The exam often describes a business situation and expects you to name the kind of AI application behind it. Most real-world uses fall into a small set of families: seeing (computer vision), understanding language (natural language processing), hearing (speech recognition), suggesting (recommendations), spotting the unusual (fraud and anomaly detection), looking ahead (forecasting), answering from a company's own documents (knowledge bases) and taking actions (agents). This lesson gives you a quick way to recognize each one.
Seeing, Reading and Hearing
Computer vision lets software interpret images and video: detecting defects on a production line, recognizing objects in photos, or flagging unsafe content. On AWS, Amazon Rekognition analyzes images and video, and Amazon Textract extracts text and data from scanned documents.
Natural language processing (NLP) works with written language: finding the sentiment of a review, pulling names and places out of an email, or summarizing a report. Amazon Comprehend is AWS's NLP service.
Speech recognition turns spoken audio into text, for example for call transcripts or video captions. Amazon Transcribe does this. The opposite direction, turning text into speech, is Amazon Polly.
Suggesting and Spotting the Unusual
Recommendation systems suggest products, films or articles based on what a person and similar people have done before. Amazon Personalize is AWS's managed recommendation service.
Fraud detection looks for transactions or behavior that don't fit the normal pattern. It's often built as a classification model trained on labeled fraud cases, or as anomaly detection when labels are rare. Both can be built on Amazon SageMaker AI.
Looking Ahead
Forecasting predicts future values from historical patterns over time: next month's demand, next week's call volume, or the energy needed tomorrow. Because the data is ordered in time, this is called time-series forecasting. SageMaker AI, including the no-code SageMaker Canvas, can build time-series forecasts.
Knowledge Bases and Agents
Knowledge bases let people ask questions in plain language and get answers drawn from an organization's own documents, such as policies, manuals or wikis. The system retrieves the relevant passages and a generative model writes an answer based on them. Amazon Bedrock Knowledge Bases provides this.
Agentic AI goes a step further: instead of only answering, an agent works out the steps needed to complete a task, calls tools or company systems to carry them out, and checks the results. Rebooking a cancelled trip across flight and hotel systems is an agent's job. Amazon Bedrock Agents and Amazon Bedrock AgentCore support building agents.
On the Exam
Match the verb in the scenario to the family:
| The business wants to... | Application |
|---|---|
| Inspect images or video | Computer vision |
| Understand or extract meaning from text | NLP |
| Turn speech into text | Speech recognition |
| Suggest items to each user | Recommendation system |
| Catch unusual transactions | Fraud or anomaly detection |
| Predict future values over time | Forecasting |
| Answer questions from its own documents | Knowledge base |
| Complete multi-step tasks using its systems | Agentic AI |
The common trap is confusing a knowledge base with an agent. If the system only answers, it's a knowledge base. If it also takes actions in other systems, it's an agent.