Digital Literacy · AI Ethics

How to understand what AI does well
and know what it gets wrong

Practical and engaging ways to teach your students how to use AI effectively.
⏱ 55–65 min
👥 Groups of 4
🎯 8 activities
Learning Goal
2 / 18
By the end of this lesson, you will be able to
Classify AI strengths and weaknesses

Given 10 scenarios, sort them accurately into what AI handles well vs. poorly, hitting at least 8 out of 10.

📝

A note worth sharing with students: The meat and potatoes of this lesson were designed by humans. The actual materials were prepared with the help of AI. That distinction matters, and it's exactly what this lesson is about.

Phase 1, Hook
3 / 18
📺
Before we say anything...

Watch this.

A short clip from Husk.IRL, real AI, real failures, real world.

No context. Just watch.

Phase 1, Hook Debrief
4 / 18
Quick reactions

What just happened there?

🤔   What did AI try to do?

❌   Where did it go wrong?

💬   Were you surprised? Why or why not?

Today we're going to figure out exactly where AI breaks down, where it genuinely excels, and why that difference matters for how you use it.

Phase 2, Activity 1
5 / 18
The Sort, 10 min

16 cards. 2 piles. 4 people.

✅ Pile A, AI Does This Well

  • The whole group must agree
  • Be ready to explain your reasoning
  • Some cards are obvious, some aren't

❌ Pile B, AI Struggles Here

  • Disagreement within your group is fine
  • Note which cards caused debate
  • Cards 13 and 14 are tricky on purpose

⏱ You have 10 minutes. When time is up, your group should be ready to share which card sparked the most debate, and why.

Phase 2, Answer Key
6 / 18
The Sort, reveal and discuss

How the 16 cards sort out

AI Does This Well

Tap a card to see why
01, Summarize a long article
Compressing and pattern-matching text is its core strength, no real understanding of meaning required.
03, Write in a specific style
Style is statistical texture; it has absorbed millions of examples and can mimic the pattern.
05, Explain things multiple ways
It can rephrase endlessly, remixing the same idea for different reading levels on demand.
07, Translate text instantly
Trained on huge bilingual datasets, mapping one language onto another is exactly what it does.
09, Generate lots of ideas fast
Volume and speed are effectively free for AI; it never runs out of combinations to offer.
12, Find patterns across research
It can scan and cross-reference far more text, far faster, than any human reader.
16, Draft a professional email
A formulaic, well-represented format it has seen countless times, easy to reproduce.

AI Struggles With This

Tap a card to see why
02, Yesterday's news
Its training data has a cutoff date, so it simply can't know what happened after that.
04, Moral judgment
It has no values or stake in the outcome, it only predicts agreeable-sounding words.
06, Detecting sarcasm
Tone depends on real-world context and intent that the text alone doesn't reveal.
08, Emotional understanding
It imitates empathy from patterns in text but doesn't actually feel or grasp emotion.
10, Recent sports scores
Same cutoff problem, it has no live connection to events from last week.
11, Humor tied to who/why
The joke depends on shared context and relationships the AI isn't part of.
15, Knowing when to stop
It has no social read on the room; it just keeps generating until told otherwise.
Cards 13 & 14 are ambiguous on purpose. 13 (personal family advice): AI can offer general frameworks but lacks your real context. 14 (argue either side): AI does this well, which is the problem, it has no actual position.
Phase 3, Activity 2
7 / 18
Catch the Mistake, 8 min

AI wrote this report.
Find what's wrong.

🏀

Sports

NBA scoring race, recent stats and records

🔬

Science

Deep-sea species discovery, 2025 expedition

🎤

Music

Sabrina Carpenter's 2025 world tour details

📱

Influencer

MrBeast / Feastables brand expansion in 2025

Choose one report as a group. Read it together. Flag anything that seems wrong, too vague, or suspicious. Be ready to share one finding and explain why you think AI got it wrong.

Phase 3, Answer Key
8 / 18
Catch the Mistake, the reveal

What was actually wrong in each report

Answers stay hidden until you tap, share each one after that group presents.
🏀
Sports
NBA Scoring Race
The planted error

"LeBron reached 45,000 career points, the second player in history." Both parts are false: LeBron is the first and only player past 40,000, and he was nowhere near 45,000 in early 2025. The rest (SGA ~32.7 PPG, the Luka-to-Lakers trade, Thunder 68–14) is true, which is exactly why the fake stat slips by.

🔬
Emerging Science
Deep-Sea Species
The planted error

It contradicts itself: the comb jelly is called "a new genus… unlike any in the phylum," then given the name Pleurobrachia ramifera, an existing genus. It also claims the "<25% of deep ocean mapped" figure has "remained stable since 2010," when seafloor mapping has grown a lot since. Names and institutions sound authoritative but don't check out.

🎤
Music & Tours
Sabrina Carpenter Tour
The planted error

Confidently invented specifics: the upcoming album "Freckles" and the fan-named ballad "Paperback" don't exist, and the precise figures ($210M gross, O2 "sold out in under four minutes," 85,000 UK attendance) are fabricated. The tour is real; AI filled the gaps with made-up numbers.

📱
Influencers
MrBeast / Feastables
The planted error

The "Feastables × Reese's collaboration" is fabricated, Reese's is a direct competitor (Hershey), so the partnership is implausible. The subscriber and view counts (350M subs, 180M views) and the "$500,000 chocolate factory challenge" are invented precision presented as fact.

Phase 4, Mini-Lecture
9 / 18
Where AI genuinely excels

What AI is actually good at

Speed at scale

Summarizing huge amounts of text, generating 20 ideas in 10 seconds, translating instantly.

🔍

Pattern recognition

Finding connections across thousands of documents or research papers that would take humans weeks.

🎭

Adapting style and format

Rewriting the same content five different ways, formal, casual, simple, technical, poetic.

💬

First drafts and starting points

Getting something on the page fast. AI excels at "good enough to edit," not "finished."

Phase 4, Mini-Lecture
10 / 18
Where AI consistently breaks down

What AI is actually bad at

📅

Recent events

AI's knowledge has a cutoff date. Anything recent, scores, news, new albums, is a gamble.

⚖️

Moral judgment

AI can argue any side of any ethical debate equally well. It has no actual values, just patterns.

🤷

Knowing when it's wrong

AI doesn't flag uncertainty unless prompted. It confidently states made-up facts the same way it states real ones.

👤

Your specific context

AI doesn't know you, your family, your school, or your situation. Generic advice can miss the point entirely.

Phase 4, Mini-Lecture
11 / 18
The reason behind all of this

AI doesn't know things.
It predicts words.

📚
Trained on text
Billions of words from the internet, books, articles
🎲
Predicts next words
Scores thousands of possible next words by probability
📢
Sounds confident
Regardless of whether it's right or completely wrong

AI doesn't know it's wrong, because it was never "knowing" anything in the first place. It was matching patterns. That's powerful. And it's a real limitation.

Phase 5, Prompt Tournament
12 / 18
The challenge

Build the prompt that stumps AI.

  • Setup
    Each pair writes one prompt designed to expose a real AI limitation. 2 min
  • 2v2 Round
    Pitch your idea in 60 seconds. The group of 4 votes. Winner advances. 3 min
  • 4v4 Round
    Merge groups. Same format. 60-second pitch. Vote. 3 min
  • 8v8 Round
    Keep merging until one class champion remains. 3 min

The judging standard: which prompt is most likely to expose a real AI limitation, not just confuse it with nonsense or trick it with wordplay.

Phase 6, Socratic Debrief
13 / 18
Was the tournament fair?

How did we pick a winner?
How does AI?

  • ? Was the tournament a fair way to decide which prompt was best?
  • ? Did the loudest or most confident pitch win, even if it wasn't the best idea?
  • ? Does that remind you of anything about how AI picks its answers?
Phase 6, The Parallel
14 / 18
What your tournament and AI have in common

Confidence ≠ Correctness

🏆 Your Tournament 🤖 How AI Picks
The most confidently pitched idea often wins early rounds The highest-probability word pattern wins, regardless of truth
Groups go along with the front-runner once momentum builds AI's answers are shaped by what was repeated most in training data
The winning idea feels right because everyone chose it AI sounds authoritative because that's how text in its training was written
But you can change your mind after the debrief AI cannot reflect on its own answers or update itself

When AI recommends a product, a song, or an answer, it's reflecting which voices were loudest in its training data. Just like the loudest room shaped your vote.

Phase 7, Live Demo
15 / 18
The moment of truth

Enter the winning prompt.

💻

Switch to the AI tool now.
Type the winning prompt live, don't paste it.
Read the response together.

Did it work?
What did AI do?
Does this change how
you'd use AI for this?
Phase 8, Exit Ticket
16 / 18
Individual, no talking, no phones

Last one: you, solo.

01 Write a persuasive essay on any topic
02 Know how your best friend is feeling right now
03 Find patterns across 10,000 customer reviews
04 Tell you if a news story is morally right or wrong
05 Translate a document from English to Japanese
06 Notice when a joke landed badly in the room
07 Explain photosynthesis 5 different ways
08 Tell you who won a game played last night
09 Draft a professional email in under 30 seconds
10 Give advice right for your specific family situation
🎯   Sort 5 into AI Strong, 5 into AI Weak, turn it in before you leave.
The Takeaway
17 / 18
What to remember from today

AI is a powerful thinking partner, not a thinking replacement. It's fast, broad, and tireless. It's also confidently wrong, shallow by default, and incapable of knowing you, your values, or your context.

🎯 Know when to use it and when not to
🧭 Know how to direct it, not just accept what it gives
🧠 Know when to trust your own brain instead
Up Next
18 / 18
This lesson is part of a series

Next time:
Is AI ethical?

🌍
Environmental Impact
What does running AI actually cost?
💼
Jobs and Labor
Who benefits and who doesn't?
🎨
Intellectual Property
Who owns what AI creates?