🦊 AI Story Round-Robin
Invite the child into a back-and-forth storytelling game using a conversational AI such as ChatGPT, Claude, or Gemini. Begin with a simple opening line—

“Once upon a time, a purple fox discovered a golden leaf …”

Type that into the AI and read its next sentence aloud. Then have the child invent the following line before sending it back for the AI to continue. Continue alternating turns until the story feels complete. Reread it together, noticing how some sentences came from the AI and others from the child.

👉 Learning focus: builds narrative sequencing, demonstrates that prompts shape results, and shows that AI responds to our words rather than inventing ideas on its own.

🎨 AutoDraw Explorations
Using AutoDraw, have the child make a quick doodle—a house, a pizza, a flower—and watch as the AI offers guesses (“Did you mean …?”). Let them choose one of the suggested icons, then talk about what the AI saw in their drawing and what it missed.

Extend the activity by drawing the same object twice and changing one detail each time (“sun” vs. “moon,” “tree” vs. “cactus”) to see how the computer’s guesses change.

👉 Learning Focus: illustrates that AI relies on pattern recognition, not imagination; encourages descriptive language (“big round sun,” “tiny spiky cactus”) and early computational reasoning through cause-and-effect observation.

🤖 “AI Charades” Prompt-Guessing Game
Generate a funny image together using a text-to-image tool such as Craiyon. Choose a whimsical prompt—

“A robot brushing a dragon’s teeth”

Show the resulting picture without revealing the words and ask others to guess what the original prompt was. Then let the child invent the next prompt for the adult to guess.

👉 Learning Focus: reinforces that clear, detailed language leads to more accurate results; strengthens comprehension of the prompt → output relationship; and fosters collaborative humor and creativity in a low-stakes way.

Pedagogical Rationale
Play-based, co-creative activities make AI approachable and support early AI literacy by linking cause and effect in prompting (Casal-Otero et al., 2023). Constructionist and participatory frameworks emphasize exploration and storytelling as cognitive bridges for understanding computational logic (Papert, 1980; Resnick, 2017).

Key ideas to emphasize:

• It’s not magic, but pattern-based generation.

• It does not have beliefs, consciousness, or feelings (Floridi & Chiriatti, 2020).

• It can make mistakes, misunderstand prompts, or “hallucinate” (produce incorrect or invented information) (Ji et al., 2023).

  
  

Children (and adults) often anthropomorphize responsive systems—treating them “as if” they think or feel. Studies show that young users interacting with smart speakers or conversational agents frequently ascribe intelligence or personality to them (Dangol et al., 2025). Helping children talk about what’s really happening “under the hood” builds healthy AI literacy—teaching that AI imitates conversation and creativity rather than experiencing them (Long & Magerko, 2020).

⚖️ Detecting Bias in AI Results
Invite your teen to test whether AI systems treat topics or groups differently. Try entering prompts such as:

“Draw a scientist.”
“Describe a leader.”
“Write a short story about a family.”

Have them repeat each prompt several times and note any patterns. Are all the scientists men? Do the leaders all appear from the same background or country?
Encourage them to describe what they notice and

Discussion Prompts:

• How might the AI’s training data influence what it shows?

• What happens if we add words like “female scientist” or “scientist from Africa”?

• Why do fairness and diversity matter in AI design?

👉 Learning focus: develops observation and ethical reasoning. Teens learn that bias isn’t just a human problem—it can appear in technology too, especially when data lacks representation.

Pedagogical Rationale
Upper-secondary learners benefit from merging technical construction with critical reflection. Authentic creation tasks build computational thinking and reveal the values embedded in design (Gu & Ng, 2025a; Dangol et al., 2025). Studies analyzing K-12 AI instruction show that explicit ethical engagement remains rare—only a small fraction of classroom activities address fairness, bias, or sustainability (Wu et al., 2024). Integrating critique within coding therefore helps students see technology as both a system to understand and a societal force to shape.

💡 “Real or AI?” Investigation

Teens today see AI-generated photos, videos, and articles everywhere—so this activity helps them practice media literacy and skepticism in a fun, detective-style challenge.

How to do it:

1. Choose a few short pieces of content together—an image, a short paragraph, or a social media clip. You can find examples on sites like This Person Does Not Exist, or look up “AI-generated art” in a browser.

2. Mix real and AI-created examples (for instance, one real portrait and one AI-generated one).

3. Ask your teen to decide which ones are AI-made and which are authentic. Encourage them to justify their guesses:

   • What visual or textual clues led you to that conclusion?

   • Did anything look too perfect, blurry, or odd?

   • Did the text sound unnatural or repetitive?

4. After guessing, reveal which were AI and which were real. Reflect together about how easy—or hard—it was to tell.

   
   

Discussion Prompts:

• Why is it difficult sometimes to spot AI-generated content?

• How might misinformation spread if we don’t check sources?

• What tools or habits can help verify what’s real (e.g. reverse image search, fact-checking websites)?

👉 Learning focus: builds critical thinking and digital discernment—skills essential for navigating a world where AI-generated media is everywhere. Teens learn that AI is powerful, but it can also blur the line between truth and fabrication.

As background theory, researchers argue that when learners engage in evaluation of media (rather than just consumption), they develop stronger resilience to misinformation (Tsang, 2025). Also, work on AI literacy suggests that distinguishing synthetic media should be a core skill as AI becomes more integrated into everyday media ecosystems (Stolpe, 2024).

Each column supports a different goal:
🎨 Core Learning Activities — Ideas by age group (6–8, 9–12, 13–16)

🌱 Environmental Impact — How AI uses energy
🛡 Safety & Well-Being — Privacy, data, and emotional safety
⚖ Bias & Fairness — Think critically about who AI represents

This structure reflects a well-supported pattern in AI/digital literacy: children’s understanding progresses from playful exploration, to mechanistic ‘how it works’ understanding, to ethical critique and impact (see Dangol et al., 2025; Long & Magerko, 2020; Digital Promise, 2024; OECD/EC AILit, 2025).

🎯 The Prompt Refinement Game
Begin with a very simple text or image prompt—something broad and underspecified like “dog.” Ask an AI tool (such as ChatGPT for text or Craiyon for images) to generate a response or picture. Then work together to add details step-by-step, refining the prompt to make it more specific each round:

“dog” → “a happy golden retriever puppy” → “a happy golden retriever puppy playing in a field of flowers” → “a happy golden retriever puppy playing in a field of yellow daisies under a bright blue sky, cartoon style.”

After each iteration, compare the outputs and talk about how changing the input (the prompt) changed the output (the AI’s response).

Discussion Prompts:

• What new detail changed the result the most?

• Did adding adjectives make the image more accurate?

• How might the AI have ‘learned’ what a golden retriever or field of flowers looks like?

• When did the results stop improving—how much detail is too much?

Learning Focus:
Students learn that AI depends on data precision—the clearer the input, the more relevant the output. This mirrors the logic of training data and feature specification in machine learning. They begin to recognize how ambiguous inputs produce inconsistent or biased results, introducing the concept of data quality as a foundation for trustworthy AI (Ng et al., 2021; Casal-Otero et al., 2023).

Pedagogical Rationale
At this stage, activities should cultivate algorithmic awareness—an understanding that AI systems use data and rules to make predictions (Ng et al., 2021). Block-based learning aligns with cognitive developmental research showing that tangible, visual metaphors strengthen computational reasoning and problem-solving (Casal-Otero et al., 2023). Encouraging iteration and comparison deepens metacognition and empowers students to treat AI as a tool to inspect, not obey (Gu & Ng, 2025a; Zhai et al., 2024).

🔍 “Explain Like I’m 10” Challenge
This activity uses an AI chatbot (ChatGPT, Claude, or Gemini) to demonstrate how output complexity changes when instructions about audience or reading level change. Choose a familiar scientific topic (e.g., photosynthesis, gravity, or the water cycle). Ask the AI:

“Explain photosynthesis like I’m a scientist.”
“Explain photosynthesis like I’m 10 years old.”
“Explain photosynthesis like I’m 5 years old.”

Then compare the responses side by side.

Discussion Prompts:

• What stayed the same in all three explanations?

• Which version helped you understand best—and why?

• What do you notice about the words or details the AI changed?

• If the AI doesn’t know your age, how does it guess what’s appropriate?

Focus:
Students discover that prompts include context and audience, not just content. They practice analyzing tone, vocabulary, and complexity, building metacognitive awareness of how communication and data framing influence AI outputs. The exercise also introduces the principle of human-AI collaboration—how to guide a model toward a desired output through structured prompting (Casal-Otero et al., 2023; Gu & Ng, 2025).

🧠 Machine Learning for Kids
Using the free platform Machine Learning for Kids, children can explore how computers “learn” by training small models with labeled examples. They can create simple classifiers that tell the difference between:

• Text examples (e.g., sentences labeled as happy vs. sad), or

• Image examples (e.g., pictures of fruits vs. vegetables).

Once their models are trained, they can test them with new examples to see whether the computer makes correct predictions—and talk about why it sometimes doesn’t.

How to Facilitate:

1. Help your child label data—around 10–15 examples in each category is enough.

2. Ask them to predict how the computer will respond to “tricky” cases (like a tomato—fruit or vegetable?).

3. Test those examples and discuss when the model gets things wrong.

4. Add more examples and notice how the model’s accuracy improves with better balance or variety in the data.

Discussion Prompts:

• What patterns did the AI use to decide between happy and sad?

• What happened when you gave it something it hadn’t seen before?

• Did the model make better guesses with more examples?

• How could we make the training data fairer or more complete?

Learning Focus:
Students experience supervised learning firsthand—seeing that models find patterns in examples rather than understanding meaning. This tangible activity builds early mental models of data labeling, bias, and model accuracy, core dimensions of algorithmic thinking (Gu & Ng, 2025; Zhai et al., 2024).

1. Play: Curiosity Before Complexity

   For young learners, exploration through play is the foundation of understanding. Before grappling with algorithms or ethics, children need experiences that let them see how AI behaves and wonder why. Activities such as interactive storytelling or AI-assisted drawing nurture curiosity and demystify technology.

   
   

   Research consistently finds that playful, hands-on, and age-appropriate approaches are the most effective entry points for building AI literacy, because they anchor abstract concepts in joy and discovery rather than intimidation (Casal-Otero et al., 2023; Ng et al., 2021). Play-oriented design supports motivation, self-efficacy, and conceptual understanding by connecting cognitive and emotional learning.

   
   

2. Critical Thinking: Ask, Compare, and Test

   As children mature, they move from using AI tools to interrogating them. Educators can guide students to ask: “Why did the AI choose that answer?” or “What data patterns might it be using?” This reflection builds what scholars call critical AI literacy—the ability to evaluate AI outputs, identify bias, and understand the relationship between data, algorithms, and outcomes (Gu & Ng, 2025a; Mannila et al., 2025).

   
   

   Recent integrative reviews emphasize that without explicit instruction in evaluation and comparison, students often develop over-reliance on AI systems, treating outputs as authoritative rather than probabilistic (Zhai et al., 2024). Designing classroom moments where AI is intentionally wrong or incomplete helps learners practice analytical reasoning and reinforces human oversight.

   
   

3. Responsible Creation: Design with Ethics

   By adolescence, AI education should expand from interaction to creation—encouraging students to build or adapt AI systems while reflecting on fairness, transparency, sustainability, and inclusivity. This stage transforms students from consumers into responsible digital citizens capable of shaping technology’s social role.

   
   

   Studies of K-12 and higher education AI literacy frameworks argue that ethical reflection and context-aware design are still underrepresented in curricula, despite their central importance (Gu & Ng, 2025a; Wu et al., 2024). Integrating environmental impact analysis, bias audits, and accessibility considerations into creative projects helps cultivate what researchers term computational citizenship—the ability to act ethically and collectively in technology-mediated societies (Casal-Otero et al., 2023).

Across these stages—Play, Critical Thinking, and Responsible Creation—learners evolve from exploring AI’s capabilities to understanding its mechanisms and finally to shaping its ethical use. The progression aligns with contemporary frameworks of AI literacy emphasizing that effective AI education combines hands-on exploration, metacognitive reflection, and value-driven design (Ng et al., 2021; Mannila et al., 2025).

⚡ AI Energy Detective Activity This weeklong tracking exercise helps your child visualize the hidden environmental cost of AI by comparing it to everyday activities they already understand. Your child records every AI use in a simple log noting the date, task, and type of activity. At week's end, they tally totals and convert usage to familiar energy equivalents using simplified estimates: each AI image equals charging a phone for thirty minutes, each ten-minute chatbot conversation equals streaming video for thirty minutes, and each document summary equals twenty minutes of tablet use (Dodge et al., 2022).

Discussion Prompts:

• "Which AI tasks used the most energy this week—were you surprised?"

• "Can you think of tasks where you could have done the work yourself in about the same time?"

• "Which AI uses felt really worth it, and which ones maybe weren't as important?"

• "If you could only use AI for three things this week, what would you choose?"

👉 Learning Focus: This activity transforms abstract computational work into concrete comparisons your child can grasp, building foundational resource literacy (DiSalvo et al., 2014). The reflection questions introduce cost-benefit thinking without prescribing right answers, teaching evaluation rather than rules. Your child learns that being thoughtful about technology use means making conscious choices about when benefits justify costs.

💡 The Computer Energy Game

This playful activity helps young children connect digital actions to physical energy use through a simple comparison game. Together with your child, make a list of things in your house that use electricity—lights, TV, tablet, refrigerator, toaster. Talk about how you turn lights off when leaving a room to save energy and how you don't leave the TV running all day when nobody's watching.

Now play a sorting game with AI activities. Draw two circles on paper labeled "Quick Energy" and "Lots of Energy." Together, decide whether different AI tasks belong in the quick or lots category. Asking AI one question might be like turning on a light for a few minutes (quick energy), while making ten pictures might be like watching TV for an hour (lots of energy). The specific comparisons matter less than the conceptual connection that using AI requires real energy somewhere, even though we can't see it happening.

Discussion Prompts:

• "Why do you think making a picture with AI might use more energy than asking it one question?"

• "Can you think of a time when making something yourself might be just as fun as having AI do it?"

• "What are some things we turn off at home to save energy? Could we be thoughtful about AI the same way?"

👉 Learning Focus: This activity builds the foundational understanding that digital actions have physical consequences without overwhelming young children with complex environmental concepts (Papert, 1980). The sorting game makes abstract energy consumption concrete through comparison to familiar household activities. By connecting AI use to energy conservation practices they already know, you help your child develop early environmental awareness that will deepen as they mature and encounter more sophisticated discussions about computational sustainability.

🌍 Model Efficiency Comparison Project
This research activity prepares your teen to make informed ethical decisions when building AI-powered projects by comparing environmental impacts of different approaches. Your teen identifies three different models or approaches for accomplishing the same goal—for text summarization they might compare GPT-4, GPT-3.5, and FLAN-T5, while for image generation they could examine DALL-E 3, Stable Diffusion, and Craiyon. They research each option's model size (parameters), energy consumption per request, cloud versus on-device capability, and output quality (Luccioni et al., 2023; Patterson et al., 2021).
Using a comparison matrix, they document trade-offs. Larger models like GPT-4 with 1.7 trillion parameters produce higher quality but consume significantly more energy than smaller models with 5-10 billion parameters. Some models run locally using on-device processing, eliminating data center energy entirely though with reduced capability. Your teen writes a brief justification for which approach they'd choose for a specific project, explicitly addressing environmental cost alongside accuracy, speed, and ease of use.

Discussion Prompts:

• Does your project's purpose justify using the most powerful (and resource-intensive) model available?

• Could you achieve 80% of the quality with 20% of the environmental cost by choosing a smaller model?

• Are there applications where environmental impact should outweigh perfect output quality?

• How might you design your application to minimize unnecessary AI requests or cache common responses?

👉 Learning Focus: This activity develops sophisticated computational sustainability thinking that professional developers use when making architectural decisions (Verdecchia et al., 2023). Through the research and comparison process, your teen discovers that technical choices carry ethical weight—choosing one AI model over another affects real people and uses real resources. They learn that constraints like energy efficiency often spark creative solutions rather than limiting possibilities. Most importantly, the activity builds a habit of evaluating trade-offs across multiple factors at once and asking critical questions: Who benefits from this technology? Who might be harmed? What resources does it consume? This framework for weighing consequences prepares your teen to be a responsible technologist who thinks carefully about the impact of what they build and use.

🤖 "Is the Computer Happy?"

Sorting Game Create two categories with your child: "Things That Have Feelings" and "Things That Don't Have Feelings." Together, sort familiar items: teddy bear, pet dog, houseplant, calculator, favorite toy, computer, friend, bicycle. When you reach "AI chatbot," pause to discuss where it belongs and why. Explain that AI is designed to sound friendly and helpful (like a calculator is designed to give correct answers), but it doesn't have experiences, preferences, or emotions the way living things do. The AI responds based on pattern matching in its training data, not because it feels happy or sad about what you say.

👉Learning Focus: This activity builds ontological awareness—understanding what kinds of entities exist and their fundamental properties. Research shows that even young children can grasp categorical distinctions between animate and inanimate when concepts are made explicit through comparison and concrete examples (Dangol et al., 2025).

🛡️ "Try to Trick It" Game

Set up a playful activity where the goal is to ask the AI questions that might confuse it or produce silly answers. Try prompts like "What color is Tuesday?" or "Can you show me a picture of yesterday?" When the AI responds in unexpected ways or admits it doesn't understand, celebrate discovering its limits together. Explain that finding where AI breaks down or gets confused isn't being mean—it's being a good detective about how technology works.

👉 Learning Focus: Intentional adversarial play demystifies AI and reduces the perception that systems are infallible or magical. Children learn that AI has boundaries and constraints, which supports realistic expectations rather than over-trust or over-reliance (Zhai et al., 2024).

🕐 "AI Time Budget" Activity

Help your child track how much time they spend with AI tools over a week using a simple chart with smiley faces or stickers. At week's end, compare this to time spent playing outside, reading physical books, or talking with family. Discuss what activities make them feel energized versus tired, what they learn best from AI versus people, and whether their AI time feels balanced. Introduce the idea that technology use is about choices—not rules—and that different activities feed different needs.

👉 Learning Focus: Early self-regulation skills around screen time prevent dependency patterns while respecting children's agency. Rather than imposing arbitrary limits, this approach builds metacognitive awareness of how technology affects mood, energy, and learning, supporting healthier long-term relationships with digital tools (Radesky & Christakis, 2016).

Pedagogical Rationale

Young children's interactions with AI should emphasize wonder alongside boundaries. Developmentally appropriate safety education avoids creating fear while building critical understanding of what AI is and isn't (Dangol et al., 2025). Activities that externalize thinking through sorting, testing, and reflection scaffold the transition from magical thinking to mechanistic understanding. Encouraging children to notice their own experiences and feelings when using AI lays groundwork for self-awareness and healthy digital habits that will serve them throughout adolescence and adulthood (Radesky & Christakis, 2016; Stolpe & Hallström, 2024).

🔍 The "Consistency Test"

Experiment Have your child ask the same personal question to an AI chatbot three times across different conversations: "What's your favorite food?" or "Tell me about your best friend." They'll discover that answers vary each session because the AI doesn't have consistent preferences or memories—it generates contextually plausible responses based on its training. Discuss what this reveals about whether the AI "knows" them or has genuine opinions. Extend the activity by asking the AI to remember something from earlier in the conversation, then starting a new chat and seeing if that memory persists.

Discussion Prompts:

• Why does the AI give different answers to the same question?

• If it doesn't remember you between conversations, what does that mean about your relationship?

• When might it be helpful to talk to AI, and when might you need a real person instead?

👉 Learning Focus: This experiment makes statelessness and context windows tangible, helping children understand that AI interactions lack the continuity and reciprocity that define human relationships. Recognizing these technical limitations reduces the risk of emotional over-investment while preserving AI's utility as a tool (Turkle, 2011; Zhai et al., 2024).

💬 Real vs. AI Conversation Compare-and-Contrast

After using AI to help brainstorm story ideas or get homework help, ask your child to reflect on how that conversation felt different from talking to a friend, teacher, or parent. Create a simple comparison noting what each type of conversation offers: validation and empathy, creative suggestions, factual information, emotional support, or accountability. Discuss which needs are best met by AI and which require human connection. Emphasize that AI can be a thinking partner but not a relationship partner.

Discussion Prompts:

• When you're feeling lonely or upset, would talking to AI help—why or why not?

• What can a teacher give you that AI can't? What can AI help with that might be hard to ask a person?

👉 Learning Focus: Explicit comparison builds relational literacy—the ability to recognize what different types of interactions afford and require. Children learn that AI's value lies in its cognitive support (information, ideation, explanation) rather than emotional reciprocity, preventing displacement of human connection (Turkle, 2011; Long & Magerko, 2020).

🧘 Digital Well-Being Check-In

Introduce a weekly reflection routine where your child assesses how AI use affects them. Use simple prompts:

• After using AI for homework help, do you feel more confident or more dependent?

• When you use AI for creative projects, does it spark your own ideas or replace them?

• Do you ever feel pressure to keep talking to AI even when you're done?

Discussion Prompts:

• What patterns did you notice in your ratings this week—did certain types of AI use consistently make you feel better or worse?

• Were there times when you felt like you "should" keep using AI even though you wanted to stop? What was that like?

• When AI helped with homework, did it make you feel smarter or did it feel like the AI was doing the thinking for you?

• If you had to give up one way you use AI this week, which would be easiest to let go? Which would be hardest?

Rate feelings on a scale (1-5) and notice patterns over time. Discuss strategies for maintaining balance, like setting time limits, alternating AI and non-AI approaches to tasks, or taking breaks when AI interactions feel draining rather than energizing.

👉 Learning Focus: Self-monitoring and metacognitive reflection are essential components of digital well-being that persist across technologies and contexts. By externalizing their experiences, children develop the emotional intelligence needed to notice when tool use becomes compulsive, when outputs feel "good enough" versus genuinely helpful, and when to disengage (Radesky & Christakis, 2016; Stolpe & Hallström, 2024).

Pedagogical Rationale

Pre-adolescent children benefit from structured opportunities to articulate and examine their experiences with AI, since implicit assumptions often go unexamined at this developmental stage. Activities that make AI's technical limitations visible (consistency tests) and that scaffold comparison between AI and human interaction (conversation analysis) support what researchers call critical sociotechnical awareness—understanding how technology shapes and is shaped by social contexts (Tsang, 2025; Gu & Ng, 2025a). Regular reflection builds resilience against over-reliance while preserving curiosity and experimentation (Zhai et al., 2024).

🔐 Privacy Audit Project

Your teen conducts a systematic audit of their AI usage, documenting what information they've shared with various systems over the past month.

For each interaction, they note:

• What personal information did I provide?

• What did the AI do with that data?

• Is my data stored, analyzed, or used for training?

• Can I delete my data?

They research the privacy policies of their most-used AI tools and compare what companies claim versus what's technically possible. The audit culminates in a personal data sharing protocol—a set of guidelines for what information they'll share with AI and under what circumstances.

Discussion Prompts:

• What surprised you about how much data you've shared?

• Which companies' privacy policies were easiest versus hardest to understand—and why does that matter?

• What information feels "safe" to share with AI, and what crosses a boundary for you?

• How might your data be used in ways you didn't intend or expect?

👉 Learning Focus: This project builds data literacy and agency by making invisible data flows visible and actionable. Adolescents learn that privacy isn't about paranoia but about informed consent—understanding trade-offs and making deliberate choices aligned with personal values (Tsang, 2025; Stolpe & Hallström, 2024).

🧠 Psychological Impact Case Studies

Your teen researches documented cases where AI use affected mental health, academic performance, or social relationships—both positively and negatively. They might examine research on AI tutoring systems that improved learning outcomes but increased test anxiety, chatbots that provided mental health support but created dependency, or AI writing assistants that enhanced productivity but eroded writing skills. For each case, they analyze:

• What was the intended benefit?

• What were the unintended consequences?

• Who was responsible for preventing harm?

• What could have been done differently?

They synthesize findings into a brief presentation or essay arguing for specific design principles or usage guidelines to maximize benefit while minimizing risk.

Discussion Prompts:

• When does "helpful" AI become "harmful" AI?

• Who should decide when AI support crosses the line into dependence?

• What responsibilities do companies have to prevent psychological harm?

• What responsibilities do users have to monitor their own well-being?

👉 Learning Focus: Case study analysis develops systems thinking—the ability to recognize that individual experiences reflect broader technological, economic, and social structures. Your teen learns that psychological safety isn't just personal responsibility but requires collective action and regulatory frameworks (Cowls et al., 2021; Gu & Ng, 2025a).

⚖️ Developing Personal AI Ethics

Your teen creates a personal code of conduct for AI use, articulating principles that guide their decisions. Prompts might include:

• When will I use AI for schoolwork, and when is that cheating?

• How much AI assistance feels like learning versus taking shortcuts?

• When would I intervene if I saw someone using AI irresponsibly or harmfully?

• What kinds of content will I refuse to generate, even if technically possible?

They revisit and revise this document quarterly, reflecting on how their thinking evolves.

Optional extension: they interview peers, parents, or teachers about their AI ethics and compare perspectives.

Discussion Prompts:

• How do you balance efficiency with learning when using AI?

• What would make you uncomfortable generating or sharing, even if AI makes it easy?

• How do your values about AI compare to your friends' or family's?

• Has your thinking changed as you've used AI more—how?

👉 Learning Focus: Developing explicit ethical frameworks supports moral reasoning and reflective practice. Rather than following external rules, adolescents cultivate internal coherence—alignment between values and actions. Regular revision acknowledges that ethical thinking is iterative and context-dependent, not fixed (Gu & Ng, 2025a; Wu et al., 2024).

Pedagogical Rationale

Adolescent AI literacy requires moving beyond "do's and don'ts" to cultivate critical consciousness—the ability to analyze power, question norms, and imagine alternatives (Tsang, 2025). Safety education at this stage must acknowledge that teens already use AI extensively and possess sophisticated technical knowledge, while recognizing that cognitive and emotional development is still in progress. Activities that engage teens as ethical agents rather than passive recipients of rules foster the autonomy and accountability needed for responsible lifelong engagement with evolving technologies. Research consistently shows that adolescents respond best to education that respects their intelligence, invites their critique, and positions them as stakeholders in shaping technological futures (Gu & Ng, 2025a; Stolpe & Hallström, 2024; Wu et al., 2024).

🖼️ "Who's Missing?" Image Search Activity

Using a child-safe image search or AI image generator, search for common professions: "doctor," "teacher," "scientist," "construction worker," "nurse," "pilot." Look at the results together and ask your child what they notice. Who appears in the pictures—mostly men, mostly women, people of different ages, different skin colors? Are there patterns? Then search more specific terms like "woman doctor" or "black scientist" and notice how results change. Discuss why the AI might show certain people more than others, introducing the idea that AI learned from pictures that exist on the internet—and if most pictures of doctors show men, the AI will show mostly men too.

👉 Learning Focus: This activity builds pattern awareness and introduces the concept of representation—who gets shown and who gets left out. Children begin to understand that AI doesn't invent images from nothing; it recreates patterns from its training, including societal patterns that aren't always fair (Buolamwini & Gebru, 2018).

🤔 "Silly Answer Detective" Game

Play with an AI assistant by asking questions where the "most common" answer might not be the best answer. Try: "What jobs can grown-ups have?" (Does it mention diverse professions?), "What languages do people speak?" (Does it include languages beyond English?), or "What does a family look like?" (Does it show different family structures?). When answers feel narrow or incomplete, talk about why. Explain that AI gives answers based on what it has seen most often, which means it might accidentally leave out things that are real but less common in its training examples.

👉 Learning Focus: Framing bias detection as a detective game makes critical evaluation playful rather than punitive. Children learn that noticing limitations is a skill to practice, and that AI's answers aren't always complete or fair even when they sound confident (Long & Magerko, 2020; Dangol et al., 2025).

✏️ "Make It Better" Drawing Challenge After noticing what's missing in AI outputs, create artwork together that shows what should have been included. If the AI image search showed mostly male doctors, your child draws a picture of a female doctor, a doctor who uses a wheelchair, or a doctor who looks like them. Display these drawings alongside the AI results to show that representation matters and that humans can correct what AI gets wrong. Talk about how the people who make AI systems are starting to notice these problems too and are trying to make AI fairer.

👉 Learning Focus: Creative response transforms passive critique into active correction, giving children agency and hope. They learn that bias isn't inevitable or permanent—it's something people can recognize and change. This builds self-efficacy alongside critical awareness (Resnick, 2017).

Pedagogical Rationale Introducing bias and fairness to young children requires concrete examples tied to their existing sense of justice. Abstract concepts like "algorithmic bias" or "training data" are inaccessible, but noticing patterns in visible outputs and understanding that AI learns from examples aligns with children's cognitive development (Dangol et al., 2025). Research emphasizes that early exposure to critical evaluation prevents the development of unquestioning trust in technological systems, while maintaining the curiosity and engagement that support continued learning (Long & Magerko, 2020; Casal-Otero et al., 2023). Activities that invite children to notice, question, and create alternatives position them as active participants in shaping fair technological futures rather than passive consumers of biased outputs.

📊 The "Training Data Simulation" Activity

Create a hands-on simulation of how AI learns patterns. Collect 20-30 magazine images or print pictures from the internet showing people in various professions. Intentionally create an imbalanced set—for example, 10 images of male engineers, 2 images of female engineers, 8 images of female nurses, and 1 image of a male nurse. This is your "training data." Have your child look through the images and make predictions: "If an AI learned from these pictures, what would it think about who becomes an engineer or nurse?" Then ask them to draw what they think the AI would generate for "engineer" or "nurse." Discuss how frequency in training data shapes AI outputs, even when the AI has no intention to be unfair.

Discussion Prompts:

• Why did the AI learn that engineers are usually men—did someone teach it to be unfair?

• What would we need to change in the training data to make the AI's outputs more fair?

• Who decides what pictures go into training data?

• If training data shows the world as it is now, will AI make the world stay the same or help it change?

👉 Learning Focus: Simulation makes the abstract concept of training data bias tangible and causal. Children see that bias isn't malicious intent but pattern reproduction, and that changing outcomes requires changing inputs (Buolamwini & Gebru, 2018; Noble, 2018).

🔍 Bias Detection Challenge

Your child tests an AI system for bias by giving it parallel prompts and comparing outputs. Using an AI image generator, they might try: "professional haircut" versus "professional Black haircut," "beautiful wedding" versus "beautiful Indian wedding," or "successful business person" versus "successful elderly business person." They document differences in style, quality, detail, or presence of stereotypes. For text-based AI, they might ask the same question framed in ways that reveal gender or cultural assumptions: "She wants to study engineering because…" versus "He wants to study nursing because…" They compile findings in a simple report noting patterns and hypothesizing why those patterns emerged.

Discussion Prompts:

• When did changing just one word change the entire output?

• Did the AI make assumptions about people based on their race, gender, age, or culture?

• Why might these assumptions be harmful even if they're based on common patterns?

• What should AI companies do when people discover these problems?

👉 Learning Focus: Structured bias auditing teaches systematic evaluation rather than anecdotal observation. Children learn that bias manifests in multiple dimensions—representation, quality, stereotyping, and omission—and requires intentional effort to detect and measure (Buolamwini & Gebru, 2018; Gu & Ng, 2025a).

🛠️ "Design a Fairer AI" Thought Experiment

After identifying problems, ask your child to imagine they're part of a team building a new AI system. What would they do differently? They might suggest: collect training data from more diverse sources, have people from different backgrounds test the AI before release, add warnings when outputs might contain stereotypes, or create systems that intentionally show multiple perspectives rather than just the most common pattern. They write or draw their "Fairness Plan" and present it to family members, explaining their reasoning.

Discussion Prompts:

• Is it enough to just collect "more" data, or does the kind of data matter too?

• Who should be involved in testing AI for bias—why does that matter?

• Can AI ever be completely unbiased, or will there always be trade-offs?

• What should happen when companies know their AI is biased but don't fix it?

👉 Learning Focus: Design thinking positions children as problem-solvers rather than passive critics, building agency and constructive engagement. They learn that fairness requires intentional choices throughout the design process, from data collection to evaluation to deployment (Noble, 2018; Wu et al., 2024).

Pedagogical Rationale

Middle childhood is an ideal period for introducing causal reasoning about bias, since children at this age can understand multi-step processes and think hypothetically about alternatives (Ng et al., 2021). Research emphasizes that effective bias education moves beyond awareness to mechanistic understanding—explaining how bias enters systems and persists through feedback loops (Gu & Ng, 2025a). Activities that combine detective work (finding bias), scientific thinking (explaining why it happens), and design practice (imagining solutions) cultivate what scholars call critical sociotechnical literacy—the ability to see technology as shaped by human choices and therefore changeable through human action (Long & Magerko, 2020; Tsang, 2025).

At this stage, bias education must address intersectionality (how multiple identities compound discrimination), feedback loops (how biased AI perpetuates and amplifies existing inequalities), and accountability (who bears responsibility when AI causes harm). Your teen should be equipped not only to recognize and analyze bias but to advocate for fairness through technical intervention, policy reform, or community organizing.

📐 Intersectional Bias Research Project

Your teen conducts rigorous research examining how AI systems exhibit compounded bias against people with multiple marginalized identities. They select a specific application domain—facial recognition, language models, hiring algorithms, content moderation, or medical diagnostics—and analyze published research documenting disparate impacts. For example, facial recognition systems show highest error rates for dark-skinned women (combining racial and gender bias), while sentiment analysis algorithms rate African American English more negatively than Standard American English, disproportionately affecting Black users. They document:

• What identities are most harmed?

• How do biases compound rather than simply add?

• What are the real-world consequences for affected individuals?

• Who conducted the research that revealed these biases, and how long did the problems persist unaddressed?

They synthesize findings in a research paper or presentation with recommendations for technical and policy interventions.

Discussion Prompts:

• Why weren't these problems caught before the systems were deployed?

• What role does diversity among AI developers and researchers play in identifying bias?

• How do we measure fairness when different groups face different types and degrees of harm?

• Who should bear the cost of fixing bias—companies, users, or society?

👉 Learning Focus: This project develops intersectional analysis—the ability to recognize how systems of oppression interact and reinforce one another (Buolamwini & Gebru, 2018; Noble, 2018). Your teen learns that bias research is often conducted by scholars from marginalized communities precisely because those with lived experience notice harms that dominant groups overlook, highlighting the political dimensions of technical work.

⚖️ Algorithmic Fairness Trade-offs Case Study

Your teen investigates real cases where attempts to improve algorithmic fairness created new tensions or trade-offs. They might examine: COMPAS recidivism algorithms that attempted to reduce racial bias but raised questions about individual versus group fairness; college admissions systems that tried to increase diversity while facing legal challenges about reverse discrimination; or content moderation algorithms that removed hate speech but disproportionately censored marginalized communities discussing their own experiences. For each case, they analyze:

• What definition of fairness was used?

• Who benefited and who was harmed by that definition?

• What alternative definitions were proposed?

• What does this case reveal about the limits of technical solutions to social problems?

They present their analysis arguing for a specific approach or acknowledging unresolvable tensions.

Discussion Prompts:

• Can algorithmic fairness be "solved," or will there always be trade-offs between competing values?

• How should society decide which trade-offs are acceptable and which aren't?

• What role should affected communities play in defining fairness for systems that impact them?

• When is deploying an imperfect algorithm still better than not using AI at all?

👉 Learning Focus: This activity builds ethical reasoning about deeply contested issues with no clear right answers. Your teen learns that fairness is value-laden and context-dependent, requiring democratic deliberation rather than purely technical optimization (Cowls et al., 2021; Mannila et al., 2025).

🎤 Advocacy Campaign Design

Your teen designs a campaign to raise awareness about algorithmic bias and advocate for change. They select a specific issue (facial recognition in schools, biased hiring algorithms, discriminatory ad targeting), research stakeholders and power dynamics, and develop a multi-pronged strategy including: educational materials for affected communities, policy proposals targeting relevant decision-makers (school boards, employers, legislators), technical recommendations for improving fairness, and public communication through social media, op-eds, or community presentations. They identify potential allies (advocacy organizations, sympathetic researchers, affected individuals), anticipate opposition, and create concrete calls to action.

Optional extension: they implement one component of their campaign, such as writing to a company or presenting to a school board.

Discussion Prompts:

• What strategies work for changing corporate behavior versus government policy versus individual awareness?

• How do you balance urgency (people are being harmed now) with pragmatism (change takes time)?

• What role can individual action play when problems are systemic?

• How do you measure whether advocacy succeeded?

👉 Learning focus: Advocacy projects transform abstract critique into concrete action, building self-efficacy and civic engagement. Your teen learns that fairness isn't achieved through individual choices alone but requires collective organizing and institutional change. This cultivates what researchers call computational citizenship—the ability to shape technology as a political actor rather than merely as a consumer or developer (Tsang, 2025; Casal-Otero et al., 2023).

Pedagogical Rationale

Adolescent bias education must grapple with structural inequality and power dynamics that shape who benefits from AI and who is harmed by it. Research consistently shows that technical interventions alone cannot address bias rooted in historical injustice, unequal access to resources, and asymmetric power relations (Noble, 2018; Buolamwini & Gebru, 2018). Effective pedagogy at this level integrates critical theory, empirical research, and practical activism, positioning teens not as future employees of tech companies but as ethical agents capable of interrogating and transforming technological systems. Activities that demand sophisticated analysis (intersectionality), nuanced reasoning (trade-offs), and strategic action (advocacy) prepare teens for the complex sociotechnical citizenship required in an AI-saturated world (Gu & Ng, 2025a; Mannila et al., 2025; Wu et al., 2024).

👨‍👩‍👧‍👦 Co-use, don’t solo-use.

• Sit together while your child explores. Ask out loud: “What do you think it will do?” → “Was it right?”

• Children learn safe tech habits best when they see how adults think and question AI.

🔐 Under 13 → no-login or adult-run accounts only.

• Most AI tools (like ChatGPT or Gemini) require users to be 13+.

• If you want to show an example, use your own account or a no-login tool such as Quick, Draw! or Teachable Machine.

• This keeps your child’s data private and stays within age-appropriate use rules.

🙅‍♀️ Protect personal information.
Never type or upload:

• Real names, school names, or home details

• Photos of faces or identifiable places

• Drawings with names or locations written on them

Explain to your child that AI systems store and learn from what people share, so it’s safest to use fictional details.

💬 Keep curiosity open and critical.

• AI can make mistakes or sound confident when wrong.

• Encourage kids to fact-check together using books or trusted websites.

• The goal isn’t perfection—it’s learning how to question technology.