AI Associate AI Fundamentals Practice Test 10 — 15 Questions

Question 1

A company wants to automatically extract key information like order numbers and dates from customer emails. Which NLP technique should be used?

Accepted Answer

Entity extraction. Entity extraction (named entity recognition) identifies and pulls structured elements such as order numbers, dates, names, and amounts out of unstructured text. Since the goal is to automatically extract specific key fields from customer emails, entity extraction is the precise NLP technique that returns those labeled spans.

Answer

Sentiment analysis

Answer

Text classification

Answer

Text summarization

Question 2

What is the term for when an AI model produces confident but incorrect information, often in generative AI?

Accepted Answer

Hallucination. Hallucination refers to the phenomenon where a generative AI model produces output that is fluent and confident but factually incorrect or fabricated. This is distinct from bias (systematic skew in outputs), overfitting (model memorizes training data), and underfitting (model fails to capture patterns). Option C is correct.

Answer

Bias

Answer

Overfitting

Answer

Underfitting

Question 3

A company wants to use AI to analyze customer feedback from surveys and social media. Which TWO capabilities are most relevant?

Accepted Answer

Sentiment analysis. Sentiment analysis (B) is correct because it is the natural language processing capability specifically designed to determine the emotional tone (positive, negative, neutral) of customer feedback from surveys and social media, which is exactly what the company wants to analyze. Text classification (D) is also correct because it categorizes unstructured text into predefined topics or labels (e.g., complaints, praise, product issues), enabling systematic analysis of large volumes of survey and social media text. Speech recognition (A) is not relevant here because it converts spoken audio to text and the scenario involves written survey responses and social media posts, not voice data. Image recognition (C) is not relevant because it analyzes visual content such as photos or videos, which is outside the scope of text-based customer feedback. Predictive lead scoring (E) is not relevant because it ranks sales prospects by likelihood to convert, which is a sales/marketing task rather than analyzing customer feedback sentiment or topics.

Answer

Speech recognition

Answer

Image recognition

Answer

Predictive lead scoring

Question 4

A data scientist is building a churn prediction model. What THREE factors are most critical for model success?

Accepted Answer

Relevant features correlated with churn. Option B is correct because feature relevance drives predictive power: features that genuinely correlate with churn (e.g., tenure, usage decline, support tickets) give the model signal, whereas irrelevant features add noise and risk overfitting. Option C is correct because a representative dataset covering all customer segments ensures the model generalizes across the population and avoids sampling bias that would make predictions unreliable for underrepresented groups. Option D is correct because supervised churn prediction depends on accurate, consistent historical labels; mislabeled or dirty data teaches the model wrong patterns regardless of algorithm choice. Option A is not critical—more parameters increase capacity and overfitting risk without guaranteeing better performance, and simpler models often suffice. Option E is not inherently critical—dataset size helps only when data is relevant, representative, and well-labeled; large but biased or noisy data can degrade results.

Answer

High number of model parameters

Answer

Large dataset size

Question 5

Which TWO statements correctly describe predictive AI compared to generative AI?

Accepted Answer

Predictive AI is used for forecasting numeric outcomes like churn probability. Option B is correct because predictive AI models (e.g., regression, classification, gradient boosting) are trained on historical data to output numeric or categorical forecasts such as churn probability, lead scores, or sales estimates. Option C is correct because generative AI (e.g., LLMs, diffusion models, GANs) is designed to create novel content — text, images, audio, or code — rather than predict a labeled outcome. Option A is wrong because lead scoring is a classic predictive classification task, not a generative one. Option D is wrong because the two paradigms use fundamentally different objectives and architectures (discriminative/predictive models vs. generative models like transformers or diffusion networks). Option E is wrong because producing realistic customer service responses is a generative AI capability, not a predictive one.

Answer

Generative AI is best suited for lead scoring

Answer

Both predictive and generative AI rely on the same algorithms

Answer

Predictive AI can generate realistic customer service responses

Question 6

Which type of machine learning is used when a model is trained on historical sales data that includes both input features and the known outcome (e.g., closed won/lost) to predict whether a new lead will convert?

Accepted Answer

Supervised learning. Supervised learning is correct because the training dataset contains labeled examples — historical sales records where each row pairs input features (lead attributes) with a known outcome (closed won/lost). The model learns a mapping function from features to labels, then applies it to new, unlabeled leads to predict conversion. This labeled-input-to-known-output pattern is the defining characteristic of supervised classification.

Answer

Semi-supervised learning

Answer

Reinforcement learning

Answer

Unsupervised learning

Question 7

A sales operations team wants to automatically categorize incoming support cases into predefined categories (e.g., Billing, Technical, General). The team has thousands of historical cases with correct category labels. Which AI approach should they use?

Accepted Answer

Text classification. Text classification is a supervised learning approach where a model is trained on labeled historical cases to predict predefined categories. With thousands of labeled examples, this is the ideal approach for automatically categorizing incoming support cases.

Answer

Topic modeling (unsupervised)

Answer

Sentiment analysis

Answer

Named entity recognition

Question 8

A data scientist trains a churn prediction model on CRM data that includes customer tenure, support ticket count, and last purchase date. The model achieves 95% accuracy on training data but only 60% on a holdout validation set. What is the most likely issue?

Accepted Answer

The model is overfitting the training data. A large gap between high training accuracy (95%) and much lower validation accuracy (60%) is the textbook signature of overfitting: the model has memorized training-specific noise rather than learning generalizable patterns. The fix typically involves regularization, more data, or simpler models. Underfitting would show low accuracy on both sets.

Answer

The validation set contains data leakage

Answer

The model needs more features

Answer

The model is underfitting the training data

Question 9

A company wants to use AI to automatically extract key information (e.g., invoice number, date, total amount) from scanned PDF invoices. Which AI capability should they use?

Accepted Answer

Computer vision with optical character recognition (OCR). Computer vision with OCR is the correct capability because it combines image processing to detect and extract text from scanned documents (invoices) with optical character recognition to convert that text into machine-readable form, enabling extraction of fields like invoice number, date, and total amount. This is a classic document AI use case.

Answer

Generative AI for text generation

Answer

Predictive analytics

Answer

Sentiment analysis

Question 10

A marketing team wants to recommend products to customers based on their past purchases and browsing behavior. Which type of AI is most appropriate?

Accepted Answer

Predictive AI. Product recommendation is a predictive AI task that uses historical data to predict which products a customer is likely to purchase.

Answer

Generative AI

Answer

Reinforcement learning

Answer

Unsupervised learning

Question 11

A customer service chatbot misinterprets user requests and often provides irrelevant answers. The development team wants to improve the chatbot's understanding of user intent. Which NLP component should they focus on?

Accepted Answer

Intent detection. Intent detection is the NLP component responsible for classifying a user's utterance into a predefined intent, such as 'check balance' or 'reset password.' If the chatbot misinterprets requests and gives irrelevant answers, the intent classification model is likely underperforming, so improving intent detection directly addresses the problem.

Answer

Text generation

Answer

Sentiment analysis

Answer

Named entity recognition

Question 12

A predictive model for lead scoring shows high precision but low recall. Which business impact is most likely?

Accepted Answer

Many high-quality leads are not followed up. High precision means that when the model predicts a lead is high-quality, it is usually correct; low recall means it misses many actual high-quality leads. The business impact is that many genuinely high-quality leads are never flagged for follow-up, resulting in lost revenue opportunities. This is the classic precision-recall trade-off: the model is conservative, so it only flags a small subset of positives.

Answer

The model is overfitted to training data

Answer

Sales team spends time on many leads that never convert

Answer

The model has high bias

Question 13

Which type of AI is designed to perform only a specific task, such as playing chess or recommending products?

Accepted Answer

Narrow AI. Narrow AI (weak AI) is specialized for one task and cannot generalize beyond its training.

Answer

Super AI

Answer

General AI

Answer

Strong AI

Question 14

A company uses an AI model to predict customer churn. The model's predictions are used to automatically assign discounts to high-risk customers. A customer complains about receiving a discount offer they did not request. Which ethical concern is most relevant?

Accepted Answer

Lack of transparency of AI decisions to customers. The customer received an automated discount they did not request and was not informed about, which directly implicates transparency — the customer cannot see or understand why the AI made that decision. Lack of transparency (explainability) is the most relevant ethical concern in this scenario.

Answer

Bias in training data

Answer

Model overfitting

Answer

Data quality

Question 15

A sales director wants to use AI to prioritize leads that are most likely to convert. The company has historical data on leads that includes whether they converted (yes/no) and various attributes. Which machine learning type should be used?

Accepted Answer

Supervised learning (classification). The problem is a binary outcome (converted: yes/no) predicted from historical labeled data with various attributes. This is a classic supervised learning classification task: the model learns from labeled examples (leads with known conversion outcomes) to predict the class of new leads. Classification is the correct ML type for predicting discrete categories.

Answer

Generative AI

Answer

Reinforcement learning

Answer

Unsupervised learning (clustering)