What AI Actually Is
Separate artificial intelligence, machine learning, and large language models, and understand what these systems do rather than what they appear to do.
By the end of this lesson you can
- Distinguish AI, machine learning, and large language models
- Explain in plain terms how a language model produces text
- Describe what training data is and why it determines a model's limits
- Explain why these systems are delivered as cloud services
Lesson Notes
Read through the key concepts before you try the challenge.
Three words that are not synonyms
You evaluate a vendor's claims at Lakeside Medical Associates.
A scheduling vendor says its product is 'AI-powered.' That could mean a large language model reading appointment requests, a statistical model predicting no-shows, or a set of if-then rules someone wrote in 2015. All three are marketed the same way, and they carry very different risks.
Your task: Understand the terms precisely enough to ask a vendor what their product actually does.
Key terms
- Artificial intelligence
- The broad field of building systems that perform tasks normally requiring human intelligence. An umbrella term, not a specific technology.
- Machine learning
- A subset of AI where a system learns patterns from data rather than following rules a person wrote. The system is trained, not programmed.
- Large language model (LLM)
- A machine learning model trained on very large amounts of text to predict likely continuations. ChatGPT, Claude, and Gemini are LLM-based products.
- Training data
- The material a model learned from. It determines what the model knows, what it is good at, and which biases it carries.
- Generative AI
- Models that produce new content — text, images, code — rather than only classifying or predicting.
- Prompt
- The input you give a model. With generative systems, the quality of the prompt substantially determines the quality of the output.
A large language model works by predicting likely next words, over and over, given everything before them. That description sounds reductive, and it genuinely is how these systems operate. The output is fluent because the patterns it learned are fluent — not because the system checked whether the content is true.
| Type | Good at | Poor at |
|---|---|---|
| Rule-based system | Consistent, auditable decisions with known logic | Anything its author did not anticipate |
| Machine learning classifier | Predicting categories from many examples — spam, no-show risk | Explaining why; handling cases unlike its training data |
| Large language model | Drafting, summarizing, rephrasing, explaining | Facts, arithmetic, citations, anything needing to be verifiably true |
These models are delivered as cloud services for a straightforward reason: training and running them requires hardware far beyond a workstation. This is the IaaS/PaaS/SaaS stack from Module 2 in its newest form — you consume AI as SaaS, over the internet, with all the shared-responsibility implications that carries. Your prompt leaves your building.
An AI assistant states a specific drug interaction with complete confidence. What does that confidence tell you about the accuracy of the claim?
Challenge
Apply what you've learned in this lesson.
Test the claim rather than accepting it.
- Ask a free AI assistant a factual question in an area you know well. Assess the answer for accuracy and note anything subtly wrong.
- Ask it for three sources supporting a claim. Attempt to locate each one. Record how many actually exist.
- Ask the same question three times in separate conversations. Compare the answers and note where they differ.
- Write a paragraph explaining to a colleague why 'it sounded very confident' is not evidence that an AI answer is correct.
Finished this lesson?
Progress is saved in this browser only. It is not a grade — official progress lives in Brightspace.