financial firm wants to build an AI system to analyze historical stock prices and predict the price for the next day. This problem is best framed as a:

Key AI problems and techniques Medium
A. Anomaly detection problem
B. Regression or Time-Series Forecasting problem
C. Clustering problem
D. Classification problem

41 John Searle's Chinese Room argument primarily targets which specific claim about artificial intelligence?

Foundations of AI Hard
A. The claim that a machine, by virtue of implementing a formal program, can have understanding or consciousness (Strong AI).
B. The claim that machines can process information faster than humans.
C. The claim that machines can pass the Turing Test by deceiving a human interrogator.
D. The claim that AI can be a useful tool for studying human cognition (Weak AI).

42 Consider a modified 8-puzzle problem on a 3x3 grid, but with an additional action: any tile adjacent to the blank space can be 'zapped' (removed from the board) at a high cost. A goal state is any configuration with tiles 1-8 in their correct positions, regardless of whether other tiles have been zapped. How does this modification fundamentally alter the state space search compared to the classic 8-puzzle?

AI Problem Modeling & Search Concepts: Defining AI problems as State Space and Search Problems Hard
A. The state space graph becomes a tree instead of a general graph, as the 'zap' action is irreversible.
B. The state space becomes infinite, as zapping can be done repeatedly on newly adjacent tiles.
C. The state space remains finite, but the problem is no longer solvable with heuristic searches like A* because the heuristic becomes inconsistent.
D. The state space remains finite, but the graph is no longer undirected (it becomes a directed graph), and the branching factor becomes variable.

43 An AI model for loan approvals shows 95% accuracy across all demographics. However, it is found that for a protected minority group, the False Rejection Rate (FRR) is 30%, while for the majority group, it is 5%. This situation best exemplifies a conflict between which two core principles of Responsible AI?

Responsible AI Hard
A. Privacy and Security
B. Accountability and Reliability
C. Overall Accuracy and Fairness (Equality of Opportunity)
D. Transparency and Robustness

44 A researcher is building a novel recurrent neural network where the computational graph's structure changes at each time step based on the input data itself (e.g., activating different sub-networks). Why would PyTorch or TensorFlow 2.x (in eager mode) be fundamentally more suitable for this task than TensorFlow 1.x (with static graphs)?

Modern AI Toolkits (TensorFlow, PyTorch) Hard
A. TensorFlow 1.x requires specialized hardware like TPUs which are not suitable for dynamic computations.
B. PyTorch and TF 2.x have a larger community and more pre-trained models for recurrent architectures.
C. TensorFlow 1.x's 'define-then-run' paradigm requires a fixed, pre-compiled graph, making it extremely cumbersome to handle data-dependent graph structures.
D. PyTorch and TF 2.x offer better visualization tools like TensorBoard for debugging dynamic graphs.

45 A team has developed a highly complex image classification model that performs poorly on images taken in foggy weather, a rare condition in their initial dataset. The team's resources for new data acquisition are limited. According to data-centric AI principles, which of the following strategies is the most direct and efficient first step to address this specific problem?

Introduction to AI Workflows & Data-Centric Modeling Hard
A. Implement a targeted data augmentation strategy by applying a synthetic fog effect to a significant portion of the existing training images.
B. Switch to a more complex model architecture like a Vision Transformer, as it may generalize better to out-of-distribution data.
C. Perform extensive hyperparameter tuning on the existing model, focusing on regularization techniques like dropout to improve robustness.
D. Increase the overall size of the training dataset by collecting 10% more general images, hoping some will include foggy conditions.

46 In the context of AI problem spaces, which characteristic of a problem would most strongly suggest that a local search algorithm (like Hill Climbing) is highly likely to fail to find the optimal solution?

Characteristics of AI Problem Spaces Hard
A. A state space where actions are irreversible (directed graph).
B. A problem that is fully observable and deterministic.
C. The presence of numerous local maxima and a 'plateau' where many states have the same heuristic value.
D. A very large or infinite branching factor.

47 Considering the transition from Artificial Narrow Intelligence (ANI) to a hypothetical Artificial General Intelligence (AGI), which conceptual leap represents the most significant and currently unsolved challenge?

Evolution, and types of AI (narrow, general) Hard
A. Achieving superhuman performance in a wider variety of specific, isolated tasks such as chess, Go, and protein folding.
B. Developing the ability for robust 'transfer learning' where knowledge gained in one domain (e.g., playing a video game) can be abstractly applied to a completely different domain (e.g., financial planning).
C. Improving the accuracy of natural language processing models to achieve near-perfect translation and summarization.
D. Scaling up computational power and memory to match the human brain's capacity.

48 A self-driving car's perception system misidentifies a large, white truck against a bright sky as being part of the sky, leading to a collision. This is a real-world example of failure due to a combination of which two fundamental AI challenges?

Challenges in AI problem solving Hard
A. The Qualification Problem and Moravec's Paradox.
B. The Frame Problem and the Symbol Grounding Problem.
C. Lack of Common Sense Knowledge and Brittleness to Out-of-Distribution Data.
D. Combinatorial Explosion and the Halting Problem.

49 A hospital deploys an AI to predict patient readmission risk. To ensure transparency, they use a simple, interpretable model (e.g., logistic regression). However, a more complex 'black box' model (e.g., a deep neural network) is shown to be 15% more accurate. This creates a direct tension between which two ethical AI principles?

Responsible AI Hard
A. Accountability and Security
B. Fairness and Privacy
C. Interpretability and Beneficence (Utility)
D. Robustness and Reliability

50 The philosophical position of 'functionalism' is a key foundation for AI. It posits that mental states are constituted by their causal relations to other mental states, sensory inputs, and behavioral outputs. How does this view directly support the possibility of creating artificial general intelligence?

Foundations of AI Hard
A. It prioritizes emotional intelligence over logical reasoning as the cornerstone of AGI.
B. It suggests that consciousness is an illusion and therefore irrelevant to creating AI.
C. It proves that intelligence can only arise from biological carbon-based structures.
D. It implies that if a machine can replicate the functional role of a mental state, it possesses that mental state, regardless of its physical substrate (e.g., silicon vs. neurons).

51 Reinforcement Learning (RL) is often modeled as a Markov Decision Process (MDP). What is the critical implication of the 'Markov Property' for an RL agent's decision-making process?

Key AI problems and techniques Hard
A. The reward function must be deterministic and cannot have stochastic components.
B. The agent must have a complete and perfect model of the environment's dynamics to make any decision.
C. The future state depends only on the current state and the chosen action, not on the sequence of states that preceded it ().
D. The effects of an action taken in a state depend on the entire prior history of states and actions.

52 When modeling a problem like vehicle routing (a variant of the Traveling Salesperson Problem) as a state-space search, what is the most significant challenge that makes simple uninformed search algorithms like Breadth-First Search (BFS) computationally infeasible?

AI Problem Modeling & Search Concepts: Defining AI problems as State Space and Search Problems Hard
A. The partial observability of the problem, where the agent doesn't know the location of all cities at once.
B. The difficulty in defining a goal state, as multiple routes can have the same minimal cost.
C. The combinatorial explosion of the state space, which grows factorially () with the number of cities.
D. The non-deterministic nature of the environment, where travel times can change unexpectedly.

53 In medical imaging AI, a convolutional neural network (CNN) is trained to detect tumors. The model achieves high accuracy but is later found to be focusing on artifacts introduced by a specific type of X-ray machine used for most of the positive cancer cases in the training data. This is a subtle example of what specific AI pitfall?

Applications across domains (business, healthcare, automation, vision, language) Hard
A. The vanishing gradient problem preventing deeper layers from learning meaningful features.
B. The model learning a 'shortcut' or a spurious correlation instead of the actual underlying pathology.
C. Overfitting to the training data's noise.
D. Catastrophic forgetting, where the model forgets previous knowledge when trained on new data.

54 What is the primary motivation for adopting a 'Data-Centric' modeling approach over a 'Model-Centric' approach, especially in mature AI projects where performance has plateaued?

Introduction to AI Workflows & Data-Centric Modeling Hard
A. Newer deep learning models are too complex to tune effectively, so focusing on data is the only remaining option.
B. In many real-world systems, the quality and consistency of the data become the biggest lever for improvement after initial model architecture has been optimized.
C. Model-centric approaches require more computational power, which is often a bottleneck.
D. Data-centric approaches allow for the use of simpler, more interpretable models.

55 The 'Frame Problem' in classical symbolic AI is notoriously difficult. In essence, what is the core challenge it describes?

Challenges in AI problem solving Hard
A. The difficulty of representing all the qualifications needed for an action to be successful.
B. The challenge of representing what remains unchanged in the world after an agent performs an action, without having to explicitly state every single non-effect.
C. The problem of grounding abstract symbols (like 'chair') to real-world perceptual data.
D. The computational intractability of planning in a large state space.

56 Which of the following scenarios best illustrates the characteristic of 'autonomy' in an AI system, as distinct from mere 'automation'?

What is Intelligence, what is AI, characteristics of artificial intelligence Hard
A. An industrial robot on an assembly line that welds car parts in the exact same predefined locations every time.
B. A chatbot that provides scripted answers to frequently asked questions based on keyword matching.
C. A software script that automatically runs at midnight to back up a database.
D. A Mars rover that, after losing communication with Earth, independently navigates around a newly-formed crater to reach its next waypoint.

57 A company uses an AI to screen resumes and observes that it disproportionately rejects female candidates for a software engineering role. Upon investigation, they find the AI learned a spurious correlation between being named 'Jared' and being a successful engineer, as 'Jared' appeared frequently in the training data of successful hires. This is a direct example of which type of bias?

Responsible AI Hard
A. Selection Bias
B. Historical Bias
C. Latent Bias
D. Interaction Bias

58 When performing distributed training of a large neural network, what is the fundamental difference between 'data parallelism' and 'model parallelism'?

Modern AI Toolkits (TensorFlow, PyTorch) Hard
A. Data parallelism involves training different models on the same data, while model parallelism trains one model on different data.
B. Data parallelism is only supported in PyTorch, while model parallelism is exclusive to TensorFlow.
C. Data parallelism replicates the entire model on multiple devices, each processing a different batch of data, while model parallelism splits a single large model across multiple devices.
D. In data parallelism, gradients are synchronized after each epoch, whereas in model parallelism, they are synchronized after each batch.

59 The concept of the 'AI Winter' refers to periods of reduced funding and interest in AI research. A primary cause of the first AI Winter in the mid-1970s was the failure of early AI systems to overcome what specific, fundamental problem?

Evolution, and types of AI (narrow, general) Hard
A. The failure of perceptrons to solve non-linearly separable problems like XOR, as highlighted by Minsky and Papert.
B. The inability to process natural language, as demonstrated by the limitations of early machine translation.
C. The extreme computational cost of running neural networks on the hardware of the time.
D. The inability of early symbolic AI systems to handle the combinatorial explosion and common-sense reasoning required for real-world problems outside of limited 'microworlds'.

60 The 'Turing Test' is often cited as a benchmark for AI. However, a significant philosophical criticism of the test, as a definitive measure of intelligence, is that it is primarily a test of:

What is Intelligence, what is AI, characteristics of artificial intelligence Hard
A. emotional intelligence and empathy.
B. mathematical and logical reasoning ability.
C. the ability to successfully deceive a human through linguistic manipulation.
D. computational efficiency and speed.