How AI Thinks: The Birds and Bees of AI Answers
Students need to understand how AI works.
Lesson Plan: Birds & Bees of AI
- Where AI answers come from
- The difference between Google search and ChatGPT
- When to use AI and when to avoid it
Step 1: A brand new language for AI answers
Tell your students they are going to learn a brand new language the same way AI learns new things.
Show them a list of words and their job is to figure out
- Which words refer to birds
- Which words refers to cats and
- Which words refer to neither
Present this list of words to your students and ask which ones are the bird words.
You’ll sit in awkward silence and be met with blank stares.
After a few moments, acknowledge your students have no clue which ones are the bird words. Explain this is exactly how AI answers start out.
Everything these AIs learn they’ve had to be trained on.
Before your students can answer the “bird word” question, they need training data.
Give them some bird words and ask them to observe what’s happening in their brain as you show these bird words.
- Show the first bird word in this new language: Briz.
- Show the second bird word: Buitle.
- Show the third bird word: Bast.
Ask your students what the last bird word is. Tell them to write it down but don’t say it out loud, that you will count them down and all of them can shout it at once.
Count down 3 . . . 2 . . . 1 . . . GO! and your students will likely all shout “BEOL!”
Step 2: AI answers through pattern recognition
Explain the way their brain works to identify an answer is the same way AI answers a question.
Your students started making connections and seeing patterns as soon as you showed them the second or third word. They could understand you were giving them the words that start with B.
Tell your students they recognized the pattern, which is exactly how we train AIs.
Tell your students you’re moving on to the cat words, and same as with the bird words, you will show them a couple words and then you’ll count them down to tell you the last cat word.
Give them some cat words and ask them to observe what’s happening in their brain as you show these cat words.
- Show the first cat word in this new language: Schluggat.
- Show the second cat word: Fissat.
Count your students down to yell the third cat word 3 . . . 2 . . . 1 . . . GO! and your students will likely yell a variety of answers.
Many students will say “Lerat.”
This answer makes sense because it ends with “at” just like “cat” and the first two cat words.
This is incorrect, because they don’t take the time to think through all the patterns you’ve given them.
It turns out AIs can be overly aggressive in pattern matching just like students who pick Lerat.
Step 3: Training data for accurate AI answers
Use the following story to highlight that an AI is only as good as its training data. If the AI doesn’t have enough training data it can can make incorrect assumptions, and if it has biased or incorrect training data, it can produce erroneous results. The point to drive home is that training data is of utmost importance.
The story is about AI researchers who were trying to train AI to detect malignant moles from images. They used a lot of images of moles to train the AI, and the AI came up with an astounding conclusion:
Rulers cause cancer.
The AI learned that any picture with a ruler in it also contained a malignant mole. So it concluded that rulers caused cancer. What happened is the AI researchers trained the tool on a number of pictures of people’s benign moles that didn’t have rulers in them, but the malignant moles did have rulers in them to indicate the size of the mole.
The AI got overly aggressive and thought anytime it sees a ruler the mole must be cancerous.
Back to the cat words.
Tell students that Lorat is not the right cat word, and ask them to shout out what is the last cat word in 3 . . . 2 . . . 1 . . . and they will shout “WRATT!”
Highlight the pattern of double letters that ends with an “at” sound.
Tell students they now have all the information they need to be an AI after these two training exercises.
Step 4: A brand new word!
As a new AI, you want them to use the patterns they learned so far to generate a brand new word that means “flying cat.”
Give your students about 30 seconds, and tell them you’ll count down for them to shout their answer. Count them down from 3, and you’ll hear some words that start with “B” and have double letters and end with an “at” sound.
This is what you want because they are using their training to combine the attributes of bird and cat words. Talk about some of the words they’re sharing and how they made incorrect assumptions and/or produced erroneous results.
For instance, maybe a student says “bat.” It does start with a “b” and end with an “at” sound, but it doesn’t have any double letters.
Maybe another student says “bullet.” It does start with a “b” and have double letters, but it does not end with an “at” sound.
This is how generative AIs work – they learn some patterns and combine them.
Step 5: Google answers vs. AI answers
Explain to your students this is the difference between these new AIs and something like Google. Google is basically a dictionary. It is a database of gathered information from around the web. So when they ask Google for that information, it searches for information on websites and creates a database of them.
When you ask Google for a definition, it retrieves information about that word and gives it to you.
Now explain the difference of generative AI answers.
ChatGPT works from information it’s gathered across the web, but it’s not a dictionary.
Tell your students to think of a generative AI like ChatGPT like an incredible Lego builder. The Legos are not individual facts, but are patterns it observed. The AI builds a database of these patterns, and uses those patterns to generate brand new content that has never before been created.
As new generative AIs, your students never saw a word that means “flying cat” but they do know:
- Things that fly are birds, and bird words start with a “b”
- A cat word needs to contain double letters and end with an “at” sound
Tell them to combine these patterns and generate a brand new word from them.
For fun, give students 30 seconds to create a new word by combining these patterns. As they share, have fun celebrating their ability to use pattern recognition to create something the world has never known before.
Get the “The Birds & Bees of AI” Lesson Plan
We’ve created a detailed lesson plan for the “The Birds & Bees of AI: Where Do Answers Come From?” exercise to walk you and your students through the process step-by-step.
It’s free for any/all entrepreneurship teachers. Please feel free to share it.
Coming Soon…
In upcoming posts, we will be sharing more engaging AI exercises like this one!
Subscribe here to get lesson plans delivered in your inbox.
Want 30+ more engaging exercises?
