Artificial Intelligence MCQs – Set-1 | 50 questions on AI applications, generative systems, language models, and responsible AI.
Mixed difficulty · Select one answer and open each explanation to review the concept.
Which statement best describes artificial intelligence?
A Systems performing intelligent tasks
B A type of computer cable
C A database storage format
D A screen resolution standard
Answer: A. Systems performing intelligent tasks. Artificial intelligence broadly describes computational systems designed to perform tasks associated with intelligent behavior.
How is machine learning related to artificial intelligence?
A It replaces every AI system
B One approach within AI
C It only describes robotics hardware
D It is unrelated to AI
Answer: B. One approach within AI. Machine learning is one approach within the broader field of artificial intelligence.
What does a rule-based expert system rely on?
A Randomly selected answers
B A large image archive
C Encoded rules and domain knowledge
D A physical robot arm
Answer: C. Encoded rules and domain knowledge. Rule-based expert systems apply encoded domain knowledge and rules to reach conclusions.
What does an inference engine do in a rule-based system?
A Compresses the knowledge base
B Encrypts user information
C Stores network packets
D Applies rules to known facts
Answer: D. Applies rules to known facts. An inference engine applies rules to facts in a knowledge base to derive conclusions.
How can a search algorithm help an AI problem solver?
A Explore possible states or actions
B Increase a processor’s voltage
C Format a storage drive
D Translate code into HTML
Answer: A. Explore possible states or actions. Search algorithms explore possible states or action sequences to find a solution.
What does natural language processing focus on?
A Computer cooling
B Human language data
C Database backups
D Network routing
Answer: B. Human language data. Natural language processing enables systems to analyze or generate human language.
Which field helps software interpret images and video?
A Speech synthesis
B Cloud storage
C Computer vision
D File compression
Answer: C. Computer vision. Computer vision develops methods for extracting useful information from images and video.
What does automatic speech recognition produce?
A A printed barcode
B A network address
C A translated database
D Text from spoken audio
Answer: D. Text from spoken audio. Automatic speech recognition converts spoken audio into text or another representation.
What does text-to-speech technology generate?
A Spoken audio from text
B Text from scanned images
C A database index
D A network firewall rule
Answer: A. Spoken audio from text. Text-to-speech systems generate spoken audio from written text.
How can AI help a robot act in its environment?
A By replacing every sensor
B By planning from sensor inputs
C By changing the robot’s paint
D By creating storage partitions
Answer: B. By planning from sensor inputs. AI can use sensor information to support perception, planning, and action selection.
What does generative AI create?
A Only database constraints
B Only numeric measurements
C New text or images
D Only file backups
Answer: C. New text or images. Generative AI models create content such as text, images, audio, or code.
How does a large language model commonly generate text?
A By selecting a random webpage
B By retrieving one fixed answer
C By translating every prompt first
D By predicting token sequences
Answer: D. By predicting token sequences. Many language models generate text by predicting likely next tokens from prior context.
What is a token in language-model processing?
A A unit of text representation
B A user access password
C A cloud storage bucket
D A graphics processing core
Answer: A. A unit of text representation. A token is a unit used to represent text for model input and output.
What does tokenization do to input text?
A Encrypts it for storage
B Splits it into model tokens
C Converts it into a picture
D Deletes repeated words
Answer: B. Splits it into model tokens. Tokenization breaks text into tokens that a model can process.
What does a model’s context window limit?
A The number of model developers
B The number of cloud servers
C Tokens processed in one interaction
D Its training data age
Answer: C. Tokens processed in one interaction. A context window limits how many tokens the model can process in a given interaction.
What is a prompt in generative AI?
A A type of processor
B A database schema
C A model evaluation score
D Input instructions and context
Answer: D. Input instructions and context. A prompt supplies instructions, context, or data that guide the model’s response.
What does zero-shot prompting provide?
A A task request without examples
B Several labeled examples
C A full model retraining
D Only a database query
Answer: A. A task request without examples. Zero-shot prompting gives a task request without providing examples of the desired output.
How does few-shot prompting guide a model?
A By disabling its context window
B By including a few examples
C By removing all instructions
D By changing its hardware
Answer: B. By including a few examples. Few-shot prompts provide a small number of examples to illustrate the requested pattern.
Why should a generated answer be checked against reliable evidence?
A Models always cite sources
B Prompts guarantee factual output
C Models can produce plausible errors
D Generated text is always copied
Answer: C. Models can produce plausible errors. Generative systems can produce fluent but inaccurate or unsupported responses.
What does grounding an AI response aim to provide?
A A faster processor
B A larger screen
C A different model name
D Relevant evidence for its answer
Answer: D. Relevant evidence for its answer. Grounding connects an answer to relevant evidence or trusted information sources.
What is retrieval-augmented generation (RAG)?
A Retrieve relevant information, then generate
B Retrain after every prompt
C Replace it with a database
D Compress prompts into fewer tokens
Answer: A. Retrieve relevant information, then generate. RAG retrieves relevant external information and supplies it as context for generation.
How does fine-tuning differ from retrieval augmentation?
A It only changes the prompt
B Adjusts model parameters using examples
C It searches a document store
D It converts text into audio
Answer: B. Adjusts model parameters using examples. Fine-tuning updates model parameters using additional training examples, unlike retrieving context at inference time.
What are embeddings commonly used to represent?
A Physical server locations
B Password expiration dates
C Data as numerical vectors
D Network cable lengths
Answer: C. Data as numerical vectors. Embeddings represent items such as text as numerical vectors that capture useful relationships.
What does a vector database help find?
A Only exact filenames
B Only encrypted passwords
C New processor instructions
D Similar vectors or embedded content
Answer: D. Similar vectors or embedded content. Vector databases store embeddings and support similarity searches for related items.
What does attention help a transformer model do?
A Weight relevant parts of context
B Encrypt a database backup
C Assign IP addresses
D Compress image files
Answer: A. Weight relevant parts of context. Attention mechanisms let a model weigh relationships among elements in its input context.
Which architecture was introduced in the paper “Attention Is All You Need”?
A Decision tree
B Transformer
C Relational database
D Optical network
Answer: B. Transformer. The 2017 paper introduced the Transformer architecture based on attention mechanisms.
What is a multimodal AI model designed to process?
A Only spreadsheet formulas
B Only one language
C More than one data modality
D Only numeric IDs
Answer: C. More than one data modality. Multimodal systems can process or generate multiple modalities such as text, images, and audio.
What does image classification assign to an image?
A A network address
B A storage location
C A spoken transcript
D A category or label
Answer: D. A category or label. Image classification assigns one or more labels to an image.
How does object detection differ from basic image classification?
A Locates objects in images
B It only translates captions
C It stores image backups
D It changes pixels into sound
Answer: A. Locates objects in images. Object detection identifies objects and their locations within an image.
What does machine translation convert?
A Images into pixels
B Text between languages
C Passwords into user names
D Audio into video frames
Answer: B. Text between languages. Machine translation converts text or speech between languages.
How can a recommendation system use AI?
A Measure computer temperature
B Encrypt user accounts
C Suggest items from user signals
D Format web pages
Answer: C. Suggest items from user signals. Recommendation systems use signals such as user preferences or past activity to suggest relevant items.
What does narrow AI describe?
A Broad human-level AI
B Program that never uses data
C Any device with a processor
D AI built for specific tasks
Answer: D. AI built for specific tasks. Narrow AI is designed for a defined task or limited set of tasks.
What does artificial general intelligence (AGI) refer to?
A A hypothetical broadly capable AI
B A single-purpose calculator
C A type of network switch
D A specific image file
Answer: A. A hypothetical broadly capable AI. AGI is a proposed form of AI with broad capabilities across many tasks; it is not a settled description of today’s systems.
What does AI explainability seek to provide?
A A longer training dataset
B Reasons or evidence for outputs
C A faster network connection
D A smaller storage device
Answer: B. Reasons or evidence for outputs. Explainability helps people understand the reasons or evidence behind an AI system’s outputs.
What does AI interpretability focus on?
A Changing the system’s user name
B Increasing training costs
C Understanding how system behavior works
D Compressing model outputs
Answer: C. Understanding how system behavior works. Interpretability concerns how well people can understand a system’s internal behavior or decision process.
Why is transparency important in AI use?
A It guarantees perfect accuracy
B It removes all system risk
C It prevents model updates
D Understand system use and limits
Answer: D. Understand system use and limits. Transparency helps people understand when and how an AI system is used and what its limitations are.
What does fairness evaluation examine?
A Unequal group-level effects
B Processor speed
C File compression
D Model storage size
Answer: A. Unequal group-level effects. Fairness evaluation looks for unjustified disparities or harmful outcomes across people or groups.
How can biased training data affect an AI system?
A It guarantees broader representation
B It can reproduce harmful patterns
C Testing is unnecessary
D It ensures perfect fairness
Answer: B. It can reproduce harmful patterns. Biased or unrepresentative data can contribute to harmful or unfair model behavior.
Which property concerns consistent performance under expected conditions?
A Color balance
B File portability
C Reliability
D Data compression
Answer: C. Reliability. Reliability concerns whether a system performs its intended function consistently.
What does robustness describe in AI?
A A system’s brand name
B A prompt’s word count
C The model’s training date
D Performance under varied conditions
Answer: D. Performance under varied conditions. Robustness is a system’s ability to maintain suitable performance under varied or challenging conditions.
What does privacy-enhanced AI aim to protect?
A Personal and sensitive information
B Image resolution
C Processor temperature
D File names
Answer: A. Personal and sensitive information. Privacy-enhanced approaches seek to protect personal information used or produced by AI systems.
Why keep human oversight for high-impact AI decisions?
A To let models avoid evaluation
B To retain review and accountability
C To eliminate system documentation
D Guarantee every answer
Answer: B. To retain review and accountability. Human oversight helps ensure review, accountability, and appropriate action when AI affects consequential decisions.
Which functions are in the NIST AI RMF Core?
A Train, prompt, deploy, sell
B Collect, label, publish, archive
C Govern, map, measure, manage
D Plan, build, test, ship
Answer: C. Govern, map, measure, manage. The NIST AI RMF Core organizes risk activities under Govern, Map, Measure, and Manage.
When should AI performance and risks be evaluated?
A Only after a public failure
B Only once during training
C Only when a user complains
D Before deployment and during use
Answer: D. Before deployment and during use. NIST guidance calls for testing before deployment and regular monitoring while systems operate.
What does human-in-the-loop design keep?
A People review selected decisions
B Every task fully automated
C Training data permanently hidden
D Model outputs unreviewed
Answer: A. People review selected decisions. Human-in-the-loop approaches include people in specified review or decision steps.
Why conduct red-team exercises on AI systems?
A To increase token counts
B Probe weaknesses with realistic tests
C Train users to ignore alerts
D To avoid evaluating model behavior
Answer: B. Probe weaknesses with realistic tests. Red teaming probes systems for weaknesses, including unsafe behavior under adversarial or unusual inputs.
What does an AI governance process establish?
A A larger model architecture
B A universal prompt template
C Roles, policies, and oversight
D A faster training algorithm
Answer: C. Roles, policies, and oversight. Governance establishes organizational roles, policies, and oversight for AI development and use.
How should an AI system’s intended use be defined?
A After it causes harm
B Only by the model vendor
C From training accuracy alone
D Users and intended context
Answer: D. Users and intended context. Intended use should be defined in context because the same system may create different risks in different settings.
What is a model card commonly used to document?
A Model use and limitations
B A user’s password history
C A server’s physical address
D A database’s primary key
Answer: A. Model use and limitations. A model card summarizes information such as a model’s intended uses, evaluation, and limitations.
What should an organization do before relying on generated material for important decisions?
A Treat fluent text as proof
B Review output against trusted sources
C Assume it has no limitations
D Disable human review
Answer: B. Review output against trusted sources. Generated material should be checked against trusted evidence, especially before consequential use.