Fundamentals of AI and ML practice questions

Agentic AI extends generative AI so a system can plan and carry out multi-step tasks with tools.

Agentic AI is best understood as generative AI that acts: it reasons with a generative model, remembers context, and calls tools to complete a goal, which matches option A. Rule-based logic, regression, and data labeling are different concepts the distractors use as look-alikes.

1. Recognize that agentic AI is built on top of generative AI.
2. Note the added ingredients: memory, tool use, and multi-step planning toward a goal.
3. Distinguish it from a system that merely answers one prompt.

Exam tip: Agentic AI is generative AI that plans and takes action using tools.

An algorithm is the learning method; a model is the trained result it produces.

The trained artifact that captures learned patterns and makes predictions is the model, so option C is correct. The algorithm is the recipe rather than the cake, a dataset is the raw input, and a feature is one input variable, which is why the other options do not fit.

1. Identify the algorithm as the procedure that learns from data.
2. Identify the model as the output of training that procedure.
3. Match the trained, prediction-making artifact to the term model.

Exam tip: The algorithm is the recipe; the model is the finished result.

Training builds a model; inferencing uses the finished model on new data.

Scoring a new, unseen transaction with an already-trained model is inferencing, so option B is correct. Training happens beforehand, while labeling and feature engineering are data-preparation steps, which is why the distractors do not match the moment described.

1. Check whether the model is still learning or already finished.
2. A finished model producing a prediction on new input is inferencing.
3. Reserve training for the earlier learning phase.

Exam tip: Making predictions on new input with a finished model is inferencing.

Supervised learning trains on labeled data where each example includes the correct answer.

Because every email is already labeled spam or not-spam and the goal is to predict that label, this is supervised learning, making option C correct. Unsupervised learning and clustering require unlabeled data, and reinforcement learning learns from rewards, so none of them fit labeled examples.

1. Ask whether the training examples carry correct answers.
2. Labeled examples with known outcomes point to supervised learning.
3. Rule out unsupervised and reinforcement options, which do not use a fixed labeled dataset.

Exam tip: Labeled examples with known answers signal supervised learning.

Unsupervised learning finds structure in unlabeled data with no correct answers provided.

With no labels available and the goal of discovering unusual patterns, the model must find structure by itself, which is unsupervised learning and makes option A correct. Supervised learning would require labels, reinforcement learning would require an environment and rewards, and transfer learning describes reusing a model, so the distractors do not apply.

1. Confirm the data has no labels or known answers.
2. Match learning that discovers structure on its own to unsupervised learning.
3. Eliminate options that require labels, rewards, or a pre-trained source model.

Exam tip: No labels plus finding hidden structure means unsupervised learning.

Reinforcement learning learns by trial and error using rewards and penalties from an environment.

The presence of rewards, penalties, and trial-and-error interaction points squarely to reinforcement learning, so option D is correct. Supervised and unsupervised learning use datasets rather than an environment with rewards, and batch is a processing style, not a learning type.

1. Look for the trigger words reward, penalty, and trial and error.
2. Match learning by interacting with an environment to reinforcement learning.
3. Rule out dataset-based and processing-based options.

Exam tip: Learning by rewards through trial and error is reinforcement learning.