14 Sep 100 Agentic AI MCQ (Multiple Choice Questions) with Answers

Posted at 21:43h in mcq by

1) What is the primary characteristic that distinguishes an Agentic AI from a standard generative AI model?
  1. The ability to generate more creative text.
  2. The ability to take autonomous actions to achieve a goal.
  3. The ability to process larger amounts of data.
  4. The ability to run on less powerful hardware.
Show Answer
Answer: b) The ability to take autonomous actions to achieve a goal.
Explanation
A standard generative model produces content in response to a prompt. An agentic system goes further: it autonomously plans, decides and executes actions in pursuit of a goal, often using tools and memory across multiple steps.

2) Which of the following is NOT a core component of a typical AI agent?

  1. Perception
  2. Reasoning/Planning
  3. Action/Execution
  4. A static, pre-programmed database
Show Answer
Answer: d) A static, pre-programmed database
Explanation
Typical agents are built around perception, reasoning/planning, action/execution and memory. A static, pre-programmed database is not a core component — and on its own it would prevent the agent from adapting to new situations.

3) The concept of an “agent” in AI is derived from which field?

  1. Neuroscience
  2. Economics and Game Theory
  3. Linguistics
  4. Hardware Engineering
Show Answer
Answer: b) Economics and Game Theory
Explanation
The AI notion of an agent comes from economics and game theory, where a “rational agent” is modelled as an entity that selects actions to maximise its expected utility given its environment and goals.

4) What is the role of a “tool” in an agentic AI system?

  1. To decorate the user interface.
  2. To provide the agent with the ability to interact with the outside world (e.g., APIs, databases).
  3. To slow down the agent’s reasoning process.
  4. To store the agent’s final answer.
Show Answer
Answer: b) To provide the agent with the ability to interact with the outside world (e.g., APIs, databases).
Explanation
Tools extend an agent beyond its training data. Through APIs, databases, search engines, calculators and code interpreters, the agent can fetch real information and take real actions in the world.

5) Which of the following is a common architectural pattern for Agentic AI?

  1. Monolithic Transformer
  2. ReAct (Reason + Act)
  3. Convolutional Neural Network (CNN)
  4. Generative Adversarial Network (GAN)
Show Answer
Answer: b) ReAct (Reason + Act)
Explanation
ReAct is a widely used agentic pattern that interleaves reasoning steps with actions and observations. The other options are neural network architectures, not agent control patterns.

6) In the context of Agentic AI, what does the “ReAct” framework stand for?

  1. React, Analyze, Correct
  2. Reason, Act, Observe
  3. Retrieve, Act, Confirm
  4. Review, Assess, Conclude
Show Answer
Answer: b) Reason, Act, Observe
Explanation
ReAct combines reasoning traces with actions: the agent reasons about what to do, acts (often by calling a tool), observes the result, and then loops back to reason again with that new information.

7) What is the primary function of a “planner” in an AI agent?

  1. To execute a single, pre-defined action.
  2. To break down a high-level goal into a sequence of smaller, actionable steps.
  3. To randomly select actions until one works.
  4. To generate a final report.
Show Answer
Answer: b) To break down a high-level goal into a sequence of smaller, actionable steps.
Explanation
The planner performs task decomposition, turning an abstract objective into an ordered set of concrete sub-tasks that the agent (or its sub-agents and tools) can execute.

8) Which type of memory in an agentic system is analogous to a human’s short-term memory?

  1. Long-term memory (LTM)
  2. Episodic memory
  3. Contextual memory
  4. Procedural memory
Show Answer
Answer: c) Contextual memory
Explanation
Contextual (working) memory holds the information currently in play — the ongoing conversation, task state and recent observations — much like human short-term memory. Long-term and episodic memory persist far beyond the current task.

9) What is “Chain of Thought” (CoT) prompting primarily used for in agentic systems?

  1. To reduce the cost of API calls.
  2. To encourage the model to break down a problem into intermediate reasoning steps.
  3. To generate more creative images.
  4. To translate text between languages.
Show Answer
Answer: b) To encourage the model to break down a problem into intermediate reasoning steps.
Explanation
Chain of Thought prompting asks the model to show intermediate reasoning before answering, which markedly improves performance on multi-step arithmetic, logic and planning problems.

10) What is the key difference between a “reflex agent” and a “goal-based agent”?

  1. A reflex agent is more intelligent.
  2. A goal-based agent considers the future consequences of its actions.
  3. A reflex agent uses a complex world model.
  4. A goal-based agent cannot learn from experience.
Show Answer
Answer: b) A goal-based agent considers the future consequences of its actions.
Explanation
A reflex agent maps a stimulus directly to an action using condition–action rules. A goal-based agent additionally maintains a goal and a model of the world, so it can evaluate how candidate actions affect future states.

11) Which of the following is a popular framework for building agentic AI systems?

  1. TensorFlow
  2. PyTorch
  3. LangChain
  4. Scikit-learn
Show Answer
Answer: c) LangChain
Explanation
LangChain provides high-level abstractions for agents, tools, memory and execution loops. TensorFlow and PyTorch are deep learning frameworks and Scikit-learn is a classical ML library.

12) What is “tool use” or “function calling” in the context of Large Language Models (LLMs)?

  1. The ability of an LLM to write its own code.
  2. The ability of an LLM to request the execution of an external function or API.
  3. The ability of an LLM to use a physical tool like a robot arm.
  4. The ability of an LLM to explain how a tool works.
Show Answer
Answer: b) The ability of an LLM to request the execution of an external function or API.
Explanation
With function calling, the model emits a structured request naming a function and its arguments. The surrounding system runs the function and feeds the result back into the model’s context.

13) What is a “multi-agent system”?

  1. A system with multiple users.
  2. A system where multiple AI agents interact and collaborate to solve a problem.
  3. A system that runs on multiple computers.
  4. A system with multiple GPUs.
Show Answer
Answer: b) A system where multiple AI agents interact and collaborate to solve a problem.
Explanation
In a multi-agent system, several agents — each possibly with its own role, tools or expertise — communicate, coordinate and sometimes compete to accomplish a task that a single agent would struggle with.

14) In a multi-agent system, what is “emergent behavior”?

  1. Behavior that is explicitly programmed by the developer.
  2. Complex, system-level behavior that arises from the simple interactions of individual agents.
  3. The behavior of the most powerful agent in the system.
  4. A bug in the system.
Show Answer
Answer: b) Complex, system-level behavior that arises from the simple interactions of individual agents.
Explanation
Emergent behaviour is not coded into any single agent; it arises from the local interactions between agents and can be beneficial (e.g. coordinated problem solving) or undesirable.

15) What is a “zero-shot” task for an AI agent?

  1. A task it has been trained on extensively.
  2. A task it can perform without any specific prior training or examples.
  3. A task that requires zero computational resources.
  4. A task that takes zero seconds to complete.
Show Answer
Answer: b) A task it can perform without any specific prior training or examples.
Explanation
Zero-shot means the model is asked to perform a task with only an instruction — no worked examples or task-specific fine-tuning — relying on generalisation from pre-training.

16) What is the purpose of a “critic” or “evaluator” agent in a multi-agent system?

  1. To perform the main task.
  2. To provide feedback on the performance of other agents.
  3. To manage the system’s memory.
  4. To interact with the user.
Show Answer
Answer: b) To provide feedback on the performance of other agents.
Explanation
The critic reviews outputs and gives feedback, enabling an iterative generate–critique–revise loop that improves quality, correctness and safety.

17) Which of the following is a potential risk associated with autonomous AI agents?

  1. They might be too slow.
  2. They might perform unintended or harmful actions.
  3. They might use too much electricity.
  4. They might be too expensive to build.
Show Answer
Answer: b) They might perform unintended or harmful actions.
Explanation
Because agents act rather than merely generate text, a mis-specified goal, a misinterpreted instruction or a faulty tool call can lead to real, unintended and potentially harmful consequences.

18) What is a “simulated environment” used for in agentic AI research?

  1. To train agents in a safe and controlled setting before deploying them in the real world.
  2. To create realistic graphics for video games.
  3. To test the hardware the agent will run on.
  4. To market the AI agent to customers.
Show Answer
Answer: a) To train agents in a safe and controlled setting before deploying them in the real world.
Explanation
Simulations let agents explore, fail and learn cheaply and safely, with full control over conditions and the ability to reset the environment at will.

19) What is the “alignment problem” in AI?

  1. Ensuring the AI’s hardware and software are compatible.
  2. Ensuring the AI’s goals and actions align with human values and intentions.
  3. Aligning the text in a document.
  4. Aligning the AI’s processing speed with the user’s expectations.
Show Answer
Answer: b) Ensuring the AI’s goals and actions align with human values and intentions.
Explanation
Alignment research addresses the risk that a capable system optimises the literal objective it was given while violating what humans actually intended or valued.

20) What is the role of a “world model” in a goal-based agent?

  1. To store the agent’s personal memories.
  2. To represent the agent’s understanding of its environment and how its actions affect it.
  3. To generate the agent’s final report.
  4. To connect the agent to the internet.
Show Answer
Answer: b) To represent the agent’s understanding of its environment and how its actions affect it.
Explanation
The world model lets the agent simulate or predict the consequences of candidate actions, which is what makes goal-directed planning possible.

21) What is “prompt engineering” in the context of agentic AI?

  1. Designing the physical casing of the AI.
  2. The practice of designing and refining prompts to guide an LLM’s behavior.
  3. Building the servers that run the AI.
  4. Writing the low-level code for the AI’s core logic.
Show Answer
Answer: b) The practice of designing and refining prompts to guide an LLM’s behavior.
Explanation
Prompt engineering is the craft of writing instructions, examples, roles and constraints that reliably steer a model toward the desired behaviour without changing its weights.

22) What is a “stateful” agent?

  1. An agent that operates only in one country.
  2. An agent that maintains an internal state or memory across multiple interactions.
  3. An agent that cannot change its behavior.
  4. An agent that has no memory.
Show Answer
Answer: b) An agent that maintains an internal state or memory across multiple interactions.
Explanation
A stateful agent carries context forward, so later turns or tasks can build on earlier ones — the opposite of a stateless agent that treats every request in isolation.

23) What is “plan-and-execute” as a strategy for AI agents?

  1. The agent plans all steps upfront and then executes them without deviation.
  2. The agent executes actions randomly and then plans what to do next.
  3. The agent plans and executes at the same time.
  4. The agent only plans and never executes.
Show Answer
Answer: a) The agent plans all steps upfront and then executes them without deviation.
Explanation
Plan-and-execute separates a single planning phase from the execution phase, which reduces the number of expensive model calls — though many implementations add re-planning when a step fails.

24) What is the main advantage of a hierarchical agent architecture?

  1. It is simpler to implement than a flat architecture.
  2. It allows for abstraction, where a high-level agent can delegate sub-tasks to lower-level agents.
  3. It requires less computational power.
  4. It is less prone to errors.
Show Answer
Answer: b) It allows for abstraction, where a high-level agent can delegate sub-tasks to lower-level agents.
Explanation
Hierarchy gives abstraction and separation of concerns: a manager agent handles the goal and strategy while specialised worker agents handle the detailed execution of sub-tasks.

25) What is “self-correction” in an AI agent?

  1. The agent’s ability to fix its own hardware.
  2. The agent’s ability to identify and correct its own mistakes in reasoning or execution.
  3. The agent’s ability to correct the user’s mistakes.
  4. The agent’s ability to correct other agents.
Show Answer
Answer: b) The agent’s ability to identify and correct its own mistakes in reasoning or execution.
Explanation
Self-correction (reflection) is the agent reviewing its own output or trajectory, spotting errors or inconsistencies and revising before finalising the answer.

26) Which of the following is a key challenge in building long-term memory for AI agents?

  1. Storing and retrieving relevant information efficiently from a large history.
  2. Making the memory last for more than a day.
  3. Ensuring the memory is not too large to fit on a hard drive.
  4. The cost of the memory hardware.
Show Answer
Answer: a) Storing and retrieving relevant information efficiently from a large history.
Explanation
As an agent’s history grows, the hard part is not storage but recall: finding the small subset of past information that is actually relevant to the current task, quickly and accurately.

27) What is a “Vector Database” often used for in agentic AI?

  1. Storing the agent’s source code.
  2. Storing and retrieving text embeddings for long-term memory (semantic search).
  3. Storing the user’s personal information securely.
  4. Generating images for the agent.
Show Answer
Answer: b) Storing and retrieving text embeddings for long-term memory (semantic search).
Explanation
A vector database indexes embeddings and supports similarity search, so the agent can retrieve semantically related memories or documents even when the wording differs.

28) What is “retrieval-augmented generation” (RAG)?

  1. A technique where an LLM retrieves relevant documents from a knowledge base to inform its generation.
  2. A technique for generating random numbers.
  3. A method for compressing large language models.
  4. A way to generate text without any external data.
Show Answer
Answer: a) A technique where an LLM retrieves relevant documents from a knowledge base to inform its generation.
Explanation
RAG retrieves relevant passages and inserts them into the model’s context, grounding the answer in up-to-date or private knowledge without retraining the model.

29) What is a “guardrail” in the context of AI safety?

  1. A physical barrier around the AI’s server.
  2. A set of rules or constraints designed to prevent an AI from taking harmful actions.
  3. A type of neural network layer.
  4. A tool for measuring the AI’s speed.
Show Answer
Answer: b) A set of rules or constraints designed to prevent an AI from taking harmful actions.
Explanation
Guardrails are enforced constraints — content filters, allowed-tool lists, permission checks, human approval steps — that keep an agent within safe and intended boundaries.

30) What is a “sim-to-real” transfer in robotics and agentic AI?

  1. Transferring data from a simulation to a real-world system.
  2. The process of transferring an agent’s learned policy from a simulated environment to a physical robot.
  3. Transferring the agent’s code from one computer to another.
  4. Selling the agent’s technology to a real company.
Show Answer
Answer: b) The process of transferring an agent’s learned policy from a simulated environment to a physical robot.
Explanation
Sim-to-real transfer moves a policy trained in simulation onto real hardware, tackling the “reality gap” caused by modelling errors, noise and unmodelled physics.

31) What is the “exploration vs. exploitation” trade-off in reinforcement learning?

  1. Choosing between exploring new actions to find better rewards and exploiting known actions to get reliable rewards.
  2. Choosing between exploring new markets and exploiting existing ones.
  3. The trade-off between the cost of exploration and the cost of exploitation.
  4. A trade-off in database management.
Show Answer
Answer: a) Choosing between exploring new actions to find better rewards and exploiting known actions to get reliable rewards.
Explanation
An agent must balance trying unfamiliar actions to discover better strategies against repeatedly using what it already knows works well.

32) What is a “Partially Observable Markov Decision Process” (POMDP)?

  1. A model for environments where the agent cannot fully observe the state of the world.
  2. A model for environments that are completely random.
  3. A model for environments where the agent has full information.
  4. A model for a deterministic environment.
Show Answer
Answer: a) A model for environments where the agent cannot fully observe the state of the world.
Explanation
A POMDP extends the MDP with observations that give only partial information about the true state, so the agent must maintain a belief over possible states.

33) What is “in-context learning” for an LLM-based agent?

  1. The ability to learn from examples provided within the prompt without any weight updates.
  2. The ability to learn from a textbook.
  3. The ability to learn by interacting with the physical world.
  4. The ability to learn from a pre-training dataset.
Show Answer
Answer: a) The ability to learn from examples provided within the prompt without any weight updates.
Explanation
In-context learning is adaptation at inference time: examples in the prompt shape the model’s behaviour immediately, with no gradient updates to its parameters.

34) What is “fine-tuning” an LLM?

  1. Adjusting the temperature of the model.
  2. Further training a pre-trained LLM on a smaller, specific dataset to adapt it to a particular task.
  3. Writing a better prompt.
  4. Compressing the model for faster inference.
Show Answer
Answer: b) Further training a pre-trained LLM on a smaller, specific dataset to adapt it to a particular task.
Explanation
Fine-tuning continues training a pre-trained model on domain- or task-specific data, updating its weights so it behaves better on that narrower use case.

35) What is a “cognitive architecture” in AI?

  1. The physical layout of the AI’s processors.
  2. A blueprint for the structure of an intelligent agent, often inspired by human cognition.
  3. The architecture of the building where the AI is developed.
  4. A type of neural network.
Show Answer
Answer: b) A blueprint for the structure of an intelligent agent, often inspired by human cognition.
Explanation
A cognitive architecture specifies the fixed components and processes of an intelligent system — memory, perception, decision making, learning — as in SOAR or ACT-R.

36) What is the role of “attention” in a Transformer-based agent?

  1. To allow the model to focus on the most relevant parts of the input when making a decision.
  2. To make the model pay attention to the user’s emotions.
  3. To slow down the model’s processing.
  4. To store the model’s long-term memory.
Show Answer
Answer: a) To allow the model to focus on the most relevant parts of the input when making a decision.
Explanation
Self-attention computes weights over all positions in the context, letting the model dynamically prioritise the tokens most relevant to the current prediction.

37) What is a “soft actor-critic” (SAC)?

  1. A method for evaluating actor performance.
  2. An off-policy reinforcement learning algorithm that incorporates an entropy maximization term.
  3. A type of neural network layer.
  4. A tool for generating realistic dialogues.
Show Answer
Answer: b) An off-policy reinforcement learning algorithm that incorporates an entropy maximization term.
Explanation
SAC maximises expected reward plus an entropy bonus, which encourages exploration and produces more robust policies; it is an off-policy actor-critic method.

38) What is the primary goal of “AI alignment” research?

  1. To ensure AI systems are aligned with the goals of the company that built them.
  2. To ensure AI systems are aligned with human values and are beneficial to humanity.
  3. To align the AI’s hardware components.
  4. To align the text in a document.
Show Answer
Answer: b) To ensure AI systems are aligned with human values and are beneficial to humanity.
Explanation
Alignment aims to make AI systems pursue goals that genuinely reflect human values and intentions, rather than narrowly optimising a proxy objective.

39) What is a “constitutional AI”?

  1. An AI that can write constitutions for countries.
  2. A method for training AI to be helpful, harmless, and honest, using a set of principles (a “constitution”).
  3. An AI that is governed by the constitution of a country.
  4. An AI that is used for legal research.
Show Answer
Answer: b) A method for training AI to be helpful, harmless, and honest, using a set of principles (a “constitution”).
Explanation
Constitutional AI gives the model a written set of principles and has it critique and revise its own outputs against them, reducing reliance on large amounts of human harmlessness labelling.

40) What is “red teaming” in the context of AI development?

  1. A team that paints the AI’s server room red.
  2. A practice where a team of experts adversarially tests an AI system to find its vulnerabilities and flaws.
  3. The team that develops the AI’s core algorithms.
  4. The team that markets the AI product.
Show Answer
Answer: b) A practice where a team of experts adversarially tests an AI system to find its vulnerabilities and flaws.
Explanation
Red teaming deliberately probes a system for failure modes, jailbreaks, harmful outputs and misuse potential so that they can be fixed before deployment.

41) What is the “temperature” parameter in an LLM?

  1. The physical temperature of the server.
  2. A parameter that controls the randomness and creativity of the model’s output.
  3. A parameter that controls the model’s speed.
  4. A parameter that controls the model’s memory usage.
Show Answer
Answer: b) A parameter that controls the randomness and creativity of the model’s output.
Explanation
Temperature rescales the next-token probability distribution. Low values make the output focused and deterministic; high values make it more varied and creative.

42) What is “top-k sampling” in text generation?

  1. A method for evaluating the top k models.
  2. A sampling strategy where the model only considers the k most likely next words.
  3. A method for ranking the top k search results.
  4. A method for selecting the top k GPUs for training.
Show Answer
Answer: b) A sampling strategy where the model only considers the k most likely next words.
Explanation
Top-k sampling truncates the distribution to the k highest-probability tokens and samples from those, cutting off the unlikely long tail that can produce incoherent text.

43) What is a “hallucination” in the context of LLMs?

  1. A visual artifact in generated images.
  2. When a model generates plausible-sounding but incorrect or nonsensical information.
  3. When a model refuses to answer a question.
  4. When a model takes too long to respond.
Show Answer
Answer: b) When a model generates plausible-sounding but incorrect or nonsensical information.
Explanation
A hallucination is fluent, confident output that is factually wrong or unsupported by any source — a core reliability problem for LLM-based systems.

44) How can agentic AI help to reduce hallucinations?

  1. By using a larger LLM.
  2. By using tools to verify information from external sources and grounding the reasoning in factual data.
  3. By asking the user to provide the correct answer.
  4. By generating longer responses.
Show Answer
Answer: b) By using tools to verify information from external sources and grounding the reasoning in factual data.
Explanation
An agent can call search engines, databases, calculators or code execution to check facts and compute results, anchoring its answer in verifiable evidence rather than recall alone.

45) What is “grounding” in the context of AI agents?

  1. Connecting the AI’s electrical circuits to the ground.
  2. The process of linking the agent’s language and reasoning to real-world, verifiable information.
  3. Training the agent on a grounded dataset.
  4. A technique for stabilizing the agent’s training process.
Show Answer
Answer: b) The process of linking the agent’s language and reasoning to real-world, verifiable information.
Explanation
Grounding ties symbols and statements to actual entities, data and observations, so the agent’s claims can be traced back to something verifiable.

46) What is a “React” agent in the LangChain framework?

  1. An agent that reacts to user input in real-time.
  2. An agent that implements the ReAct (Reason + Act) framework.
  3. An agent that uses a reaction-based reward system.
  4. An agent that is easily angered.
Show Answer
Answer: b) An agent that implements the ReAct (Reason + Act) framework.
Explanation
LangChain’s ReAct agent follows the reason–act–observe loop, deciding which tool to call at each step and incorporating the observation into the next reasoning step.

47) What is a “Plan-and-Execute” agent in LangChain?

  1. An agent that creates a full plan and then executes it step-by-step.
  2. An agent that plans and executes simultaneously.
  3. An agent that only executes pre-defined plans.
  4. An agent that cannot create plans.
Show Answer
Answer: a) An agent that creates a full plan and then executes it step-by-step.
Explanation
The Plan-and-Execute pattern produces an explicit list of steps first, then runs them one at a time — often with a re-planning step if a result invalidates the plan.

48) What is the purpose of the “AgentExecutor” in LangChain?

  1. To execute the LLM’s code.
  2. To manage the agent’s execution loop, calling the agent, tools, and handling the output.
  3. To execute trades on the stock market.
  4. To train the agent.
Show Answer
Answer: b) To manage the agent’s execution loop, calling the agent, tools, and handling the output.
Explanation
AgentExecutor is the runtime that repeatedly invokes the agent, executes the tools it selects, feeds observations back, and terminates when a final answer is produced (or a limit is reached).

49) What is “self-consistency” as a reasoning technique for agents?

  1. The agent’s ability to consistently produce the same wrong answer.
  2. A technique where the agent generates multiple reasoning paths and chooses the most consistent answer.
  3. A technique for ensuring the agent’s hardware is consistent.
  4. A technique for making the agent’s personality consistent.
Show Answer
Answer: b) A technique where the agent generates multiple reasoning paths and chooses the most consistent answer.
Explanation
Self-consistency samples several chain-of-thought solutions and takes the majority answer, which is more reliable than a single greedy reasoning path.

50) What is a “tree of thoughts” (ToT) approach?

  1. A method for generating a tree diagram of the agent’s thoughts.
  2. A reasoning framework that explores multiple reasoning paths and can backtrack, like a tree search.
  3. A method for counting the number of thoughts an agent has.
  4. A method for visualizing the agent’s neural network.
Show Answer
Answer: b) A reasoning framework that explores multiple reasoning paths and can backtrack, like a tree search.
Explanation
Tree of Thoughts treats reasoning as a search over intermediate “thoughts”, allowing the agent to evaluate partial solutions and backtrack from dead ends.

51) What is the role of a “router” in a multi-agent system?

  1. To connect the system to the internet.
  2. To direct a task or query to the most appropriate agent for the job.
  3. To physically route cables in the server room.
  4. To manage the system’s power supply.
Show Answer
Answer: b) To direct a task or query to the most appropriate agent for the job.
Explanation
A router (dispatcher) classifies the incoming request and forwards it to the specialised agent or tool best suited to handle it, improving efficiency and accuracy.

52) What is a “blackboard system” in AI?

  1. A system for writing on a blackboard.
  2. A shared memory or data structure that multiple agents can read from and write to, in order to collaborate.
  3. A system for teaching AI.
  4. A type of display screen.
Show Answer
Answer: b) A shared memory or data structure that multiple agents can read from and write to, in order to collaborate.
Explanation
In a blackboard architecture, independent knowledge sources post partial results to a common workspace, and other agents react to whatever appears there.

53) What is a “contract net protocol” in multi-agent systems?

  1. A legal contract for AI agents.
  2. A protocol for task allocation where an agent announces a task and other agents bid to perform it.
  3. A protocol for secure communication between agents.
  4. A protocol for shutting down agents.
Show Answer
Answer: b) A protocol for task allocation where an agent announces a task and other agents bid to perform it.
Explanation
The contract net protocol uses announcement, bidding and award phases, so the agent best able to complete the task wins the contract.

54) What is “swarm intelligence”?

  1. The intelligence of a single, very smart agent.
  2. The collective behavior of decentralized, self-organized systems, natural or artificial.
  3. A type of AI that is shaped like a swarm of bees.
  4. A type of AI that is very fast.
Show Answer
Answer: b) The collective behavior of decentralized, self-organized systems, natural or artificial.
Explanation
Swarm intelligence describes how simple individuals following local rules produce sophisticated collective behaviour, as seen in ant colonies, bird flocks and particle swarms.

55) Which of the following is an example of a swarm intelligence algorithm?

  1. Particle Swarm Optimization (PSO)
  2. Gradient Descent
  3. Backpropagation
  4. Convolution
Show Answer
Answer: a) Particle Swarm Optimization (PSO)
Explanation
PSO optimises a function using a population of particles that share information, inspired by flocking behaviour. Gradient descent, backpropagation and convolution are not swarm methods.

56) What is the “frame problem” in AI?

  1. The problem of fitting an AI into a single frame.
  2. The challenge of representing in a formal logic system what does not change when an action is performed.
  3. The problem of displaying images on a screen.
  4. The problem of building the physical frame of a robot.
Show Answer
Answer: b) The challenge of representing in a formal logic system what does not change when an action is performed.
Explanation
The frame problem is the difficulty of specifying, without an explosion of axioms, everything that remains unaffected by an action in a logical world model.

57) What is the “symbol grounding problem”?

  1. The problem of grounding electrical symbols.
  2. The challenge of how to connect abstract symbols in an AI to their real-world meanings.
  3. The problem of using too many symbols in a prompt.
  4. The problem of creating new symbols for AI.
Show Answer
Answer: b) The challenge of how to connect abstract symbols in an AI to their real-world meanings.
Explanation
Symbol grounding asks how a system that only manipulates tokens can attach genuine meaning to them — how “apple” connects to actual apples rather than to other symbols.

58) What is “embodied AI”?

  1. AI that has a physical body (e.g., a robot).
  2. AI that is disembodied and exists only in the cloud.
  3. AI that is used in the fashion industry.
  4. AI that is used in the medical industry.
Show Answer
Answer: a) AI that has a physical body (e.g., a robot).
Explanation
Embodied AI learns and acts through a physical body, using sensors and actuators to interact with the real world — an idea closely tied to solving symbol grounding.

59) What is a “digital twin”?

  1. A virtual replica of a physical entity, process, or system.
  2. A person who looks identical to another person.
  3. A backup of a computer’s hard drive.
  4. A type of AI agent.
Show Answer
Answer: a) A virtual replica of a physical entity, process, or system.
Explanation
A digital twin mirrors a real system in software, often fed by live sensor data, so it can be monitored, simulated and optimised without touching the physical asset.

60) What is “reinforcement learning from human feedback” (RLHF)?

  1. A technique for training agents using a reward model learned from human preferences.
  2. A technique for training agents by having humans manually control them.
  3. A technique for training agents using a pre-defined reward function.
  4. A technique for training agents without any feedback.
Show Answer
Answer: a) A technique for training agents using a reward model learned from human preferences.
Explanation
RLHF collects human comparisons between outputs, trains a reward model on those preferences, and then optimises the policy with reinforcement learning against that model.

61) What is a “reward model” in RLHF?

  1. A model that predicts the reward a human would give for a given action or output.
  2. A model that gives itself a reward.
  3. A model that calculates the monetary cost of an action.
  4. A model that is used to reward the developers.
Show Answer
Answer: a) A model that predicts the reward a human would give for a given action or output.
Explanation
The reward model is trained on human preference data and acts as a differentiable stand-in for human judgement, so the policy can be optimised at scale.

62) What is “direct preference optimization” (DPO)?

  1. A method for directly optimizing a model on human preference data without training a separate reward model.
  2. A method for optimizing the direct deposit of paychecks.
  3. A method for optimizing the direct path of a robot.
  4. A method for optimizing the direct connection between two agents.
Show Answer
Answer: a) A method for directly optimizing a model on human preference data without training a separate reward model.
Explanation
DPO reformulates the preference objective so the language model itself can be optimised directly on preference pairs, avoiding the separate reward model and RL loop used in RLHF.

63) What is “proximal policy optimization” (PPO)?

  1. A popular reinforcement learning algorithm used for training agents.
  2. A method for optimizing the proximity of sensors.
  3. A method for optimizing the proximity of agents to each other.
  4. A method for optimizing the proximity of a database.
Show Answer
Answer: a) A popular reinforcement learning algorithm used for training agents.
Explanation
PPO is a policy-gradient method that constrains how far each update can move the policy, giving stable and reliable training; it is widely used in RLHF pipelines.

64) What is “curiosity-driven exploration” in reinforcement learning?

  1. An intrinsic motivation for an agent to explore its environment to reduce uncertainty.
  2. An agent’s desire to learn about human culture.
  3. An agent’s desire to explore the internet.
  4. A method for making the agent more interesting to talk to.
Show Answer
Answer: a) An intrinsic motivation for an agent to explore its environment to reduce uncertainty.
Explanation
Intrinsic curiosity rewards the agent for reaching novel or unpredictable states (for example, via prediction error), which drives exploration when external rewards are sparse.

65) What is a “sparse reward” environment?

  1. An environment where the agent receives rewards very infrequently.
  2. An environment where the rewards are very small.
  3. An environment where the rewards are very large.
  4. An environment where there are no rewards.
Show Answer
Answer: a) An environment where the agent receives rewards very infrequently.
Explanation
With sparse rewards, the agent gets signal only rarely, making credit assignment hard and learning slow — a common motivation for reward shaping and curiosity.

66) What is “reward shaping”?

  1. The process of designing the physical shape of the reward.
  2. The practice of modifying the reward function to guide the agent’s learning process.
  3. The process of making the reward more aesthetically pleasing.
  4. The process of distributing rewards to the development team.
Show Answer
Answer: b) The practice of modifying the reward function to guide the agent’s learning process.
Explanation
Reward shaping adds intermediate or auxiliary reward terms so the agent gets more frequent guidance — though poorly designed shaping can introduce unintended behaviour.

67) What is a “Markov Decision Process” (MDP)?

  1. A mathematical framework for modeling decision-making in situations where outcomes are partly random and partly under the control of a decision-maker.
  2. A process for making decisions in a deterministic environment.
  3. A process for making decisions without any state.
  4. A process for making decisions based on a fixed set of rules.
Show Answer
Answer: a) A mathematical framework for modeling decision-making in situations where outcomes are partly random and partly under the control of a decision-maker.
Explanation
An MDP is defined by states, actions, transition probabilities and rewards, and it underpins most of reinforcement learning theory.

68) What is the “Bellman equation” used for in reinforcement learning?

  1. To calculate the speed of an agent.
  2. To relate the value of a state to the values of its possible successor states.
  3. To calculate the cost of training an agent.
  4. To measure the agent’s intelligence.
Show Answer
Answer: b) To relate the value of a state to the values of its possible successor states.
Explanation
The Bellman equation expresses a state’s value as the immediate reward plus the discounted value of the states that follow — the recursive relationship behind value iteration and Q-learning.

69) What is “Q-learning”?

  1. A model-free reinforcement learning algorithm that learns the value of taking a given action in a given state.
  2. A method for learning how to ask questions.
  3. A method for learning how to queue tasks.
  4. A method for learning how to be quiet.
Show Answer
Answer: a) A model-free reinforcement learning algorithm that learns the value of taking a given action in a given state.
Explanation
Q-learning estimates Q(s, a) — the expected return of taking action a in state s and acting optimally thereafter — without needing a model of the environment’s dynamics.

70) What is a “Deep Q-Network” (DQN)?

  1. A Q-learning algorithm that uses a deep neural network to approximate the Q-value function.
  2. A Q-learning algorithm that uses a deep physical network.
  3. A Q-learning algorithm that is very deep in thought.
  4. A Q-learning algorithm that is used for deep-sea exploration.
Show Answer
Answer: a) A Q-learning algorithm that uses a deep neural network to approximate the Q-value function.
Explanation
DQN replaces the Q-table with a neural network, using experience replay and a target network to stabilise training — famously applied to Atari games.

71) What is the “credit assignment problem” in reinforcement learning?

  1. The problem of assigning credit to the right actions for a successful outcome, especially when rewards are delayed.
  2. The problem of assigning credit to the developers of the AI.
  3. The problem of assigning credit to the users of the AI.
  4. The problem of getting a loan for an AI project.
Show Answer
Answer: a) The problem of assigning credit to the right actions for a successful outcome, especially when rewards are delayed.
Explanation
When a reward arrives long after the actions that caused it, the learner must work out which of many preceding actions deserves credit — a central difficulty in RL.

72) What is an “actor-critic” method?

  1. A reinforcement learning method that combines a policy-based “actor” and a value-based “critic”.
  2. A method for evaluating the performance of actors in a movie.
  3. A method for critiquing the work of other AI agents.
  4. A method for acting and critiquing at the same time.
Show Answer
Answer: a) A reinforcement learning method that combines a policy-based “actor” and a value-based “critic”.
Explanation
The actor proposes actions while the critic estimates value and supplies the learning signal, which reduces the variance of pure policy-gradient updates.

73) What is “off-policy” learning?

  1. Learning about a policy that is different from the one being used to make decisions.
  2. Learning about a policy that is no longer in use.
  3. Learning about a policy that is not important.
  4. Learning about a policy that is used by a different company.
Show Answer
Answer: a) Learning about a policy that is different from the one being used to make decisions.
Explanation
Off-policy methods (such as Q-learning) can learn about a target policy while behaving according to a different, more exploratory behaviour policy.

74) What is “on-policy” learning?

  1. Learning about the same policy that is being used to make decisions.
  2. Learning about a policy that is currently in use.
  3. Learning about a policy that is popular.
  4. Learning about a policy that is on a website.
Show Answer
Answer: a) Learning about the same policy that is being used to make decisions.
Explanation
On-policy methods (such as SARSA and PPO) evaluate and improve exactly the policy that is generating the experience.

75) What is a “policy gradient” method?

  1. A reinforcement learning method that directly optimizes the policy without learning a value function.
  2. A method for calculating the gradient of a policy.
  3. A method for making policies more gradual.
  4. A method for evaluating the performance of a policy.
Show Answer
Answer: a) A reinforcement learning method that directly optimizes the policy without learning a value function.
Explanation
Policy gradient methods parameterise the policy and adjust its parameters in the direction that increases expected return, without needing a separate value function (though actor-critic variants add one).

76) What is “meta-learning” in AI?

  1. “Learning to learn” – the ability of a model to learn new tasks quickly with few examples.
  2. Learning about metal.
  3. Learning about the meta-data of a dataset.
  4. Learning about the history of AI.
Show Answer
Answer: a) “Learning to learn” – the ability of a model to learn new tasks quickly with few examples.
Explanation
Meta-learning trains a model across many tasks so that it can adapt rapidly to a new task from only a handful of examples.

77) What is a “few-shot” learning task?

  1. A task where the model is given a few examples to learn from.
  2. A task that requires a few shots to complete.
  3. A task that is not very important.
  4. A task that is performed by a few agents.
Show Answer
Answer: a) A task where the model is given a few examples to learn from.
Explanation
Few-shot prompting supplies a small number of worked examples in the prompt so the model can infer the pattern and apply it to a new input.

78) What is “transfer learning”?

  1. The ability to transfer knowledge learned from one task to another.
  2. The ability to transfer money between bank accounts.
  3. The ability to transfer files between computers.
  4. The ability to transfer an agent from one environment to another.
Show Answer
Answer: a) The ability to transfer knowledge learned from one task to another.
Explanation
Transfer learning reuses representations or parameters learned on a source task to speed up and improve learning on a related target task.

79) What is “continual learning” or “lifelong learning”?

  1. The ability of an agent to learn continuously over time without forgetting previous knowledge.
  2. The ability of an agent to learn for its entire lifetime.
  3. The ability of an agent to learn continuously without stopping.
  4. The ability of an agent to learn about life.
Show Answer
Answer: a) The ability of an agent to learn continuously over time without forgetting previous knowledge.
Explanation
Continual learning aims at acquiring new skills sequentially while retaining earlier ones — the challenge being to avoid catastrophic forgetting.

80) What is “catastrophic forgetting”?

  1. When a neural network forgets previously learned information upon learning new information.
  2. When a robot forgets how to walk.
  3. When an AI forgets its own name.
  4. When a computer’s hard drive fails and it forgets all data.
Show Answer
Answer: a) When a neural network forgets previously learned information upon learning new information.
Explanation
Gradient updates on new data can overwrite the weights that encoded old tasks, causing an abrupt loss of previously acquired performance.

81) What is “elastic weight consolidation” (EWC)?

  1. A technique to mitigate catastrophic forgetting by slowing down the learning of weights that are important for previous tasks.
  2. A technique for consolidating the weight of a physical object.
  3. A technique for making a neural network more elastic.
  4. A technique for consolidating multiple neural networks into one.
Show Answer
Answer: a) A technique to mitigate catastrophic forgetting by slowing down the learning of weights that are important for previous tasks.
Explanation
EWC adds a penalty that resists changes to weights identified as important for earlier tasks, letting the network learn new tasks while preserving old ones.

82) What is “prompt injection”?

  1. A security vulnerability where a user can manipulate an AI’s behavior by injecting malicious text into a prompt.
  2. A type of medical injection for computers.
  3. A method for injecting new knowledge into an AI.
  4. A method for speeding up an AI’s response time.
Show Answer
Answer: a) A security vulnerability where a user can manipulate an AI’s behavior by injecting malicious text into a prompt.
Explanation
Prompt injection embeds instructions in user input (or in retrieved documents and web pages) that hijack the model’s original instructions — a serious risk for tool-using agents.

83) What is “jailbreaking” an LLM?

  1. The process of breaking an LLM out of a physical jail.
  2. The process of using clever prompts to bypass an LLM’s safety and ethical restrictions.
  3. The process of installing an LLM on a jailbroken phone.
  4. The process of making an LLM run faster.
Show Answer
Answer: b) The process of using clever prompts to bypass an LLM’s safety and ethical restrictions.
Explanation
Jailbreaking uses role-play, obfuscation or adversarial phrasing to get a model to produce content its safety training was designed to refuse.

84) What is a “backdoor” attack on an AI model?

  1. A physical door at the back of the server room.
  2. A hidden vulnerability deliberately introduced into a model during training that can be triggered by a specific input.
  3. A method for accessing the model’s source code.
  4. A method for bypassing the model’s safety filters.
Show Answer
Answer: b) A hidden vulnerability deliberately introduced into a model during training that can be triggered by a specific input.
Explanation
A backdoored (trojaned) model behaves normally on ordinary inputs but produces attacker-chosen behaviour when it sees a secret trigger pattern.

85) What is “model inversion”?

  1. A type of privacy attack where an attacker can reconstruct the training data from a trained model.
  2. A technique for inverting a model’s predictions.
  3. A method for turning a model upside down.
  4. A method for reversing the training process.
Show Answer
Answer: a) A type of privacy attack where an attacker can reconstruct the training data from a trained model.
Explanation
Model inversion exploits a model’s outputs or gradients to recover sensitive properties of the data it was trained on, such as a face or a specific attribute.

86) What is “membership inference”?

  1. A privacy attack where an attacker tries to determine if a specific data point was part of a model’s training set.
  2. A method for inferring the membership of a club.
  3. A method for inferring the number of members in a group.
  4. A method for adding new members to a model’s training set.
Show Answer
Answer: a) A privacy attack where an attacker tries to determine if a specific data point was part of a model’s training set.
Explanation
Because models often behave more confidently on data they were trained on, an attacker can use output patterns to infer whether a given record was in the training set.

87) What is “federated learning”?

  1. A machine learning technique that trains an algorithm across multiple decentralized devices or servers holding local data samples, without exchanging them.
  2. A method for learning about different federations.
  3. A method for learning how to be a federal agent.
  4. A method for learning in a group.
Show Answer
Answer: a) A machine learning technique that trains an algorithm across multiple decentralized devices or servers holding local data samples, without exchanging them.
Explanation
In federated learning, devices train locally and share only model updates, keeping the raw data private on each device.

88) What is “differential privacy”?

  1. A system for publicly sharing information about a dataset by describing the patterns of groups within the dataset while withholding information about individuals in the dataset.
  2. A method for treating all data points differently.
  3. A method for calculating the difference between two datasets.
  4. A method for making a dataset private.
Show Answer
Answer: a) A system for publicly sharing information about a dataset by describing the patterns of groups within the dataset while withholding information about individuals in the dataset.
Explanation
Differential privacy adds calibrated noise so that the presence or absence of any single individual changes the published result only negligibly.

89) What is “AI explainability” (XAI)?

  1. The ability to explain the decisions and predictions made by an AI model in a way that humans can understand.
  2. The ability of an AI to explain a joke.
  3. The ability of an AI to explain its own existence.
  4. The ability of an AI to explain how to build it.
Show Answer
Answer: a) The ability to explain the decisions and predictions made by an AI model in a way that humans can understand.
Explanation
XAI develops methods that make a model’s reasoning and outputs interpretable, which matters for trust, debugging, auditing and regulatory compliance.

90) What is “LIME” (Local Interpretable Model-agnostic Explanations)?

  1. A technique for explaining the predictions of any machine learning classifier in a local, interpretable way.
  2. A type of citrus fruit.
  3. A type of software for building AI.
  4. A type of neural network.
Show Answer
Answer: a) A technique for explaining the predictions of any machine learning classifier in a local, interpretable way.
Explanation
LIME perturbs the input around a single prediction and fits a simple interpretable model locally to reveal which features drove that prediction.

91) What is “SHAP” (SHapley Additive exPlanations)?

  1. A game-theoretic approach to explain the output of any machine learning model.
  2. A type of shape.
  3. A type of shampoo.
  4. A type of software for building AI.
Show Answer
Answer: a) A game-theoretic approach to explain the output of any machine learning model.
Explanation
SHAP uses Shapley values from cooperative game theory to fairly distribute the credit for a prediction among the input features.

92) What is “automated machine learning” (AutoML)?

  1. The process of automating the end-to-end process of applying machine learning to real-world problems.
  2. The process of building a machine that can learn by itself.
  3. The process of automating a physical machine.
  4. The process of building an AI that can build other AIs.
Show Answer
Answer: a) The process of automating the end-to-end process of applying machine learning to real-world problems.
Explanation
AutoML automates data preparation, feature engineering, model selection and hyperparameter tuning so that non-experts can build effective models.

93) What is “neural architecture search” (NAS)?

  1. A technique for automating the design of neural network architectures.
  2. A technique for searching for neural networks on the internet.
  3. A technique for searching for neurons in a brain.
  4. A technique for searching for a new type of neural network.
Show Answer
Answer: a) A technique for automating the design of neural network architectures.
Explanation
NAS searches over possible layer types, connections and hyperparameters to discover high-performing architectures automatically.

94) What is a “foundation model”?

  1. A large AI model trained on a vast quantity of data at scale, which can be adapted to a wide range of downstream tasks.
  2. A model that is used to build the foundation of a building.
  3. A model that is the first of its kind.
  4. A model that is very important.
Show Answer
Answer: a) A large AI model trained on a vast quantity of data at scale, which can be adapted to a wide range of downstream tasks.
Explanation
Foundation models are pre-trained broadly (often self-supervised) and then adapted by prompting or fine-tuning to many different downstream applications.

95) What is a “generalist agent”?

  1. An agent that can perform a wide variety of tasks, as opposed to a “specialist” agent that is good at one thing.
  2. An agent that is a general in the army.
  3. An agent that is very general in its approach.
  4. An agent that is not very specific.
Show Answer
Answer: a) An agent that can perform a wide variety of tasks, as opposed to a “specialist” agent that is good at one thing.
Explanation
A generalist agent handles many different tasks with a single set of parameters, whereas a specialist agent is optimised for one narrow domain.

96) What is “Gato” from DeepMind an example of?

  1. A generalist agent.
  2. A specialist agent.
  3. A type of robot.
  4. A type of cat.
Show Answer
Answer: a) A generalist agent.
Explanation
Gato is a single transformer-based model trained to perform hundreds of tasks — text, image captioning, robotic control and more — making it a generalist agent.

97) What is the “Turing Test”?

  1. A test of a machine’s ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human.
  2. A test for building a Turing machine.
  3. A test for a type of computer.
  4. A test for a type of AI.
Show Answer
Answer: a) A test of a machine’s ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human.
Explanation
Proposed by Alan Turing, the test judges intelligence by whether a human interlocutor cannot reliably tell the machine’s responses from a human’s.

98) What is the “Chinese Room” argument?

  1. A philosophical argument against the idea that a computer can have a “mind” or “consciousness” simply by manipulating symbols.
  2. An argument about the best place to put a computer in a house.
  3. An argument about the best type of tea to drink while programming.
  4. An argument about the size of a computer.
Show Answer
Answer: a) A philosophical argument against the idea that a computer can have a “mind” or “consciousness” simply by manipulating symbols.
Explanation
Searle’s Chinese Room argues that following syntactic rules to produce correct answers does not imply semantic understanding or consciousness.

99) What is “AGI” (Artificial General Intelligence)?

  1. A hypothetical type of AI that possesses the ability to understand, learn, and apply intelligence to a wide variety of problems, at a level comparable to a human being.
  2. A type of AI that is very good at one specific task.
  3. A type of AI that is used for general purposes.
  4. A type of AI that is generally intelligent.
Show Answer
Answer: a) A hypothetical type of AI that possesses the ability to understand, learn, and apply intelligence to a wide variety of problems, at a level comparable to a human being.
Explanation
AGI contrasts with narrow AI: rather than excelling at one task, it would transfer knowledge and reasoning flexibly across domains at human level.

100) What is the “singularity” in the context of AI?

  1. A hypothetical future point in time when technological growth becomes uncontrollable and irreversible, resulting in unforeseeable changes to human civilization.
  2. A type of black hole.
  3. A type of computer.
  4. A type of AI.
Show Answer
Answer: a) A hypothetical future point in time when technological growth becomes uncontrollable and irreversible, resulting in unforeseeable changes to human civilization.
Explanation
Popularised by writers such as I. J. Good and Vernor Vinge, the singularity is the speculative moment when self-improving machine intelligence accelerates beyond human ability to predict or control it.
100 AI Agents MCQ (Multiple Choice Questions) with Answers
100 Generative AI MCQ (Multiple Choice Questions) with Answers
Print page
0 Likes
Studyopedia Editorial Staff
contact@studyopedia.com

We work to create programming tutorials for all.

No Comments

Post A Comment