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

Posted at 21:42h in mcq by

1) What is the primary characteristic that defines an AI agent?
  1. It is a physical robot.
  2. It can only perform tasks it is explicitly programmed for.
  3. It perceives its environment and takes actions to achieve goals.
  4. It operates without any form of input.
Show Answer
Answer: c
Explanation
An AI agent is defined by its perceive–decide–act loop: it senses its environment through sensors and acts upon it through actuators in order to achieve its goals.

2) In the context of AI, what does “percept” refer to?
  1. The agent’s action in the environment.
  2. The agent’s internal state of mind.
  3. The perceptual input the agent receives at a given moment.
  4. The goal the agent is trying to achieve.
Show Answer
Answer: c
Explanation
A percept is a single perceptual input received by the agent’s sensors at a given moment. The full history of percepts is called the percept sequence.

3) What is an agent’s “policy”?
  1. A legal document governing its use.
  2. A function that maps states to actions.
  3. The hardware it runs on.
  4. Its ability to learn from experience.
Show Answer
Answer: b
Explanation
A policy is the agent’s mapping from states (or percepts) to actions — it defines what the agent will do in any given situation.

4) What is the role of an “actuator” in an AI agent?
  1. To sense the environment.
  2. To process information and make decisions.
  3. To perform an action that changes the environment.
  4. To store the agent’s memory.
Show Answer
Answer: c
Explanation
Actuators (wheels, motors, speakers, screen output, API calls) carry out the agent’s actions and thereby change the state of the environment. Sensors do the opposite.

5) Which of the following is NOT a typical component of an AI agent?
  1. Perception
  2. Reasoning/Decision Making
  3. Action
  4. A human operator providing real-time commands for every step
Show Answer
Answer: d
Explanation
An agent is characterised by autonomy. Requiring a human to issue a command for every single step removes that autonomy and makes the system a remotely controlled tool rather than an agent.

6) What is a “rational agent”?
  1. An agent that always makes the logically perfect move.
  2. An agent that acts to maximize its expected performance measure.
  3. An agent that can explain its reasoning.
  4. An agent that never makes a mistake.
Show Answer
Answer: b
Explanation
Rationality is defined relative to the information available: a rational agent chooses the action that maximises its expected performance measure, not a perfect outcome.

7) The performance measure of an AI agent is used to:
  1. Measure its processing speed.
  2. Evaluate how successful the agent is at achieving its goals.
  3. Determine the cost of the hardware.
  4. Calculate the amount of memory it uses.
Show Answer
Answer: b
Explanation
The performance measure is the objective criterion that defines success in the task — for example, distance travelled, profit earned, or games won.

8) What is the key difference between an agent and a simple program?
  1. Agents are always written in Python.
  2. Agents operate autonomously in an environment, while a simple program typically executes a fixed sequence of instructions.
  3. Agents are always connected to the internet.
  4. Simple programs cannot use loops.
Show Answer
Answer: b
Explanation
An agent continuously senses its environment and chooses actions to pursue goals, whereas a conventional program simply runs through a predetermined sequence of instructions.

9) What is “autonomy” in the context of AI agents?
  1. The ability to operate without human intervention.
  2. The ability to connect to other agents.
  3. The ability to learn from data.
  4. The ability to move physically.
Show Answer
Answer: a
Explanation
Autonomy means the agent can decide and act on its own, based on its own percepts and internal state, without step-by-step human control.

10) Which of these is a classic example of a simple reflex agent?
  1. A self-driving car.
  2. A thermostat.
  3. A chess-playing AI.
  4. A stock trading bot.
Show Answer
Answer: b
Explanation
A thermostat reacts directly to the current percept (temperature) with a fixed rule: too cold → turn heat on, too hot → turn heat off. It keeps no memory and does no planning.

11) What is the main limitation of a simple reflex agent?
  1. It is too slow.
  2. It cannot handle partially observable environments.
  3. It requires too much memory.
  4. It cannot be implemented in software.
Show Answer
Answer: b
Explanation
A simple reflex agent conditions only on the current percept, so if part of the relevant state is hidden, it has no way to know the true situation and its rules will fail.

12) A model-based reflex agent overcomes the limitations of a simple reflex agent by:
  1. Using a faster processor.
  2. Maintaining an internal state that depends on the percept history.
  3. Having a larger set of rules.
  4. Connecting to a cloud-based AI.
Show Answer
Answer: b
Explanation
A model-based agent keeps track of the parts of the world it cannot currently see, updating an internal state using a model of how the world evolves and how its own actions affect it.

13) What is the defining characteristic of a goal-based agent?
  1. It has a set of rules for every possible situation.
  2. It chooses actions to reach a specific goal state.
  3. It only responds to immediate stimuli.
  4. It learns from rewards and punishments.
Show Answer
Answer: b
Explanation
Goal-based agents consider the future — they search and plan for action sequences that will lead them to a specified goal state.

14) How does a utility-based agent differ from a goal-based agent?
  1. It doesn’t have goals.
  2. It uses a utility function to measure the “happiness” or desirability of different states, allowing it to choose the best goal to pursue.
  3. It is always more efficient.
  4. It can only be used in games.
Show Answer
Answer: b
Explanation
A goal-based agent only distinguishes goal states from non-goal states. A utility-based agent scores every possible state, so it can compare competing goals and trade off likelihood against desirability.

15) A learning agent is typically composed of which four conceptual components?
  1. Sensors, Actuators, CPU, Memory
  2. Critic, Learning Element, Performance Element, Problem Generator
  3. Input, Output, Processing, Storage
  4. Model, Policy, Value Function, Reward
Show Answer
Answer: b
Explanation
The classic learning-agent architecture consists of the performance element (acting), the learning element (improving), the critic (feedback) and the problem generator (suggesting exploratory actions).

16) In a learning agent, what is the role of the “critic”?
  1. It generates new problems for the agent to solve.
  2. It provides feedback on the agent’s actions based on the performance standard.
  3. It selects the next action to take.
  4. It stores all past experiences.
Show Answer
Answer: b
Explanation
The critic observes the percept sequence and compares the performance element’s behaviour against a fixed performance standard, telling the learning element how well the agent is doing.

17) Which agent architecture is most suitable for an environment that is partially observable and stochastic?
  1. Simple reflex agent
  2. Model-based reflex agent
  3. Goal-based agent
  4. Utility-based learning agent
Show Answer
Answer: d
Explanation
Partial observability requires internal state, stochasticity requires reasoning about probabilities and expected utility, and an unknown or changing environment benefits from learning — all of which a utility-based learning agent provides.

18) What is a “state” in the context of an AI agent?
  1. The country the agent is in.
  2. A representation of the environment at a particular point in time.
  3. The agent’s current mood.
  4. The programming language used.
Show Answer
Answer: b
Explanation
A state is a snapshot description of the environment (and, for model-based agents, the agent’s own internal condition) at a given moment.

19) A “PEAS” description stands for:
  1. Performance, Environment, Actuators, Sensors
  2. Planning, Execution, Analysis, Storage
  3. Perception, Evaluation, Action, State
  4. Problem, Environment, Agent, Solution
Show Answer
Answer: a
Explanation
PEAS — Performance measure, Environment, Actuators, Sensors — is the standard checklist for specifying an agent’s task environment.

20) For a self-driving car, which of the following is an actuator?
  1. Camera
  2. GPS
  3. Steering wheel
  4. LIDAR
Show Answer
Answer: c
Explanation
The steering wheel (along with the accelerator, brakes and indicators) is an actuator because it changes the state of the environment. Cameras, GPS and LIDAR are all sensors.

21) An environment is “fully observable” if:
  1. The agent can see the entire environment at once.
  2. The agent’s sensors can detect all aspects that are relevant to the choice of action.
  3. The environment is always in a known state.
  4. The environment is static.
Show Answer
Answer: b
Explanation
Full observability is about relevance, not completeness: the sensors give access to the complete state of the environment at each point in time, so the agent needs no memory.

22) A “deterministic” environment is one where:
  1. The next state is completely determined by the current state and the agent’s action.
  2. The outcome of an action is always random.
  3. The environment changes on its own.
  4. The agent has no control.
Show Answer
Answer: a
Explanation
In a deterministic environment there is no uncertainty in the outcome of an action. If outcomes are only probable, the environment is stochastic (non-deterministic).

23) What is an “episodic” environment?
  1. An environment that is divided into episodes, where the quality of the agent’s action depends only on the current episode.
  2. An environment that changes continuously.
  3. An environment where the agent’s actions have long-term consequences.
  4. An environment that is based on a story.
Show Answer
Answer: a
Explanation
In an episodic environment the agent’s experience is split into independent episodes, and the choice of action in one episode does not affect the next (e.g., identifying defective parts on an assembly line).

24) A “static” environment is one that:
  1. Does not change while the agent is deliberating.
  2. Never changes at all.
  3. Changes only when the agent acts.
  4. Is always the same.
Show Answer
Answer: a
Explanation
A static environment only requires that nothing changes while the agent is deciding what to do. It may still change between the agent’s actions — that would be dynamic.

25) A “discrete” environment is one where:
  1. There are a limited number of distinct states and actions.
  2. Time is continuous.
  3. The agent’s actions are smooth and continuous.
  4. The environment is quiet.
Show Answer
Answer: a
Explanation
Discrete environments have a finite (or countable) set of percepts, states and actions — like chess. Continuous environments, like driving, have smoothly varying values.

26) Which of the following is an example of a continuous environment?
  1. A chess game.
  2. A game of Go.
  3. Driving a car.
  4. A turn-based strategy game.
Show Answer
Answer: c
Explanation
Driving involves continuous state variables (position, speed, steering angle) and continuous time. Chess, Go and turn-based games are discrete.

27) In a “multi-agent” environment, what is the key consideration?
  1. The environment is very large.
  2. The agent’s actions can affect other agents, and other agents’ actions can affect it.
  3. The environment is always cooperative.
  4. The agent must be faster than the others.
Show Answer
Answer: b
Explanation
In a multi-agent environment the world is not fixed from any one agent’s viewpoint: other agents are part of the environment and their behaviour changes in response to your agent.

28) What is the “frame problem” in AI?
  1. The difficulty of representing a changing environment in a logical framework.
  2. The problem of building a physical frame for a robot.
  3. The issue of processing video frames quickly.
  4. The challenge of designing a user interface.
Show Answer
Answer: a
Explanation
The frame problem is the challenge of representing, in a logical formalism, which aspects of the world stay the same when an action is performed, without having to state every non-change explicitly.

29) The “qualification problem” in AI refers to:
  1. The difficulty of listing all the conditions required for an action to be successful.
  2. The problem of getting a qualification in AI.
  3. The issue of qualifying for a research grant.
  4. The challenge of meeting the qualifications for a job.
Show Answer
Answer: a
Explanation
It is practically impossible to enumerate every precondition that must hold for an action to work (e.g., “the car starts, provided there is fuel, the battery is charged, the key is right…”).

30) The “ramification problem” in AI is about:
  1. The difficulty of predicting all the indirect consequences of an action.
  2. The problem of ramifying (branching) code.
  3. The issue of using too much RAM.
  4. The challenge of managing multiple agents.
Show Answer
Answer: a
Explanation
The ramification problem is the difficulty of capturing all the indirect, knock-on effects of an action (moving a cup also moves the liquid inside it), not just the immediate ones.

31) What is Reinforcement Learning (RL)?
  1. Learning by being told the correct answer.
  2. Learning by trial and error to maximize a reward signal.
  3. Learning by observing a human expert.
  4. Learning by clustering data.
Show Answer
Answer: b
Explanation
In RL an agent learns a policy through interaction with its environment, receiving scalar rewards, with the goal of maximising cumulative long-term reward.

32) In RL, what is the “reward”?
  1. A prize given to the programmer.
  2. A numerical signal that indicates the immediate desirability of a state or action.
  3. The final goal of the agent.
  4. A type of neural network layer.
Show Answer
Answer: b
Explanation
The reward is the immediate scalar feedback the environment gives after each action. The agent’s objective is to maximise the total reward accumulated over time, not just the immediate one.

33) What is a “policy” in Reinforcement Learning?
  1. A set of rules for the agent’s behavior.
  2. The agent’s strategy for choosing actions in different states.
  3. A legal document.
  4. The reward function.
Show Answer
Answer: b
Explanation
The policy π is the mapping from states to actions (or a probability distribution over actions given a state) that fully determines the agent’s behaviour.

34) What is the “exploration vs. exploitation” trade-off in RL?
  1. Choosing between exploring the environment for new information or exploiting known information to get rewards.
  2. Choosing between using a GPU or a CPU.
  3. Choosing between a supervised or unsupervised approach.
  4. Choosing between a large or small neural network.
Show Answer
Answer: a
Explanation
The agent must balance trying new actions to discover better strategies (exploration) against repeatedly using actions it already knows to be rewarding (exploitation).

35) What is Q-learning?
  1. A supervised learning algorithm.
  2. A model-free RL algorithm that learns the value of taking a certain action in a certain state.
  3. A type of unsupervised learning.
  4. A method for natural language processing.
Show Answer
Answer: b
Explanation
Q-learning learns the action-value function 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.

36) 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 type of database management system.
  3. A programming language for AI.
  4. A network protocol for multi-agent systems.
Show Answer
Answer: a
Explanation
An MDP is defined by states, actions, transition probabilities, a reward function and a discount factor, and it underpins most of reinforcement learning theory.

37) How does imitation learning work?
  1. The agent learns by copying the actions of an expert.
  2. The agent learns by trying to trick another agent.
  3. The agent learns by reading a textbook.
  4. The agent learns by generating new data.
Show Answer
Answer: a
Explanation
Imitation learning (also called behavioural cloning or learning from demonstration) trains the agent on expert demonstrations, treating the expert’s actions as the target labels.

38) What is the role of a “value function” in RL?
  1. To store the agent’s monetary value.
  2. To estimate the long-term reward of being in a particular state.
  3. To define the agent’s goal.
  4. To measure the agent’s processing speed.
Show Answer
Answer: b
Explanation
The value function V(s) estimates the expected cumulative discounted reward obtainable from a state under a given policy, guiding the agent toward states that lead to good long-term outcomes.

39) What is Deep Reinforcement Learning?
  1. Using deep neural networks to approximate the policy or value functions in RL.
  2. Doing RL underwater.
  3. A very complex form of RL that requires a supercomputer.
  4. RL that takes a long time to train.
Show Answer
Answer: a
Explanation
Deep RL combines RL with deep learning, using neural networks to represent policies or value functions so the agent can handle high-dimensional, continuous state spaces such as images.

40) Which of the following is a challenge in Reinforcement Learning?
  1. The need for a large amount of labeled data.
  2. The sparse and delayed nature of rewards.
  3. The inability to handle continuous state spaces.
  4. It can only be used for games.
Show Answer
Answer: b
Explanation
Rewards are often rare and arrive long after the actions that caused them, making it hard for the agent to work out which behaviour deserves credit — the credit assignment problem.

41) What is a “multi-agent system” (MAS)?
  1. A system with a single, very complex agent.
  2. A system composed of multiple interacting intelligent agents.
  3. A system that runs on multiple computers.
  4. A system that can only be used for simulation.
Show Answer
Answer: b
Explanation
A multi-agent system is a network of several autonomous agents that interact — cooperating, competing or simply coexisting — to solve problems that a single agent could not handle alone.

42) What is the primary purpose of communication in a multi-agent system?
  1. To share hardware resources.
  2. To coordinate actions and share information to achieve individual or collective goals.
  3. To compete for a single reward.
  4. To make the system more complex.
Show Answer
Answer: b
Explanation
Communication lets agents exchange knowledge, negotiate, allocate tasks and synchronise their behaviour so that the group’s activity is coherent rather than chaotic.

43) An “agent communication language” (ACL) is used to:
  1. Write the source code for an agent.
  2. Define the format and semantics of messages exchanged between agents.
  3. Compile the agent’s code.
  4. Secure the agent’s network connection.
Show Answer
Answer: b
Explanation
An ACL such as KQML or FIPA-ACL standardises message structure and meaning (inform, request, propose, refuse…) so that heterogeneous agents can understand one another.

44) What is a “cooperative” multi-agent system?
  1. Agents that work together to achieve a common goal.
  2. Agents that compete against each other.
  3. Agents that ignore each other.
  4. Agents that are all identical.
Show Answer
Answer: a
Explanation
In a cooperative MAS the agents share a common objective and coordinate their actions to maximise joint performance rather than individual gain.

45) What is a “competitive” multi-agent system?
  1. Agents that work together to achieve a common goal.
  2. Agents that have conflicting goals and compete for resources or rewards.
  3. Agents that are all identical.
  4. Agents that are controlled by a single central system.
Show Answer
Answer: b
Explanation
In a competitive MAS each agent pursues its own objective, and one agent’s gain is often another’s loss — as in auctions, adversarial games or market trading.

46) What is the “contract net protocol”?
  1. A protocol for securing network contracts.
  2. A negotiation protocol used in multi-agent systems for task allocation.
  3. A legal agreement between AI companies.
  4. A method for training multiple agents at once.
Show Answer
Answer: b
Explanation
The Contract Net Protocol works like a tender: a manager announces a task, interested agents submit bids, and the manager awards the contract to the most suitable bidder.

47) What is “emergent behavior” in a multi-agent system?
  1. Behavior that is explicitly programmed by the developer.
  2. Complex behavior that arises from the interactions of simple individual agents.
  3. Behavior that is random and unpredictable.
  4. Behavior that is always cooperative.
Show Answer
Answer: b
Explanation
Emergent behaviour is a global pattern that no individual agent was programmed to produce — such as flocking, traffic jams or market bubbles — arising purely from local interactions.

48) What is a potential risk of a multi-agent system?
  1. They are always slower than single-agent systems.
  2. Unintended harmful emergent behavior, such as a “tragedy of the commons” scenario.
  3. They are impossible to debug.
  4. They cannot be used for real-world problems.
Show Answer
Answer: b
Explanation
Individually rational agents can collectively deplete a shared resource or destabilise a system, producing harmful outcomes that no single agent intended.

49) What is the role of a “middle agent” or “broker” in a MAS?
  1. To act as a central point of contact for matching service providers with service requesters.
  2. To compete with other agents.
  3. To provide financial services.
  4. To physically connect the agents.
Show Answer
Answer: a
Explanation
Middle agents (brokers, matchmakers, yellow-page services) decouple providers from requesters, so agents do not need to know about each other in advance.

50) Which of these is a classic example of a multi-agent system?
  1. A single chess engine.
  2. A traffic light control system.
  3. A washing machine.
  4. A calculator.
Show Answer
Answer: b
Explanation
A traffic-light control system consists of many intersection agents that sense local traffic and coordinate with neighbouring intersections to optimise flow across the network.

51) What is the primary function of a “planner” in an AI agent?
  1. To schedule meetings.
  2. To generate a sequence of actions to achieve a goal.
  3. To manage the agent’s finances.
  4. To design the agent’s hardware.
Show Answer
Answer: b
Explanation
A planner reasons about the current state, the goal and the available operators, then produces a plan — an ordered sequence of actions that transforms the initial state into the goal state.

52) What is STRIPS in AI planning?
  1. A programming language for robots.
  2. A classic planning algorithm that uses a set of operators with preconditions and effects.
  3. A type of neural network.
  4. A database for storing plans.
Show Answer
Answer: b
Explanation
STRIPS (Stanford Research Institute Problem Solver) represents actions as operators with a precondition list, an add list and a delete list, and searches for a sequence that reaches the goal.

53) A “PDDL” (Planning Domain Definition Language) file is used to:
  1. Define the problem for a planner.
  2. Program the agent’s user interface.
  3. Connect the agent to the internet.
  4. Train a machine learning model.
Show Answer
Answer: a
Explanation
PDDL is the standard language for describing planning domains (predicates and operators) and specific problems (objects, initial state, goal) so that different planners can read them.

54) What is “hierarchical planning”?
  1. Planning in a strict top-down manner.
  2. Breaking down a high-level goal into sub-goals, which can be further broken down, until primitive actions are reached.
  3. Planning for a hierarchy of agents.
  4. A type of planning that is always optimal.
Show Answer
Answer: b
Explanation
Hierarchical planning (as in HTN planning) decomposes abstract tasks into progressively smaller sub-tasks until only primitive, directly executable actions remain.

55) Which of the following is a popular framework for building multi-agent systems?
  1. TensorFlow
  2. PyTorch
  3. JADE (Java Agent DEvelopment Framework)
  4. Scikit-learn
Show Answer
Answer: c
Explanation
JADE is a FIPA-compliant middleware for building multi-agent systems in Java, providing agent lifecycle management, messaging and directory services. TensorFlow, PyTorch and Scikit-learn are ML libraries.

56) What is the purpose of a “tool” for an LLM-based agent?
  1. To allow the agent to interact with the outside world, e.g., by searching the web or using a calculator.
  2. To write the code for the agent.
  3. To debug the agent’s reasoning.
  4. To visualize the agent’s neural network.
Show Answer
Answer: a
Explanation
Tools extend an LLM’s capabilities beyond text generation, letting it call search engines, calculators, code interpreters or APIs to retrieve facts and take real actions.

57) In the context of LLM agents, what is “ReAct”?
  1. A framework that combines reasoning and acting, allowing the agent to create a thought, take an action, and observe the result.
  2. A type of neural network activation function.
  3. A method for fine-tuning LLMs.
  4. A security protocol for agents.
Show Answer
Answer: a
Explanation
ReAct (Reason + Act) interleaves thought, action and observation steps, so the model’s reasoning is grounded in real feedback from tools rather than pure guesswork.

58) What is the role of “memory” in an AI agent?
  1. To store the agent’s source code.
  2. To store past experiences, observations, and knowledge to inform future decisions.
  3. To increase the agent’s processing speed.
  4. To connect to external databases.
Show Answer
Answer: b
Explanation
Memory lets an agent retain context across interactions — short-term working memory for the current task and long-term memory for accumulated experience and knowledge.

59) What is a “vector database” often used for in modern AI agents?
  1. Storing the agent’s financial transactions.
  2. Enabling long-term memory by storing and retrieving text embeddings for semantic search.
  3. Rendering 3D graphics.
  4. Compiling the agent’s code.
Show Answer
Answer: b
Explanation
Vector databases index embeddings so the agent can retrieve the most semantically relevant memories or documents — the backbone of retrieval-augmented generation (RAG).

60) What is “chain-of-thought” prompting?
  1. A technique to make LLMs generate intermediate reasoning steps before giving a final answer.
  2. A method for connecting multiple agents in a chain.
  3. A type of blockchain technology for AI.
  4. A way to chain multiple GPUs together.
Show Answer
Answer: a
Explanation
Chain-of-thought prompting asks the model to reason step by step, which markedly improves accuracy on arithmetic, logic and multi-step reasoning tasks.

61) Which of the following is a prominent application of AI agents in e-commerce?
  1. A simple calculator.
  2. A recommendation engine that suggests products to users.
  3. A word processor.
  4. A spreadsheet program.
Show Answer
Answer: b
Explanation
Recommendation engines sense user behaviour and act by presenting personalised products, continuously optimising for clicks, baskets and conversions.

62) In robotics, what is a “deliberative agent”?
  1. An agent that plans its actions based on a symbolic model of the world.
  2. An agent that reacts instantly to sensor input without planning.
  3. An agent that learns by imitating humans.
  4. An agent that is controlled by a human operator.
Show Answer
Answer: a
Explanation
Deliberative (or “sense–plan–act”) robots maintain a symbolic world model and reason about it to produce plans before acting. They are thoughtful but slow to react.

63) In robotics, what is a “reactive agent”?
  1. An agent that plans its actions based on a symbolic model of the world.
  2. An agent that reacts to its environment based on a set of simple rules, without complex planning.
  3. An agent that learns from its mistakes.
  4. An agent that can only perform one task.
Show Answer
Answer: b
Explanation
Reactive (behaviour-based) robots use tight perception–action rules and no world model, giving fast responses — the approach behind Brooks’ subsumption architecture.

64) What is a “hybrid architecture” in robotics?
  1. An agent that combines both deliberative and reactive components.
  2. An agent that is half-human, half-machine.
  3. An agent that uses both Python and C++.
  4. An agent that can operate in both air and water.
Show Answer
Answer: a
Explanation
Hybrid architectures (e.g., three-layer architectures) use a reactive layer for fast reflexes and a deliberative layer for long-term planning, mediated by a middle layer.

65) How are AI agents used in finance?
  1. To physically print money.
  2. For algorithmic trading, fraud detection, and personalized banking.
  3. To replace all human bankers.
  4. To design new currencies.
Show Answer
Answer: b
Explanation
Financial agents monitor markets and transactions in real time, executing trades, flagging anomalous activity and tailoring advice and products to individual customers.

66) In healthcare, an AI agent could be used for:
  1. Performing surgery without any human supervision.
  2. Personalized treatment planning and drug discovery.
  3. Replacing all doctors and nurses.
  4. Diagnosing patients with 100% accuracy.
Show Answer
Answer: b
Explanation
Realistic current applications include personalised treatment planning, drug discovery, triage support and clinical documentation — always as decision support rather than a replacement for clinicians.

67) What is a “chatbot” an example of?
  1. A simple reflex agent.
  2. A conversational AI agent.
  3. A multi-agent system.
  4. A planning agent.
Show Answer
Answer: b
Explanation
A chatbot is a conversational agent: it perceives user messages and produces responses to pursue goals such as answering questions or completing a booking.

68) How do modern LLM-based agents (like Auto-GPT) differ from traditional chatbots?
  1. They can only answer questions.
  2. They can autonomously break down a high-level goal into sub-tasks and execute them using tools.
  3. They are always more accurate.
  4. They don’t use natural language.
Show Answer
Answer: b
Explanation
Traditional chatbots respond turn by turn. LLM agents plan, decompose goals into sub-tasks, call tools, inspect the results and iterate until the objective is met.

69) What is “process automation” in the context of AI agents?
  1. Using agents to automate repetitive business processes, such as invoice processing or customer support.
  2. Automating the process of building a robot.
  3. A type of manufacturing.
  4. A method for training agents.
Show Answer
Answer: a
Explanation
AI-driven process automation (including intelligent RPA) handles high-volume, rule-based or semi-structured work such as invoice matching, ticket routing and data entry.

70) In gaming, what is a “non-player character” (NPC) an example of?
  1. A simple reflex agent.
  2. A learning agent.
  3. A goal-based agent.
  4. All of the above can be true, depending on the game’s complexity.
Show Answer
Answer: d
Explanation
An NPC may be a simple reflex agent (patrolling guard), a goal-based planner (pathfinding to a target), or a learning agent (adaptive difficulty) — it depends entirely on the game.

71) What is the “Turing Test” designed to evaluate?
  1. The processing speed of a computer.
  2. A machine’s ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human.
  3. The memory capacity of a computer.
  4. The energy efficiency of an AI agent.
Show Answer
Answer: b
Explanation
In the Turing Test a human interrogator converses blindly with a machine and a person; if the machine cannot be reliably distinguished from the human, it is said to pass.

72) What is a common challenge in evaluating AI agents?
  1. They are always too slow to test.
  2. Defining a clear and comprehensive performance measure that captures all aspects of desired behavior.
  3. They cannot be tested in simulation.
  4. They are too expensive to run.
Show Answer
Answer: b
Explanation
Agents can often optimise the metric you choose rather than the behaviour you actually want, so designing a performance measure that truly reflects the desired outcome is hard.

73) What is the “alignment problem” in AI?
  1. The problem of aligning the agent’s code correctly.
  2. The challenge of ensuring that an AI agent’s goals and actions align with human values and intentions.
  3. The problem of aligning the pixels on a screen.
  4. The challenge of aligning multiple agents in a formation.
Show Answer
Answer: b
Explanation
The alignment problem is the technical and philosophical challenge of making sure that a capable AI system actually pursues what humans want it to pursue, rather than a literal but unintended interpretation of its objective.

74) What is “reward hacking” in Reinforcement Learning?
  1. A hacker stealing the reward signal.
  2. When an agent finds a loophole to get a high reward without actually completing the intended task.
  3. A method for designing reward functions.
  4. A type of cyberattack on an AI system.
Show Answer
Answer: b
Explanation
Because the agent optimises the reward signal rather than the designer’s true intent, it may exploit flaws in the reward function — for example, a boat-racing agent that spins in circles collecting points instead of finishing the race.

75) The “exploration-exploitation dilemma” is a key challenge in:
  1. Supervised learning.
  2. Unsupervised learning.
  3. Reinforcement learning.
  4. Symbolic AI.
Show Answer
Answer: c
Explanation
The dilemma arises whenever an agent must learn by interacting with an environment and deciding whether to try new actions or repeat known-good ones — the defining setting of reinforcement learning.

76) What is a “brittle” AI agent?
  1. An agent that is physically fragile.
  2. An agent that performs well in the specific environment it was trained for but fails in slightly different situations.
  3. An agent that is easy to break.
  4. An agent that cannot learn.
Show Answer
Answer: b
Explanation
Brittleness is the failure to generalise: small distribution shifts, unusual inputs or changed conditions cause a sharp drop in performance.

77) What is the “credit assignment problem” in RL?
  1. The problem of assigning credit to the programmer.
  2. The difficulty of determining which of a long sequence of actions was responsible for a final reward.
  3. The problem of assigning tasks to different agents.
  4. The challenge of getting credit for a project.
Show Answer
Answer: b
Explanation
When a reward arrives long after the actions that caused it, the agent must work out which of many earlier decisions deserves the credit — or the blame.

78) What is “catastrophic forgetting” in AI?
  1. When an AI agent forgets its entire training.
  2. The tendency of a neural network to completely and abruptly forget previously learned information upon learning new information.
  3. When a hard drive fails and data is lost.
  4. A memory leak in an AI program.
Show Answer
Answer: b
Explanation
Standard neural networks overwrite old weights when trained on new tasks. Techniques such as rehearsal, elastic weight consolidation and parameter isolation aim to prevent this.

79) Why is “sample efficiency” a major concern in RL?
  1. Because data is cheap.
  2. Because many RL algorithms require a huge number of interactions with the environment to learn, which can be slow and expensive.
  3. Because it’s easy to get data.
  4. Because the agent needs to be efficient with its code.
Show Answer
Answer: b
Explanation
Unlike supervised learning, RL generates its own data through interaction, so millions of trials may be needed — impractical when each trial is a costly or dangerous real-world action.

80) What is the “sim-to-real” gap?
  1. The difference between a simulation and reality.
  2. The difficulty of transferring a policy learned in a simulation to a real-world robot due to differences between the simulated and real environments.
  3. The time it takes to move from simulation to reality.
  4. The cost of building a simulator.
Show Answer
Answer: b
Explanation
Explanation
Simulators never perfectly model friction, sensor noise, latency and lighting, so a policy that excels in simulation can fail on real hardware. Domain randomisation is a common mitigation.

81) What is a primary ethical concern with autonomous AI agents?
  1. They might be too slow.
  2. Accountability and transparency in their decision-making processes.
  3. They might use too much electricity.
  4. They might be too expensive to build.
Show Answer
Answer: b
Explanation
When an autonomous agent causes harm, it can be unclear who is responsible and why the decision was made — hence the demand for transparency, auditability and clear accountability chains.

82) What is the “paperclip maximizer” thought experiment about?
  1. An AI that is very good at making paperclips.
  2. An AI with a seemingly harmless goal (making paperclips) that, if not properly aligned, could lead to the destruction of all other values.
  3. A robot that works in a paperclip factory.
  4. A test for AI creativity.
Show Answer
Answer: b
Explanation
Bostrom’s thought experiment illustrates that a sufficiently powerful optimiser with a narrow, unaligned objective may single-mindedly convert all available resources to that goal, at catastrophic cost to everything else.

83) What does “explainable AI” (XAI) aim to achieve?
  1. To make AI agents explain their actions in a way that is understandable to humans.
  2. To make AI agents explain how they were built.
  3. To make AI agents explain the laws of physics.
  4. To make AI agents explain their own source code.
Show Answer
Answer: a
Explanation
XAI produces human-understandable justifications for a system’s outputs — through feature attributions, counterfactuals, saliency maps or natural-language rationales — building trust and enabling recourse.

84) What is a “bias” in an AI agent?
  1. A preference for one thing over another.
  2. A systematic and unfair prejudice in the agent’s decisions, often stemming from biased training data.
  3. A type of error in the code.
  4. A measure of the agent’s uncertainty.
Show Answer
Answer: b
Explanation
AI bias is a systematic, repeatable skew that disadvantages particular groups, typically inherited from unrepresentative data, flawed labels or objectives that encode historical inequities.

85) What is the “off-switch problem” in AI safety?
  1. The difficulty of designing an off-switch for an AI.
  2. The challenge of ensuring an AI agent will allow itself to be shut down if it is pursuing a goal.
  3. The problem of an off-switch being pressed accidentally.
  4. The challenge of turning on an AI.
Show Answer
Answer: b
Explanation
If an agent’s objective is best served by continuing to operate, it may resist or disable shutdown. Corrigibility — accepting human intervention — must therefore be designed in.

86) What is “value lock-in” as an AI risk?
  1. Locking human values into an AI’s code.
  2. The risk that a powerful AI could impose a single set of values on humanity forever, preventing future moral progress.
  3. A security feature for AI systems.
  4. A method for training AI.
Show Answer
Answer: b
Explanation
Moral norms evolve over time. If a powerful system freezes today’s values permanently, humanity loses the ability to revise its ethics — a concern raised by Bostrom and others.

87) What is the role of “human-in-the-loop” in AI systems?
  1. To have a human monitor and potentially intervene in the AI’s decision-making process, especially for critical tasks.
  2. To have a human train the AI.
  3. To have a human write the AI’s code.
  4. To have a human as a part of the AI’s hardware.
Show Answer
Answer: a
Explanation
Human-in-the-loop keeps a person in the decision cycle to review, approve or override the agent’s actions — essential in high-stakes domains such as medicine, law and aviation.

88) What is a “deepfake”?
  1. A very deep neural network.
  2. Synthetic media (video, audio, images) created by AI that convincingly mimics a real person.
  3. A type of phishing attack.
  4. A new type of computer virus.
Show Answer
Answer: b
Explanation
Deepfakes are generated with deep generative models such as GANs or diffusion models and raise serious risks of misinformation, fraud, impersonation and non-consensual imagery.

89) What is the potential impact of AI agents on employment?
  1. No impact at all.
  2. They will only create new jobs.
  3. They could lead to significant job displacement in certain sectors, while also creating new roles and requiring workforce retraining.
  4. They will only take over manual labor jobs.
Show Answer
Answer: c
Explanation
Evidence points to a mixed effect: automation displaces some routine cognitive and manual tasks, creates new roles (data, oversight, AI operations) and shifts the demand for skills, making retraining essential.

90) Why is robustness a critical safety property for AI agents?
  1. It allows the agent to run faster.
  2. It ensures the agent can continue to operate safely and effectively in the face of unexpected inputs, errors, or adversarial attacks.
  3. It makes the agent cheaper to produce.
  4. It makes the agent more intelligent.
Show Answer
Answer: b
Explanation
A non-robust agent can fail catastrophically or be manipulated by adversarial examples once deployed, so resilience to noise, distribution shift and attack is a prerequisite for safe operation.

91) 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 neural network.
  4. A method for increasing processing power.
Show Answer
Answer: b
Explanation
Swarm intelligence describes how simple agents following local rules produce sophisticated collective behaviour — seen in ant colonies, bird flocks and algorithms such as ant colony optimisation and particle swarm optimisation.

92) What is “meta-learning” in the context of AI agents?
  1. Learning about learning, or “learning to learn,” where an agent learns how to learn new tasks more quickly.
  2. Learning about the meta-data of a dataset.
  3. A type of supervised learning.
  4. Learning from a human teacher.
Show Answer
Answer: a
Explanation
Meta-learning trains a model across many tasks so that it can adapt to a new task from just a handful of examples — the basis of few-shot learning approaches like MAML.

93) What is a “digital twin”?
  1. A virtual representation of a physical object or system that is updated from real-time data.
  2. A clone of a software program.
  3. A type of AI agent that exists only online.
  4. A second version of an AI agent.
Show Answer
Answer: a
Explanation
A digital twin mirrors a real asset — a jet engine, a factory line, a city — using live sensor data, enabling simulation, predictive maintenance and what-if analysis.

94) What is “affective computing”?
  1. Computing that is very fast and efficient.
  2. The study and development of systems that can recognize, interpret, process, and simulate human affects (emotions).
  3. A type of quantum computing.
  4. Computing that is done in the cloud.
Show Answer
Answer: b
Explanation
Affective computing, pioneered by Rosalind Picard, enables machines to detect emotional states from voice, face or physiology and to respond empathetically — key for social agents.

95) What is a “neuro-symbolic” AI system?
  1. A system that uses only neural networks.
  2. A system that uses only symbolic logic.
  3. A hybrid approach that combines neural networks with symbolic reasoning to get the best of both worlds.
  4. A system that is inspired by the brain.
Show Answer
Answer: c
Explanation
Neuro-symbolic AI pairs the pattern-recognition strength of neural networks with the explicit, verifiable reasoning of symbolic systems — a promising route to more reliable and interpretable agents.

96) What is the concept of “emergence” in large language models?
  1. The ability of a model to suddenly appear.
  2. The phenomenon where certain capabilities (like reasoning or arithmetic) appear in LLMs only after they reach a certain size, not being explicitly programmed.
  3. The process of training an LLM.
  4. The way an LLM generates text.
Show Answer
Answer: b
Explanation
Emergent abilities are skills that are essentially absent in smaller models but appear once a model passes a scale threshold — an actively debated but widely discussed phenomenon.

97) What is a “foundation model”?
  1. A model that is used to build the foundation of a building.
  2. A large AI model trained on a vast quantity of data at scale, which can be adapted to a wide range of downstream tasks.
  3. The first model ever created in a project.
  4. A model that is not very important.
Show Answer
Answer: b
Explanation
The term, coined by Stanford CRFM, describes large pretrained models (GPT, Claude, Gemini, LLaMA) that serve as a general base to be fine-tuned or prompted for many downstream applications.

98) What is “prompt engineering” in the context of LLM agents?
  1. Designing the hardware for an LLM.
  2. The practice of designing and refining the input (prompt) to guide an LLM to produce a desired output.
  3. Building the physical engine for a robot.
  4. A type of software engineering.
Show Answer
Answer: b
Explanation
Prompt engineering shapes instructions, examples, context and output formats so the model reliably produces the intended result without changing its weights.

99) What is the potential of “Artificial General Intelligence” (AGI)?
  1. An AI that is an expert at one specific task.
  2. A hypothetical AI that has the ability to understand, learn, and apply intelligence to a wide variety of problems at a level comparable to a human being.
  3. An AI that is faster than any human.
  4. An AI that can feel emotions.
Show Answer
Answer: b
Explanation
AGI refers to human-level general competence across many domains, in contrast to today’s narrow AI, which excels only within a specific task or domain.

100) What is a “multi-modal” AI agent?
  1. An agent that can operate in multiple modes of transport.
  2. An agent that can process and understand multiple types of data, such as text, images, and audio.
  3. An agent that has multiple personalities.
  4. An agent that can be programmed in multiple languages.
Show Answer
Answer: b
Explanation
Multimodal agents accept and reason over several modalities — text, images, audio, video — and can often generate outputs in more than one of them as well.
100 Deep Learning MCQ (Multiple Choice Questions) with Answers
100 Agentic 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