A plain-English introduction to artificial intelligence for IT professionals. Learn the difference between AI, machine learning, deep learning and generative AI, see real examples from tools you already administer, and understand how systems like ChatGPT and Microsoft Copilot produce answers. This is the foundation for the entire AI Mastery curriculum.
Artificial intelligence (AI) is any software that performs tasks which normally require human intelligence — recognising images, understanding language, predicting outcomes, or generating content. Modern AI is built on machine learning: instead of a developer hand-writing every rule, the system learns statistical patterns from large datasets. Generative AI — the type behind ChatGPT, Microsoft Copilot and Gemini — goes one step further: it creates new text, images or code from those learned patterns rather than only classifying or predicting. A useful one-liner: AI is the goal, machine learning is the method, and generative AI is the newest branch of the family.
AI is already inside the stack you administer. Windows 11 ships with Copilot, Microsoft 365 Copilot writes your users' emails and summaries, Defender for Endpoint classifies threats with ML models, and Intune is gaining AI-assisted insights. If you cannot explain the difference between a rule and a model, you cannot evaluate vendor claims, set sensible data-protection policies, or answer the 'is this safe?' question your CISO will ask. AI literacy is quickly becoming as basic for IT careers as knowing what DNS does — and it is now listed in endpoint, cloud and security job descriptions.
At its core, AI is pattern matching at scale. A model is trained on examples: during training the system adjusts millions or billions of internal numbers (weights) until its predictions match the data. This produces a trained model. When you use it, the model runs 'inference': your input is converted into numbers, pushed through those learned patterns, and the most likely output is produced. Traditional software follows rules the author wrote ('if score > 700, block'); a model instead outputs a probability learned from data. That is why AI can handle messy, unpredictable input like typos in a user query — and also why it can be confidently wrong.
Human question or task
↓
AI system (model + training data)
↓
Learned patterns → prediction
↓
Answer / action / generated contentYou already manage AI-powered features today. Outlook's spam filter is classic machine learning trained on millions of flagged messages. Defender for Endpoint scores suspicious behaviour with ML models instead of a static signature list. Windows Hello uses a deep neural network to match your face against an encrypted template stored on device. And Microsoft Copilot combines an LLM with your Microsoft Graph data to draft emails, summarise meetings and explain settings. In each case the pattern is identical: data went in during training, and a model now makes fast, probabilistic decisions in production.
Open ChatGPT (or Copilot) and paste: "Explain the difference between AI, machine learning, deep learning and generative AI in plain English, with one real Microsoft product example for each." Read the answer twice — it is the mental model this whole curriculum builds on.
Paste a 200-word internal email or ticket into the chat and ask: "Summarise this in 3 bullet points and describe the sender's tone." You just used a generative model for extraction and tone analysis — two things rule-based software does poorly.
Ask: "What is a token in an LLM? Show how the sentence 'Intune failed to install the app' would be split into tokens." This connects today's big picture to the concrete mechanics covered in Topic 2.
On paper, classify these into traditional software, classic ML, deep learning and generative AI: Excel forecast sheet, Outlook spam filter, Windows Hello face unlock, Microsoft 365 Copilot, a firewall rule. Then ask the AI the same question and compare answers.
Ask: "List 10 AI terms a Windows admin will meet in 2025 with a one-line definition each, written for an IT audience." Save the answer in a notes file — Topic 5 turns this into a full 100+ term glossary.
Treat any public AI tool as an external service: data you paste leaves your control. Before using ChatGPT or a free chatbot with work content, check your organisation's AI usage policy, and prefer enterprise offerings — Microsoft 365 Copilot and Azure OpenAI instances do not train on your tenant data and keep prompts inside your compliance boundary. Never paste passwords, API keys, personal employee data or customer records into a public chatbot, even 'just to test'. Classify first, paste second.
AI is the broad field of making software perform tasks that need human-like intelligence. Machine learning is a subset of AI where systems learn patterns from data instead of being explicitly programmed. Deep learning is a subset of ML that uses multi-layer neural networks, and it powers almost all modern generative AI — so every deep learning system is ML and AI, but not the reverse.
Traditional AI models classify or predict — a spam filter labels mail, a forecast model predicts next month's load. Generative models produce new content: text, images, code. They are trained on huge corpora to model language itself, so they can write an answer they have never seen verbatim. For IT, the practical difference is that generative AI can draft scripts, documentation and explanations, while traditional AI scores and sorts.
Outlook spam filtering is classic ML. Defender for Endpoint uses ML models to score files and behaviours beyond signatures. Windows Hello face recognition is on-device deep learning. Microsoft 365 Copilot grounds an LLM in your Microsoft Graph data to summarise mail and meetings. Even Intune's reporting trend lines and risk-based message centre insights lean on statistical models.
Write down five systems you touch every week (mail, endpoint security, identity, ticketing, monitoring). For each, guess which of the four buckets it falls into — traditional software, classic ML, deep learning, generative AI — and note one way AI could assist your work there. Then ask ChatGPT to do the same classification for you and compare. Keep the list; you will reuse it in the Level 3 topics.
No. Using and evaluating AI as an IT professional requires understanding concepts — models, training, inference, context — not calculus. The math matters only if you go on to train or fine-tune models yourself, which is a Level 5 concern in this curriculum, not a Level 1 one.
ChatGPT is one product built on one type of AI — a large language model (LLM). AI also includes image classifiers, speech recognition, recommendation engines and classic ML. Thinking ChatGPT equals AI is like thinking Outlook equals email servers: a popular face of a much larger field.
AI automates tasks, not accountability. It can draft a PowerShell script, but someone must validate it, understand blast radius, own the change process and answer for it in an audit. Admins who use AI well replace admins who refuse to — that is the realistic near-term picture, and this curriculum is built around it.