Every student says things like "AI is smart" and "AI learned from what I wrote." These aren't just colloquial. They're misconceptions dressed up as casual language. And they shape how students think about what AI actually is, what it can do, and what it's responsible for.
You can't spend a week on "AI literacy" and call it done. But you *can* build real vocabulary by defining terms in context, all year, as they come up.
The vocabulary that actually matters in a classroom
You don't need to teach neural networks and training sets (unless you're in a CS class). You need students to distinguish:
- Trained on (not "learned from"): The model was shown millions of examples during training. "AI was trained on text from the internet" is different from "AI learned from your essay." Training happened months ago, not when you typed.
- Generated (not "wrote" or "created"): The model predicted the next word, then the next, based on patterns in its training data. "The AI generated text that sounds like..." is different from "The AI understands..." It's statistical prediction, not understanding.
- Pattern-matching (not "thinking"): AI finds the most likely next word based on what usually comes next. Thinking implies deliberation. "The AI matches patterns in its training data" is more accurate than "The AI figured out..."
- Hallucination (not "made a mistake"): When the model generates something that sounds right but is factually false, that's a hallucination—a predictable failure mode, not a whoops. "The AI hallucinated a source that doesn't exist" tells you something different than "The AI made an error."
- Bias (what kind?): AI can reproduce bias from its training data ("historical bias") or fail equally on different groups ("performance bias"). These are different problems. "The AI is biased against X" is less useful than "The AI was trained on data where X was underrepresented, so it performs worse on X."
These definitions *matter* because they change how students think about what to trust and what to scrutinize.
How to teach them
Don't give a vocabulary quiz. Teach them in context.
In action:
- Student asks: "Can I ask the AI to summarize my essay?"
- You say: "Go ahead. Then we'll look at what it generated. It's matching patterns in its training data about what usually comes after 'In conclusion.' You'll probably see if it's actually true or if it hallucinated something."
Now "generated" and "hallucinated" are real words describing something they just saw.
In discussion:
- Student claims: "AI is thinking about this problem like I am."
- You respond: "Actually, it's pattern-matching. It's finding what usually comes next. If it thinks, does it know it's thinking? Does it have beliefs? Let's look at what it actually does." Now "thinking" vs. "pattern-matching" is a real distinction they can see.
In writing:
- Assignment: "Describe what you asked the AI to do and what it generated. What bias, if any, might be in the output? How do you know?"
Now they're using the vocabulary because they need it to be precise.
The vocabulary wall (optional, not required)
Some teachers keep a running chart:
- Term | What it means | Example
- Trained on | Shown examples during training | "Trained on text from 2024 and earlier"
- Generated | Predicted next word repeatedly | "Generated this essay intro"
- Hallucination | Confident false statement | "Made up a source"
Refer to it when new terms come up. Don't treat it as a thing to memorize. Treat it as a reference you all use together.
Where this gets reinforced
Our [learning objectives tool](/learning-objectives) helps you define exactly what vocabulary students should master in your class, and [discussion questions](/discussion-questions) prompt students to think deeply about what these terms mean in practice.
The test of vocabulary that stuck
Listen to how students talk about AI three weeks later. Are they still saying "AI thinks" or are they saying "AI generates" or "AI pattern-matches"? Are they noticing when something might be a hallucination?
If so, the vocabulary did its job. It's not about being pedantic. It's about thinking more clearly about something they're using every day.
