CAIC Certification Exam Guide + Practice Questions Updated 2026

Explanation:
Reinforcement learning is the correct answer because the AI agent learns by interacting with an environment and improving its behavior based on rewards and penalties. The goal of reinforcement learning is to learn a policy or strategy that maximizes cumulative reward over time. This differs from supervised learning, where the model learns from labeled input-output examples. It also differs from unsupervised learning, where the model searches for hidden patterns without labels or rewards. Semi-supervised learning is incorrect because the scenario does not involve a mix of labeled and unlabeled data. Regression learning is also incorrect because regression predicts continuous numerical values, while this example focuses on action selection and reward optimization. Therefore, the correct answer is C. reinforcement learning.

Explanation:
The correct answer is E. b and c only because few-shot learning means a model learns or adapts to a new task using only a small number of examples. In generative AI and large language model usage, few-shot prompting often provides a few demonstrations so the model can understand the expected pattern, format, classification logic, or response style. Option B is correct because few-shot learning uses a limited number of examples rather than a large training dataset.
Option C is also correct because few-shot learning depends on the model’s prior knowledge learned during pretraining. The model uses that existing knowledge to generalize from the small set of examples and apply the same logic to new inputs. Option A is not the best statement because “a large number of examples” does not match the idea of few-shot learning. Therefore,
the most correct answer is E. b and c only.

Explanation:
The correct answer is D. a and b only because statements A and B are the most incorrect for retrospective/concurrent risk-scoring methodology. Retrospective/concurrent risk assessment is mainly used to evaluate model risk based on past or present evidence, current model behavior, observed incidents, model performance changes, risk indicators, and investigation findings. It is not primarily a future-prediction method.
Statement A is incorrect because it says retrospective/concurrent methods “predict” model risk after analyzing historical model performance. Historical performance may be reviewed, but retrospective/concurrent risk scoring is more about assessing or investigating past and current risk conditions, not predicting future risk. Statement B is also incorrect because using current model risk to predict overall model risk for future cycles describes prospective risk, not retrospective/concurrent risk. Statement C is correct because retrospective/concurrent review is suitable when there are changes in model behavior, risk indicators, attacks, data loss, or investigation needs. Therefore, the most incorrect statements areA and B only.

Explanation:
Artificial General Intelligence, or AGI, is commonly referred to as Strong AI because it describes an AI system with human-like cognitive ability across many different tasks and domains. Unlike narrow or weak AI, which is designed to perform a specific task such as image recognition, language translation, recommendation, fraud detection, or chatbot response generation, AGI would be able to understand, learn, reason, adapt, and solve problems broadly in a way similar to human intelligence.
Weak AI is incorrect because it refers to task-specific AI systems that operate within limited boundaries. General AI is related in meaning, but the commonly used expression for AGI in AI classification is Strong AI. Super AI is different because it refers to intelligence that would exceed human intelligence, while Expert AI is not the standard term for AGI. Therefore, the correct answer is B. Strong AI.

Explanation:
The correct answer is B. Supervised Learning. Supervised learning is the machine learning approach used when a model is trained with labeled data. Labeled data means each training example includes both the input and the correct output or target label. The model studies these examples and learns the relationship between the input features and the expected result. After training, it can make predictions or classifications on new data.
Unsupervised learning is incorrect because it uses unlabeled data and focuses on finding hidden patterns, clusters, or structures without predefined answers. Reinforcement learning is incorrect because it involves an agent learning through actions, rewards, and penalties in an environment. Semi-supervised learning is also not the best answer because it uses a mix of labeled and unlabeled data. Support Vector refers to part of the Support Vector Machine method, not a learning type by itself. Therefore, the correct learning type for labeled data is B. Supervised Learning.