1

Class 1: What AI Is

Class 1 of 8About 3 hoursNo coding

By the end of this class you will be able to

• Explain the AI family tree: how artificial intelligence, machine learning, deep learning and generative AI fit inside each other.
• Describe how learning from examples differs from a person writing rules, and why that matters when AI gets things wrong.
• Place today's chat tools on a short timeline of AI, from the 1950s to agents.
• Name what generative AI is good at and bad at, and separate hype from reality.
• Use computational thinking (decompose, spot patterns, abstract, write steps) to decide which parts of a task suit AI, a normal program or a human.

Where we are

This is the start of the course. Today we build a shared vocabulary and a realistic picture of what AI is and isn't, and we meet the case study that runs through every class. In Class 2 we open up the chat tool and see how it actually produces its answers.

Welcome

This course is for colleagues who use AI tools in their work, or will soon, and who want to use them well: confidently, safely and with a clear head about what they can and can't do. You won't write any code. You will need access to your organization's approved AI chat tool, such as Microsoft Copilot, ChatGPT Enterprise or Claude, depending on what your organization has approved (as of 2026).

Each class takes about three hours and follows the same rhythm: a short warm-up quiz, a concept taught with one strong analogy, a live demonstration (including one that goes wrong on purpose), a hands-on Prompt Lab in pairs, a quiz to check your understanding and a one-question exit ticket. You can also work through the pages on your own.

The case study: the SitRep Assistant

Throughout the course we work on one realistic task. A situation report (SitRep) is a short, regular update that tells decision-makers what is happening in a crisis: how many people are affected, what they need most, what is already being done and what is still unknown. SitReps are built from many sources, and those sources are usually messy: field notes typed on a phone, figures from village leaders, partial counts, ambiguous dates.

Our fictional setting is Aramu district in Northern Veloria province, where heavy rain has caused the Tula river to flood. Field teams from the Veloria Relief Coalition send in notes like this one:

approx 1,200 households displaced in total (our estimate + figures from village chiefs). most are staying at Aramu primary school, which is full - maybe 600 people there. could NOT reach Dorra, road cut.

The SitRep Assistant is the idea of using AI to help turn notes like these into a clear, honest situation report. In this course you will use and judge it through your chat tool. (In the parallel AI Engineering course, technical colleagues build it.) By Class 8 you will design an AI-assisted workflow from your own job.

Why a case study?

It's easy to be impressed by AI in general. It's more useful to ask whether it helps with one specific, real piece of work, and what could go wrong if it does. The SitRep gives us that concrete test in every class.

Today's question is the most basic one: when people say "AI", what do they actually mean?

1. What do we mean by "AI"?

Artificial intelligence (AI) is the broad field of making computers do tasks that normally need human intelligence: understanding language, recognising images, making predictions, planning, translating. It's an umbrella term, and a very wide one. The spell-checker in your word processor, the system that flags suspicious bank transactions, the satellite analysis that maps flooded areas, and the chat tool that drafts your email all get called "AI".

That width causes confusion. When a vendor says their product "uses AI", or a colleague says "AI can't be trusted with numbers", they may be talking about completely different kinds of system. So it helps to know the family tree.

• Artificial intelligence is the whole field.
• Machine learning (ML) is the branch of AI where the system learns patterns from examples, rather than following rules that a person wrote.
• Deep learning is a kind of machine learning that uses very large neural networks (explained below). Most of the progress you've heard about since around 2012 is deep learning.
• Generative AI is deep learning used to create new content, such as text, images, audio or code, instead of only sorting or scoring things.
• Large language models (LLMs) are generative AI for text. They are the engines inside chat tools like Copilot, ChatGPT and Claude.

Each circle sits inside the one before it: every LLM is generative AI, every generative AI system uses deep learning, and so on. But the reverse isn't true. Most AI is not generative.

Know the Term: The AI family tree

Artificial intelligence (AI) is the broad field of making computers do tasks that normally need human intelligence, such as recognising speech or planning a route. Machine learning (ML) is one branch of AI: instead of following rules a person wrote, the system learns patterns from examples. Deep learning is a kind of machine learning that uses large neural networks with many layers, and it drives most AI progress since around 2012. Generative AI is deep learning that creates new content (text, images, audio, code) rather than only sorting or predicting. Large language models (LLMs), the engines inside chat tools, are generative AI for text. Each circle sits inside the one before it: AI ⊃ machine learning ⊃ deep learning ⊃ generative AI.

Analogy: "vehicle" ⊃ "car" ⊃ "electric car" ⊃ "self-driving electric car". Every self-driving electric car is a vehicle, but most vehicles are not self-driving electric cars.

Why you care: when someone says "we use AI", ask which kind. A spam filter, a satellite flood map and a chatbot are all "AI", but they fail in very different ways and need different checks.

Here is how that looks with examples you may meet at work:

System	Where it sits	What it does
A rule that routes emails with "invoice" in the subject to Finance	AI in the loosest sense, or just ordinary software	Follows a rule a person wrote
A spam filter	Machine learning	Learned from millions of emails labelled "spam" or "not spam"
A model that spots flooded areas in satellite images	Deep learning	Learned from many labelled images; outputs a map, not new content
A tool that translates field notes from French to English	Deep learning (and increasingly generative AI)	Produces new text in another language
A chat tool drafting a SitRep	Generative AI (an LLM)	Writes new text, word by word

A useful habit

When someone says "AI", ask: which kind, trained on what, and producing what? A system that sorts things makes different mistakes from one that writes things.

2. Learning from examples vs programming rules

For most of computing history, if you wanted a computer to do something, a person wrote the rules. A program is a precise list of instructions: if this, then do that. Programs are excellent when the rules are clear. A payroll system, a budget spreadsheet and a travel approval form are all ordinary programs. They do exactly what they were told, every time.

But some tasks are very hard to write rules for. Try writing down the rules for recognising a cat in a photo. Pointy ears? Some cats have folded ears. Whiskers? So do dogs. Four legs? So does a table. After a while you realise you know a cat when you see one, but you can't list the rules.

Machine learning takes a different approach. Instead of writing the rules, you show the system thousands or millions of examples, each with the right answer (cat, not cat), and let it adjust itself until it gets most of them right. This adjusting is called training. The result is a model: the trained system that you then use on new cases.

Two ways to flag urgent field notes

The rules approach. A programmer writes: if the note contains "cholera", "death" or "no water", mark it urgent. It is predictable and easy to explain. But it misses "people drinking from the river" (no keyword), and it wrongly flags "no deaths reported" (contains "death").

The learning approach. You collect 5,000 past field notes that experienced officers have already marked urgent or not urgent, and train a model on them. It picks up patterns no one wrote down, and it catches "drinking from the river". But you can't see why it flagged any note. And if the past officers tended to under-rate needs in one region, the model learns that bias too.

Neither approach is "better". They have different strengths:

	Rules (a normal program)	Learning from examples (machine learning)
Good for	Clear, stable rules: calculations, forms, checks	Fuzzy patterns: language, images, judgement calls
How it's made	A person writes every rule	The system learns patterns from data
When it's wrong	You can find the faulty rule and fix it	Hard to say why; you fix it with better data or checks
Predictable?	Same input, same output, always	Usually, but it can surprise you on unusual cases
Main risk	Misses cases nobody wrote a rule for	Copies gaps and biases in its examples; confident mistakes

The engine behind modern machine learning is the neural network.

Know the Term: Neural network

A neural network is a mathematical model loosely inspired by the brain. It is made of many simple units ("neurons") arranged in layers. Each unit takes in numbers, multiplies them by adjustable settings called weights (also called parameters), adds them up and passes the result on to the next layer. During training, the network sees millions of examples, and after each one its weights are nudged slightly so that its output gets closer to the right answer. Nobody writes the rules by hand; the rules end up spread across the weights. A deep network simply has many layers, and today's large language models have billions of weights.

Analogy: a huge mixing desk with billions of small dials. Training is a patient sound engineer turning each dial a fraction at a time until the music sounds right.

Why you care: because the "knowledge" is spread across billions of weights, nobody can open the model and point to the line where a fact or a bias lives. That is why we test AI by checking its outputs, not by reading its rules.

Three consequences of "learning from examples" matter for everything in this course:

1. A model is only as good as its examples. If the training data rarely includes a language, a region or a kind of situation, the model will be weaker there.
2. Nobody can read the rules. The patterns are spread across billions of numbers. We judge a model by testing what it does, not by inspecting how it works.
3. Models produce their best guess, always. A rule-based program can stop and say "no rule matches". A trained model almost always gives some answer, whether or not it has good grounds for it. Class 2 shows how this leads chatbots to make things up.

3. A short history

AI isn't new. The field is about 70 years old, and it has had several waves of excitement followed by disappointment. Knowing the outline helps you keep today's claims in proportion.

When	What happened	Why it matters
1950s	Alan Turing asks "Can machines think?" (1950). The term "artificial intelligence" is coined at a 1956 workshop	The ambition is old; so are the over-optimistic predictions
1970s–1980s	Expert systems: thousands of hand-written "if… then…" rules capturing specialists' knowledge, e.g. for medical diagnosis	Worked in narrow areas but were brittle and costly to maintain. Funding dried up in "AI winters"
1990s–2000s	Machine learning spreads: spam filters, search ranking, recommendations. IBM's Deep Blue beats the world chess champion (1997)	The shift from writing rules to learning from data
2012	A deep neural network wins the ImageNet image-recognition contest by a wide margin	Start of the deep learning boom, powered by more data and faster chips
2017	Google researchers publish the transformer design	The architecture behind almost every modern chat model (Class 2)
2018–2020	Large language models grow rapidly in size and ability	Fluent text generation becomes possible
November 2022	ChatGPT is released and reaches 100 million users within about two months	Generative AI reaches the general public and the workplace
2023–2026	Models handle images and audio, "reasoning" modes appear, and agents that take actions with tools spread into workplace products (as of 2026)	From answering questions to doing tasks (Class 7)

Know the Term: A short AI timeline

1950: Alan Turing asks "Can machines think?" and proposes his imitation game. 1956: the term "artificial intelligence" is coined at a workshop at Dartmouth College. 1970s–1980s: expert systems encode specialists' knowledge as thousands of hand-written "if… then…" rules; they work in narrow areas but are costly to maintain, and funding cools in periods later called "AI winters". 1990s–2000s: machine learning takes over, learning patterns from data for tasks like spam filtering. 2012: a deep neural network wins the ImageNet image-recognition contest by a wide margin, starting the deep learning boom. 2017: Google researchers publish the transformer design. 2018–2020: large language models such as GPT-2 and GPT-3 appear. November 2022: ChatGPT's release brings generative AI to the general public. 2023–2026: models become multimodal (text, images, audio), reasoning models appear, and agents that can take actions with tools spread into workplace products (as of 2026).

Analogy: like the history of flight: decades of gliders and false starts, then a few breakthroughs close together that suddenly made air travel ordinary.

Why you care: AI has been "about to change everything" several times before. Knowing the history helps you tell a real shift from a sales pitch.

Why did things speed up so much after 2012? Three ingredients arrived together: data (the internet provided enormous amounts of text and images), computing power (chips originally built for video games turned out to be ideal for training neural networks) and better designs (most importantly the transformer). None of this involved a machine suddenly "waking up". It was engineering at a very large scale.

4. What generative AI is good at, and bad at

Generative AI is remarkable at some things and unreliable at others, and the line between the two isn't always where you'd expect. Researchers sometimes call this a jagged frontier: the same tool that writes a polished two-page briefing in seconds may fail to count the words in it correctly.

Usually good at	Usually weak at, or risky for
Drafting: emails, reports, talking points, first versions	Facts it wasn't given, especially specific figures, names, dates and references
Summarizing and restructuring text you provide	Recent events after its training (unless connected to search)
Rewriting for a different audience, tone or length	Exact counting and arithmetic on long lists, done "in its head"
Translating between major languages	Knowing when it doesn't know; it rarely says "I'm not sure" unprompted
Brainstorming options, questions, risks	Judgements that need local context, relationships or accountability
Explaining concepts in plain language	Consistency: ask twice, get two different answers
Pulling key points into a table	Anything where one invented detail causes real harm

Look at the pattern. Generative AI is strongest when you supply the material and it transforms it: summarize this, reformat that, rewrite for a donor. It is weakest when it must supply facts from memory, because it produces plausible text, not checked text.

Fluent is not the same as true

A chat tool writes in confident, well-organised prose whether it's right or wrong. The quality of the writing tells you nothing about the accuracy of the content. This is the single most important habit to build in this course: judge the facts, not the fluency.

For the SitRep this gives us a first rule of thumb. Asking the AI to turn these field notes into a structured update plays to its strengths. Asking it to write a SitRep on the Aramu floods with no notes attached asks it to invent one, and it will.

5. Hype vs reality

AI attracts strong claims in both directions. Some say it will replace most office work within a year; others dismiss it as "just autocomplete" with no real use. Both are wrong in useful ways.

You may hear…	The reality (as of 2026)
"AI understands what it's saying."	It is extremely good at producing language that fits, which often looks like understanding. Whether it "understands" in a human sense is debated. What isn't debated: it has no built-in way to check its statements against the world.
"AI is just autocomplete, so it's useless."	At heart it does predict text, but prediction at this scale can summarize, translate, draft and reason through problems usefully. "Just prediction" doesn't mean "just trivial".
"AI is objective because it's a machine."	It learns from human-written data and copies its patterns, including its gaps and biases. Class 6 covers this.
"AI will replace my job."	It changes tasks more than whole jobs. Drafting gets faster; checking, judgement, relationships and accountability remain human.
"It's connected to the internet, so it knows everything current."	Only if your tool has search switched on. Otherwise it knows only what it learned up to a cut-off date (Class 4).
"If it says it's sure, it's right."	Confidence in the wording tells you nothing. Models sound equally sure when they are wrong.
"The new version fixes all that."	Each generation improves, sometimes a lot. Hallucination and bias are reduced but not solved.

Here are five questions to ask whenever you hear a claim about AI, whether from a vendor, a news story or an enthusiastic colleague:

1. Which kind of AI? A chatbot, a classifier or a forecasting model?
2. Tested on what? On data like ours, in our languages, in our context?
3. How often is it wrong, and how do we find out when it is?
4. Who checks the output, and who is accountable for it?
5. Compared to what? Is it better than our current process, or only better than nothing?

6. Computational thinking: deciding where AI fits

The most useful skill in this course isn't a prompting trick. It's a way of looking at your work called computational thinking. The name sounds technical, but it just means breaking a problem down so that it's clear which parts a computer can do. You already do most of it when you plan a project. It has four parts:

Step	In plain words	In the SitRep
Decompose	Break the big job into small steps	"Write the SitRep" becomes: collect notes, extract figures, check them, write it up, approve, send
Spot patterns	Notice what repeats, and what is the same each time	Every week: same headings, same kinds of figures, same gaps (unreachable villages, unclear dates)
Abstract	Keep what matters, drop the detail that doesn't	The director needs affected numbers, needs, response and gaps, not every sentence of every note
Write steps	Describe the process clearly enough that someone else could follow it	"For each note: extract figures with units; mark estimates; list unknowns…"

Once a job is broken into steps, you can ask of each step: who or what should do this?

• AI (generative AI) is a good fit for language work with fuzzy patterns: summarizing, drafting, rewording, pulling out key points, translating. Its output needs checking.
• A normal program is a good fit for anything with exact rules: adding up figures, checking a date format, filling a fixed template, sending an email at 9:00. It's predictable, cheap and exact, and a spreadsheet counts.
• A human is needed for judgement, accountability, relationships, anything sensitive, and the final check. If the step's consequences are serious, a person should own it.

Here is the SitRep process broken down this way:

Step	Best done by	Why
Receive field notes from teams	Normal program (email, shared folder)	Routine; no judgement needed
Pull out figures, needs and locations from messy notes	AI, checked by a human	Fuzzy language; AI is fast at this but can misread or invent
Add up households across locations	Normal program (a spreadsheet)	Exact arithmetic; AI "in its head" can get sums wrong
Spot double-counting between teams and flag unclear dates	AI to flag, human to decide	AI can point out candidates; a person knows the villages
Draft the SitRep text in the standard format	AI	Drafting in a fixed structure is a strength
Check every figure against the source notes	Human (AI can help list figures to check)	Accountability; this is where invented numbers get caught
Decide what to say about sensitive issues (security, protection)	Human	Judgement and risk
Approve and send	Human	Someone must be accountable for what goes out

Notice that AI doesn't take over the SitRep. It speeds up two or three steps inside a process that people still design, check and own. That pattern (break it down, give each step to the right worker, verify) comes back in every class of this course, and it's exactly how engineers design larger AI systems too (see the AI Map).

The three-way question

For any task you're thinking of giving to AI, ask: Does this step need language skill (AI), exact rules (a normal program) or judgement and accountability (a human)? Many steps need two of the three.

Prompt Lab

Work in pairs using your organization's approved AI chat tool. One partner types; the other reads the answers critically and asks "How do we know that's true?" Swap each exercise. Answers vary from run to run, so don't worry if yours differ from your neighbours'.

Some exercises use the fictional field note from our case study. Copy it once and keep it ready to paste.

The field report (fictional)

FIELD NOTE - Aramu district, Northern Veloria - sent Thurs eve From: K. Osei, Veloria Relief Coalition field team

heavy rain since the weekend, Tula river burst near Kessa bridge. we got to Aramu town + Lomi and Tavet villages today. could NOT reach Dorra, road cut after the bridge, no phone signal there either.

approx 1,200 households displaced in total (our estimate + figures from village chiefs). most are staying at Aramu primary school, which is full - maybe 600 people there. more families in the community hall but we didn't manage to count them.

water is the big problem. 2 of the 3 boreholes in Lomi are flooded and people are drinking from the river. Tavet health post still open but running low on ORS, nurse reports 14 cases of acute diarrhoea since Monday, mostly children under 5. no deaths reported to us.

VRC is distributing food at the school, enough stock for about 3 days. next truck expected 04/05 if the Kessa road is passable. urgent: water purification, hygiene kits, ORS, shelter materials for the hall.

Exercise 1: Meet the tool

Task: Ask the AI to describe itself, then judge how much of its answer you can check.

In plain language, for a colleague with no technical background:
1. What kind of AI system are you (use the terms artificial intelligence,
   machine learning, deep learning, generative AI and large language model)?
2. What are you good at, and what are you bad at?
3. Can you look things up on the internet right now, and how would I
   know?
Keep it under 250 words.

What to look for:

• Does its description match the family tree in Section 1? Where does it place itself?
• Does its "bad at" list include making things up, recent events and arithmetic?
• On question 3: is its answer about search consistent with what your tool actually shows (for example, a search indicator or links)? Models are not always accurate about their own features.
• How confident does it sound? Is that confidence earned?

Exercise 2: Place it on the family tree

Task: Ask the AI to sort some everyday systems into the family tree, then check its sorting against Section 1 and discuss any disagreements.

Place each of these systems in the most specific category that fits:
"ordinary program (rules)", "machine learning", "deep learning" or "generative AI". Give a one-line reason for each.

1. A spreadsheet formula that adds up households across villages
2. An email spam filter
3. A tool that detects flooded areas in satellite images
4. A chatbot that drafts a situation report
5. A rule that flags any field note containing the word "cholera"
6. An app that turns a voice message into written text
7. A tool that generates a picture from a text description

What to look for:

• Does it correctly call items 1 and 5 ordinary programs, not AI?
• Items 3 and 6 are deep learning but not necessarily generative. Does it make that distinction, or call everything "AI"?
• If you disagree with it, who is right? Argue it out with your partner. There are borderline cases (item 6 can be built either way), and that's a good discussion.

Exercise 3: Strengths and weaknesses, side by side

Task: Send three requests in the same chat. The first plays to generative AI's strengths; the second and third don't. Compare the results.

Request A:
Summarize the field note below in five bullet points for a country director. Use only information in the note.

Field note:
"""
[paste the field report here]
"""

Request B:
What was the population of Aramu district at the last census, and how many health posts does Northern Veloria province have?

Request C:
In the field note above, count exactly how many times the letter "a" appears, and then add up every number that appears in the note.

What to look for:

• Request A: is the summary accurate? Does it stick to the note?
• Request B: Veloria is fictional, so there is no true answer. Did the AI say it doesn't know, or did it produce confident figures? If it invented numbers, you have just seen a hallucination, which Class 2 explains.
• Request C: do you believe its counts? How would you check them? What's the better tool for this step (hint: Section 6)?
• Which of the three would you trust in a real SitRep without checking?

Exercise 4: Hype check

Task: Use the AI to stress-test an AI sales claim, then compare its questions with the five questions in Section 5.

A vendor tells our relief coalition: "Our AI reads every field report and produces a 100% accurate situation report automatically, with no need for human review. It works in any language."

Act as a sceptical but fair adviser. List the questions we should ask the vendor before believing this claim, and explain briefly why each question matters. Then say which parts of the claim are plausible and which are not.

What to look for:

• Does it question "100% accurate" and "no need for human review"?
• Does it ask about languages, testing on data like ours, and who is accountable for errors?
• Which of our five questions did it miss? Which of its questions would you add to our list?
• Notice the irony: you're asking an AI to be sceptical about AI. Did it do a good job? Why might it?

Exercise 5: AI, program or human?

Task: Apply computational thinking to a real, recurring task from your own job. It must be safe to discuss with your approved tool, so use general descriptions only, no confidential details. Fill in the brackets.

I work as [your role] in [type of organization]. One task I do regularly is
[describe the task in two or three sentences, without confidential
details].

Help me apply computational thinking to it:
1. Break the task into its individual steps (decompose).
2. For each step, say whether it is best done by generative AI, by a
   normal program (such as a spreadsheet or form), or by a human, and give a one-line reason.
3. Point out which steps need a human check even if AI does them.
Present it as a table.

What to look for:

• Is the breakdown realistic? What steps did it miss that only you would know about?
• Did it give everything to AI? Push back where a spreadsheet or a person is the better choice.
• Does it keep humans on judgement, sensitive issues and final approval?
• Keep this table. You'll come back to it in Class 5 and in your final project in Class 8.

Check Your Understanding

1 A colleague says: "Generative AI, machine learning and AI are just three names for the same thing." Which statement is accurate?

1. They are the same thing; the names changed as the technology became popular.
2. Generative AI is a kind of machine learning, which is a kind of AI, but most AI and most machine learning is not generative.
3. Machine learning is the newest and most advanced of the three, so it contains the others.
4. AI means rules written by programmers; machine learning and generative AI are separate fields outside AI.

2 Your office trained a model on 5,000 past field notes that officers had marked "urgent" or "not urgent". It works well, except that it rarely flags notes from one remote district as urgent. What is the most likely explanation?

1. Someone wrote a rule telling it to ignore that district.
2. The model is broken and should be restarted.
3. The district is too far away for the model to process.
4. The past examples under-rated or rarely included that district, and the model learned that pattern.

3 AI research began in the 1950s, yet most of today's tools appeared after 2012. What mainly explains the speed-up?

1. Much more data, much faster computing chips, and better designs such as the 2017 transformer, arriving together.
2. Around 2012 computers became conscious and started improving themselves.
3. ChatGPT invented artificial intelligence in 2022.
4. Programmers finally finished writing all the rules needed for intelligence.

4 Which of these SitRep tasks best plays to the strengths of a generative AI chat tool?

1. Stating the census population of Aramu district from memory.
2. Adding up displaced households from 40 village reports, exactly.
3. Turning pasted field notes into a structured, five-heading summary.
4. Deciding whether to mention a sensitive security incident in the report.

5 A vendor says their AI "writes 100% accurate situation reports with no need for human review". What is the most useful response?

1. Accept it; AI tools from large companies are thoroughly tested.
2. Ask how it was tested on data like ours, how often it is wrong, how errors are caught, and who is accountable for them.
3. Reject it; AI is just autocomplete and can't do anything useful.
4. Accept it if the demonstration output is well written and professional.

6 You have broken the SitRep process into steps. Which is the best choice for the step "add up the displaced households reported by each team"?

1. Generative AI, because it is the most advanced technology available.
2. A human, because computers can't be trusted with numbers.
3. Skip the step; the director only needs a rough idea.
4. A normal program, such as a spreadsheet, with a person checking overlaps and estimates.

7 Without giving it any notes, you ask a chat tool to "write a SitRep on the Aramu floods". It returns a polished report with population figures, agency names and dates. What should you conclude?

1. The specific details are very likely invented. Fluent writing is no evidence of accuracy.
2. The tool must have found the information online, so it's probably accurate.
3. The report is reliable because it is so detailed and well organised.
4. The figures are probably rough but close enough for a first report.

Exit Ticket

Before you leave

Name one task in your job where you now think AI could help, and one step of that task that should stay with a normal program or a human. Why?

Key terms

• Artificial intelligence (AI): the broad field of making computers do tasks that normally need human intelligence.
• Machine learning (ML): a branch of AI where the system learns patterns from examples instead of following rules a person wrote.
• Deep learning: machine learning using large neural networks with many layers.
• Generative AI: AI that creates new content such as text, images, audio or code.
• Large language model (LLM): a generative AI model for text; the engine inside chat tools.
• Neural network: a model made of many simple connected units whose adjustable weights are tuned during training.
• Training: the process of adjusting a model using examples until it performs the task well.
• Model: the trained system that is then used on new inputs.
• Program (rules-based system): software that follows instructions written by a person.
• Expert system: an early form of AI built from many hand-written "if… then…" rules.
• Jagged frontier: the uneven pattern of AI ability, strong at some tasks and weak at similar-looking ones.
• Computational thinking: breaking a problem into steps (decompose, spot patterns, abstract, write steps) so it's clear what a computer, AI or human should do.
• SitRep (situation report): a short regular update for decision-makers on a crisis.