AI-901
Deep Dive
The AI-901 tests whether you can identify the right type of AI solution for a given scenario. You don't need to know how to implement them — you need to know what each one does and when to apply it.
Contents
Artificial Intelligence is software that imitates human behaviors: learning from experience, understanding language, recognizing images, making predictions, and making decisions.
Key for the exam: AI is not magic — it's statistics and mathematics applied to data. Machine Learning is a subset of AI. Deep Learning is a subset of ML. Generative AI is a subset of Deep Learning.
Hierarchy the exam tests
Artificial Intelligence
The broadest field — all software that imitates human cognition
↳ Machine Learning
Systems that learn from data without being explicitly programmed
↳↳ Deep Learning
ML with deep neural networks — the foundation of modern vision and language
↳↳↳ Generative AI
Deep learning that generates new content: text, images, code
Create a single Azure AI services resource and discover how one key unlocks 15+ AI services.
+1 more...
The AI-901 has "what type of AI would you use for…?" questions — you need to recognize the scenario and map it to the correct type.
Machine Learning
Predicting values or categories based on historical data.
Computer vision
Extracting information from images and video.
NLP
Understanding and generating written or spoken human language.
Generative AI
Creating new content: text, images, code, audio.
Anomaly detection
Identifying unusual patterns in data.
Knowledge Mining
Extracting structured information from unstructured documents.
ML is the broadest type of AI. Instead of programming explicit rules (if price > 1000 then premium), the system learns the rules from the data. You show it thousands of houses with their prices and it learns on its own which factors matter.
🏷️
Supervised learning
You train with labeled data (examples with the correct answer). The model learns to predict the label.
Use cases
Spam classification, price prediction, medical diagnosis
🗂️
Unsupervised learning
You train without labels. The model discovers structure and patterns on its own.
Use cases
Customer segmentation, data compression, anomaly detection
🎮
Reinforcement learning
The model learns by trial and error: it receives rewards for correct actions and penalties for incorrect ones.
Use cases
Games, robotics, route optimization
Image classification
Determining what is in an image. Output: a category.
"This photo contains a dog" (confidence: 97%)
Object detection
Locating and classifying multiple objects within an image. Output: categories + bounding boxes (coordinates).
"There is a dog at [150,80,300,250] and a cat at [400,100,550,280]"
Optical character recognition (OCR)
Extracting text from images, photos of documents, or scanned PDFs.
Digitizing invoices, reading license plates, processing paper forms
Face detection
Detecting the presence of faces in an image. Output: coordinates of each detected face.
Different from face recognition — detecting ≠ identifying who it is
Face recognition
Identifying or verifying the identity of a specific person by comparing against known faces.
Biometric access system, identity verification
Common exam trap
Face detection ≠ Face recognition. Detecting = knowing there is a face (without knowing whose). Recognizing = identifying the person. The exam distinguishes them explicitly.
Sentiment analysis
Determining whether a text is positive, negative, or neutral.
Azure AI Language — Sentiment Analysis
Key phrase extraction
Identifying the most relevant terms in a text.
Azure AI Language — Key Phrase Extraction
Named entity recognition (NER)
Extracting names of people, organizations, places, dates, etc.
Azure AI Language — Named Entity Recognition
Translation
Converting text from one language to another.
Azure AI Translator
Speech recognition (STT)
Converting spoken audio into written text.
Azure AI Speech — Speech to Text
Speech synthesis (TTS)
Converting written text into spoken audio.
Azure AI Speech — Text to Speech
Conversational language understanding (CLU)
Extracting the intent and entities from a user phrase.
Azure AI Language — Conversational Language Understanding
Question answering (QA)
Answering questions based on a knowledge base.
Azure AI Language — Question Answering
Unlike predictive ML (which classifies or predicts existing values), generative AI creates entirely new content: text, images, code, audio, or video that did not exist before.
Text generation
Large language models (LLMs) like GPT-5 generate coherent, contextual text. The foundation of conversational chatbots.
Azure OpenAI in Foundry — GPT-5 family
Image generation
Diffusion models that create photorealistic or artistic images from a text description (prompt).
Azure OpenAI in Foundry — gpt-image models
Code generation
LLMs trained specifically on code that suggest, complete, or generate code snippets in multiple languages.
Azure OpenAI Service + GitHub Copilot
Audio generation
Models that transcribe audio (Whisper) or synthesize speech with natural, emotional characteristics.
Azure OpenAI Service — Whisper
Identifies values that deviate significantly from the expected pattern in time series or datasets. It doesn't predict a label — it predicts whether something is normal or unusual.
💳
Financial fraud
A $5,000 transaction at 3am in an unfamiliar country when the usual pattern is $50 in the local city.
🏭
Industrial monitoring
Turbine temperature rises from 80°C to 140°C — a possible imminent equipment failure.
🔒
Network security
Massive data transfer to an unknown IP outside working hours — possible exfiltration.
In Azure
Azure AI Anomaly Detector is retired on October 1, 2026 (it has not accepted new resources since 2023). Microsoft recommends anomaly detection in Microsoft Fabric (Real-Time Intelligence) or the open-source anomaly-detector project. In AI-901, anomaly detection appears as a machine learning concept, not as a service.
Organizations hold enormous volumes of information in PDFs, emails, images, and documents. Knowledge Mining uses AI to index and extract useful information from those unstructured documents, making them searchable and analyzable.
In Azure: Azure AI Search
Ingests documents from multiple sources (Blob Storage, SQL, SharePoint)
Applies AI skills: OCR, entity extraction, language detection, sentiment analysis
Creates an intelligent search index over the enriched content
Enables semantic search (by meaning, not just exact word match)
| Scenario | Type of AI | Azure service |
|---|---|---|
| Predicting whether a customer will churn | ML — Classification | Azure ML / AutoML |
| Automatically reading scanned invoices | Vision + Knowledge Mining | Azure AI Vision / Document Intelligence |
| Detecting fraudulent transactions in real time | Anomaly detection | Microsoft Fabric (Real-Time Intelligence) |
| Chatbot that answers support questions | NLP + Generative AI | Azure AI Language + Azure OpenAI |
| Automatically generating product descriptions | Generative AI | Azure OpenAI in Foundry (GPT-5) |
| Searching information across 10,000 PDF contracts | Knowledge Mining | Azure AI Search |
| Summarizing video recordings by scene | Computer vision | Azure Content Understanding (video) |
| Analyzing customer opinions on social media | NLP | Azure AI Language — Sentiment Analysis |
Compare the Free and Standard tiers of Azure AI Vision against a real production scenario and pick the right one.
+1 more...
Did you get this topic?
Test what you just learned
A company wants to automate the review of thousands of PDF legal contracts to extract dates, party names, and amounts. Which type of AI workload is most appropriate?
Put what you learned to the test with original AI-901 exam questions.
Start AI-901 mock examUnlimited mock exams with the Pro plan
Log in to track your progress.