AI Coding — Applied Artificial Intelligence & Machine Learning

AI Coding: Understanding, Building, and Questioning Intelligent Systems

Grades K–12 | Aurora Curriculum Pack

Aurora AI Coding standard: Artificial intelligence is not magic — it is mathematics applied to data. Every AI system was built by humans, trained on data chosen by humans, and reflects the decisions and biases of those humans. A scholar who understands how AI works will use it critically. A scholar who does not understand it will be used by it.

How to use this sheet: This worksheet is organized into four grade bands. Use the section that matches your scholar's current level. Work through the concepts in order — each builds on the last.

Grade Band: K–2 — What Is AI?

Grades K–2

Key idea: AI stands for Artificial Intelligence. "Artificial" means made by humans, not natural. "Intelligence" means the ability to learn, solve problems, and make decisions. AI is a computer program that learns from examples — just like you learn from experience.

Activity 1 — Sorting by Example

AI learns by looking at many examples. Below are pictures described in words. Circle which group each belongs to (the AI is learning to tell cats from dogs):

Description	Cat or Dog?	How did you know?
Has whiskers, purrs, and says "meow"	_______________	_______________
Barks, wags its tail, fetches a ball	_______________	_______________
Has fur, climbs trees, hisses when scared	_______________	_______________
Has fur, loves to swim, follows commands	_______________	_______________

The clues you used to decide are called features. AI programs learn which features matter most from thousands of examples.

Activity 2 — Teaching the Computer

Imagine you are teaching a computer to recognize happy faces. Write 3 features (clues) you would tell it to look for:

What might go wrong if you only showed the computer 5 examples instead of 5,000?

Grade Band: 3–5 — Training Models and Recognizing Bias

Grades 3–5

Key idea: Machine learning is a type of AI where the computer improves at a task by studying examples (training data). If the training data has a problem, the AI will have the same problem — this is called bias.

Real example — documented bias in AI:
In 2018, Amazon built an AI to screen job applications. The AI was trained on 10 years of resumes from people Amazon had hired. Because most of those hires were men, the AI learned to score male applicants higher and penalize words like "women's" (as in "women's chess club"). Amazon discontinued the tool when the bias was discovered.

The lesson: An AI is only as fair as the data it was trained on.

Activity 1 — Training Data Analysis

A school wants to train an AI to recommend books to students. They plan to use the checkout records from the last 5 years to train it. Identify two potential bias problems with this plan:

How could they make the training data more fair?

Activity 2 — Unplugged Classifier

Your teacher will give you a set of 20 "fruit cards" (circles and squares, red and yellow). Without looking at the answers, create a rule that correctly sorts at least 15 out of 20. Write your rule:

How many did your rule get right? _______ out of 20

What would you change to improve your rule?

Grade Band: 6–8 — Prompt Engineering and AI Ethics

Grades 6–8

Key idea: Prompt engineering is the skill of giving clear, specific instructions to an AI language model to get useful, accurate, and appropriate outputs. It requires understanding how the AI "thinks" — pattern matching, not genuine understanding.

Aurora AI standard: AI language models do not "know" things. They predict which words are most likely to follow other words, based on their training data. A confident-sounding AI response can be entirely wrong. All AI outputs must be verified against primary sources.

Activity 1 — Prompt Comparison

Compare these two prompts and predict which will give a more useful response. Explain why.

Prompt A:

Tell me about Africa.

Prompt B:

Summarize three documented economic achievements of the Mali Empire between 1200-1500 CE. Include one primary source for each. Format as a numbered list.

Which prompt will produce a more useful result? _______

Why?

Rewrite Prompt A to be as specific as Prompt B on a topic of your choice:

Activity 2 — AI Ethics Case Study

A school district is considering using AI to predict which students are likely to drop out so teachers can intervene early. The AI would be trained on attendance records, grades, and zip code data.

List one argument FOR this use of AI:

List one argument AGAINST this use of AI (consider: who might be harmed? what assumptions does the model make?):

What safeguards would you require before this system could be used?

Grade Band: 9–12 — Machine Learning in Python and Capstone Project

Grades 9–12

Key idea: Supervised learning is a machine learning approach where a model is trained on labeled examples (input + correct answer) and learns to predict the correct answer for new inputs. The model is evaluated on data it has never seen before (test data).

Concept — The Train/Test Split

You never evaluate a model on the same data it trained on — that would be like giving a student the exact test questions to study. Standard practice:

from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(
    features, labels, test_size=0.2, random_state=42
)
# 80% of data used to train, 20% held back to evaluate

Activity 1 — Analyze a Model

A student trains a model to classify news articles as "reliable" or "unreliable." The model achieves 94% accuracy on training data but only 61% accuracy on test data. What does this tell you about the model?

What is this problem called? (Research: overfitting vs. underfitting)

What would you do to fix it?

Activity 2 — Capstone Project Plan

Design a machine learning project on a topic that matters to your community. Complete the planning document:

Planning Question	Your Answer
What problem are you solving?
What data will you collect? Where will you get it?
What type of ML model will you use? (classifier, regressor, etc.)
What bias risks does your dataset have? How will you address them?
How will you measure success?
Who could be harmed by an error in your model? How will you protect them?