What AI is, and what it is not
About 20 minutes · plain language
People use the words “AI”, “algorithm”, “machine learning” and “ChatGPT” as if they mean the same thing. They do not. In this lesson you will learn what each one is, in plain words, so you can tell what kind of tool you are using and what can go wrong with it.
- Tell apart rule-based systems, machine learning, deep learning and generative AI.
- Explain why an AI system can sound confident and still be wrong.
Key terms
- Artificial intelligence (AI)
- Computer systems that do tasks we usually link with human thinking, such as recognising patterns, making predictions, understanding language or making recommendations.Example: A program that reads an ECG and flags atrial fibrillation.
- Algorithm
- A set of step-by-step instructions a computer follows. Every AI system uses algorithms, but not every algorithm is AI.Example: A fixed rule: “if heart rate > 130, show an alert”.
- Rule-based system
- A system that follows rules written by people. It does exactly what the rules say and nothing else.Example: A drug-interaction alert in the electronic record.
- Machine learning (ML)
- A type of AI that learns patterns from examples (data) instead of being given rules.Example: A model that learned from 50,000 past admissions which patients later needed intensive care.
- Training data
- The examples a machine-learning model learns from.Example: Records of past patients labelled “developed a pressure injury” or “did not”.
- Deep learning
- Machine learning that uses many-layered “neural networks”. It works well with images, sound and text.Example: Software that finds signs of diabetic eye disease in retinal photos.
- Generative AI
- AI that creates new content, such as text, images or audio, in response to a request (a “prompt”).Example: A chatbot that writes a patient information leaflet.
- Large language model (LLM)
- A generative AI model trained on huge amounts of text. It writes by predicting which words are likely to come next.Example: ChatGPT, Claude, Gemini and similar tools.
- Hallucination
- When a generative AI tool produces content that sounds right but is false or invented.Example: A list of references in which some articles do not exist.
- Foundation model
- A very large model trained once and then adapted to many tasks and products.Example: One language model used inside a note-writing tool, a chatbot and a coding assistant.
AI is a family, not one thing
Think of AI as a family with several members. The figure above shows how they fit inside each other. The oldest members are rule-based systems: people write the rules, and the computer follows them. They are predictable, but they cannot cope with situations nobody wrote a rule for.
Machine learning works differently. Instead of rules, it is given many examples and finds the patterns itself. If you show a model thousands of patient records, some marked “deteriorated” and some not, it learns which combinations of signs came before deterioration. It does not understand illness; it recognises patterns in past data.
Deep learning and why it changed things
Deep learning is a powerful kind of machine learning. It made it practical for computers to read images, recognise speech and work with language. Most of the AI you hear about in the news since about 2012, from image reading to chatbots, is built on deep learning.
Generative AI and large language models
Generative AI creates new content. Large language models (LLMs) are the best-known example. An LLM has read a huge amount of text and learned how words usually follow each other. When you ask a question, it writes an answer one piece at a time, choosing words that are likely to come next.
This explains both its strength and its weakness. It is very good at producing fluent, well-organised text. But nothing in this process checks whether the text is true. That is why LLMs can produce hallucinations: confident statements, numbers or references that are wrong or invented. In one study, more than half of the references an early version of ChatGPT produced did not exist, and a later version still invented almost one in five.
Why this matters for you
Knowing the type of AI tells you what to check. For a rule-based alert, check that the rules are right. For a machine-learning model, ask what data it learned from and whether it was tested on patients like yours (Lesson 2). For a generative AI tool, check every fact it gives you (Lesson 4).
Remember also that AI tools do not “know” your patient, your ward or this week’s new guideline. They only know what was in their data, up to the date they were trained, unless they search current sources, and even then they can misread what they find (Lesson 5).
Case: the perfect reference list
Mariam, a second-year nursing student, asks a chatbot for five recent studies on preventing pressure injuries in older adults. The answer looks perfect: authors, titles, journals, years and page numbers. Before using them, she searches each title in PubMed. Three exist. Two do not: the authors are real people, but the articles were never written. Mariam removes them, finds two real studies instead, and adds a note to her assignment saying how she used the chatbot.
Common mistakes
- Treating every AI tool as the same kind of thing.
- Believing an answer because it is well written.
- Assuming the tool knows your patient or the latest evidence.
Check your understanding
1. A ward alert says “check potassium” whenever a patient is prescribed two specific drugs together. What kind of system is this?
A rule-based system. People wrote the rule; it does not learn from data.
2. Why can a large language model give a confident but wrong answer?
Because it is designed to produce likely text, not to check facts. Fluency is not evidence of truth.
3. Name one thing you should check for a machine-learning tool that you would not need to check for a rule-based tool.
What data it learned from, and whether it was tested on patients like yours.
Try it
- Name one tool you used this week and place it in the circles in the figure.
- Ask a chatbot for three references on a topic you know well, then check each one in PubMed.
Remember
- AI is a family: rule-based systems, machine learning, deep learning and generative AI.
- Machine learning learns patterns from past data; it does not understand patients.
- Generative AI writes likely text. Always check facts and references.
- World Health Organization. Ethics and governance of artificial intelligence for health: guidance on large multi-modal models. Geneva: WHO; 2024. www.who.int
- National Institute of Standards and Technology. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile. NIST AI 600-1. Gaithersburg (MD): NIST; 2024. doi.org
- Walters WH, Wilder EI. Fabrication and errors in the bibliographic citations generated by ChatGPT. Sci Rep. 2023;13(1):14045. pubmed.ncbi.nlm.nih.gov
- Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019;25(1):44-56. pubmed.ncbi.nlm.nih.gov
Lesson text is our own plain-language summary of these sources; it is not endorsed by their authors. Check the original before relying on any detail.