Fundamental AI and Machine Learning Concepts
AI-900 starts with the building blocks. Artificial intelligence: systems that perform tasks that normally require human intelligence — perception (recognising images, understanding speech), reasoning (making decisions), learning (improving from data). Machine learning: instead of programming rules explicitly, you train models on data and the algorithm discovers patterns. Types of ML: supervised learning (labelled training data — regression for continuous output, classification for categorical output), unsupervised learning (unlabelled data — clustering groups similar items, anomaly detection finds outliers), reinforcement learning (agent learns through rewards and penalties — used for game-playing AI, robotics). Deep learning: neural networks with many layers — powerful for image recognition, natural language processing, and speech synthesis, but requires large datasets and significant compute. Azure Machine Learning is the platform for building, training, and deploying custom ML models at any scale.
Azure AI Services
Azure AI Services (formerly Cognitive Services) are pre-built AI capabilities accessible via REST API. Vision: Azure AI Vision (image analysis, object detection, OCR — extract text from images and documents), Face API (face detection and verification), Video Indexer (extract insights from video — transcription, speakers, scenes). Language: Azure AI Language (sentiment analysis, key phrase extraction, named entity recognition, language detection, question answering, conversational language understanding), Azure AI Translator (real-time multi-language translation), Azure AI Speech (speech-to-text, text-to-speech, speech translation). Azure OpenAI Service: access to GPT-4, DALL-E, Whisper, and embedding models within Azure's security and compliance boundary — enterprise-grade with private endpoints, Entra ID auth, and content filtering. Document Intelligence: extract structured data from forms, invoices, receipts, and ID documents — pre-built models for common document types or custom models trained on your documents.
Responsible AI Principles
Microsoft's responsible AI framework has six principles, and AI-900 tests all of them. Fairness: AI systems should treat all people fairly — bias in training data leads to biased outcomes (e.g., facial recognition accuracy varying by demographic). Reliability and Safety: AI must perform reliably and safely — unexpected failures in autonomous vehicles or medical diagnosis can be life-threatening. Privacy and Security: AI systems process sensitive data — must protect personal information, comply with regulations, allow consent and opt-out. Inclusiveness: AI should benefit all people — accessibility features (captioning, screen readers powered by AI). Transparency: people should understand how AI makes decisions — explainable AI (XAI) techniques reveal feature importance, confidence scores, and model behaviour. Accountability: humans must be accountable for AI systems — governance processes, oversight boards, human review for high-stakes decisions. AI-900 expects you to identify which principle applies to a given scenario — know all six and their practical implications.