Building on Week 6
In Week 6, the key conceptual move at K–5 was replacing magical thinking with rule-following: computers do not know things, they follow rules. Decision trees gave that idea a concrete form students could build by hand.
Week 7 introduces a twist: what if the computer can figure out its own rules by looking at lots of examples? That is the seed of machine learning at the elementary level — and it is fully accessible without any mathematics. The move from "rules a person wrote" to "patterns a computer found in examples" is the most important conceptual extension available at this level.
Learning from Examples at the Elementary Level
Young students have already experienced learning from examples themselves — they learned to read by seeing many words, not by memorizing a complete rule set. That experience is the bridge.
The Sorting Game, Extended
If students built a decision tree in Week 6, return to it now with a new question: how did the computer learn which questions to put in the tree? The answer is that a person showed it many examples where they already knew the answer, and the computer looked for patterns in what separated one group from another. It did not understand the examples — it counted and compared.
A simple demonstration: show students ten pictures of cats and dogs and ask them to point out what they look at to tell the difference. Then tell them: a computer learning system does something similar — it looks at thousands of pictures where a human has already said "cat" or "dog," and it learns which visual patterns tend to appear with which label. It does not know what a cat is. It knows that certain patterns tend to co-occur with the label "cat."
The Practice Makes Better Framing
Elementary students understand that practicing something makes you better at it. Neural networks improve with practice in a way that maps to this intuition: they see an example, make a guess, find out if they were right, and adjust. Then they do it again — millions of times. The technical term for this is training, but the concept is just: getting better by doing over and over.
This framing is honest and age-appropriate. It does not require explaining weights or gradient descent. It gives students a mental model that is not wrong.
Large Language Models at the Elementary Level
Many K–5 students have already used tools like ChatGPT or AI writing assistants, either at home or through school. They have intuitions about these tools that are worth addressing directly.
The Most Important Thing to Know at This Level
LLMs predict the next word. They do not look things up. They do not know if what they say is true. They produce text that sounds like good answers because they have seen a lot of good answers, and they learned what good answers look and sound like.
A useful framing for K–5: imagine someone who has read every book in the library but cannot go outside. They can answer questions that sound exactly like questions in books. But if you ask them about something that just happened, or something that requires actual checking, they might give you a very confident answer that is just wrong. That is roughly what these tools do.
Classroom Conversations Worth Having
The most valuable K–5 conversation about AI tools is not about technical details — it is about the habit of checking. If an AI tool gives you an answer, what would you do to find out if it is right? Where could you look? Who could you ask? These are information literacy habits that apply everywhere and are reinforced, not replaced, by the AI context.
Resist the framing that AI tools are either "good" or "bad." The productive question is: for this task, does this tool help? And how would I know if what it gave me was right?
Connections to the Broader K–5 CS Curriculum
- Data and analysis: The idea that a computer learns from labeled examples connects directly to data literacy. Labeled data is just data where someone has already recorded what the answer is. Students who have done simple data investigations understand this instinctively.
- Digital citizenship: "AI tools can sound confident and be wrong" is one of the most practically useful things a K–5 student can know. It grounds every conversation about checking sources, evaluating information, and not treating a computer's output as automatically authoritative.
- Algorithms: Neural networks learn algorithms from data rather than having algorithms written by people. That distinction — programmed versus learned — is a natural extension of the algorithmic thinking strand.