Stage 4: AI literacy for ages 14 to 15

1. AI Systems, Autonomy and Responsibility

I can map an AI-supported decision as a socio-technical system and design human authority that can really stop it.

From Automated Task to Socio-Technical System · Design Meaningful Human Authority

2. Research Claims and Reproducibility

I can trace a technology claim to its evidence, recompute the number from the data and report the limits honestly.

Trace a Claim to Its Evidence · Reproduce, Replicate and Report

3. Systems Engineering and Requirements

I can write testable requirements, map dependencies and failure modes, and keep safety constraints in the design.

From Problem Statement to Testable Requirement · Dependencies, Failure Modes and Resilience

4. Human-Centred Product Design

I can run ethical discovery, state a problem without a built-in solution and test whether a design works for real people.

Discover the Problem Before Choosing AI · Build, Test and Revise for Real Use

5. Experiments, Causality and Evidence

I can tell correlation from causal evidence and run a fair randomised comparison with a permutation test.

Correlation, Causality and Alternative Explanations · Design a Fair and Reproducible Experiment

6. Data Pipelines and Governance

I can trace data through a pipeline, find where rows go missing and assign who decides about the data.

Engineer a Traceable Data Pipeline · Create a Data Governance Plan

7. Statistics, Sampling and Uncertainty

I can simulate sampling, build a bootstrap interval and spot charts and averages that overstate certainty.

Samples, Distributions and What the Average Hides · Uncertainty, Confidence and Misleading Visuals

8. Representation, Features and Data Leakage

I can choose features for a real model, catch a leaking feature and explain why validation scores can lie.

How Representation Shapes Learning · Detect Leakage, Proxies and Shortcuts

9. Training, Validation and Generalisation

I can train a model with separate training, validation and test data, read a loss curve and test it on a shifted district.

Train, Validate and Test Without Fooling Yourself · Overfitting, Shift and Robust Generalisation

10. Model Evaluation, Calibration and Robustness

I can set a threshold from the cost of each error, read a reliability chart and write an evidence-based model card.

Choose Metrics for the Decision, Not the Leaderboard · Calibration, Stress Testing and the Model Card

11. Neural Networks and Embeddings

I can train a two-layer network that beats a straight line and judge what word embeddings capture or distort.

From Weighted Signals to Learned Representations · Embeddings, Similarity and Representation Risk

12. How Language Models Work

I can explain tokens, next-token prediction and temperature with a real tiny model and evaluate example AI answers with a rubric.

Tokens, Attention and Next-Token Generation · Evaluate Language-Model Outputs Systematically

13. Retrieval, Grounding and Knowledge Systems

I can design search with access and freshness rules, build an answer only from retrieved passages and audit its citations.

Design Retrieval Before Generation · Build and Audit a Grounded Answering System

14. Agents, Tools and Orchestration

I can configure a sandbox agent's permissions, limits and kill switch, read its log and recover the system.

Architecture of an Agentic Workflow · Control, Evaluate and Recover Agent Behaviour

15. AI Security and Adversarial Testing

I can threat-model an AI system, red-team it safely and choose layered controls that keep legitimate work running.

Threat-Model an AI System · Red-Team Safely and Verify Mitigations

16. Fairness, Explainability and Contestability

I can compare group error rates on a real model, explain one decision and design a route to challenge it.

Measure and Debate Fairness · Explain, Challenge and Seek Recourse

17. Privacy, Consent and Data Rights

I can design privacy into a data project, judge re-identification risk and complete a privacy impact assessment.

Privacy by Design and Meaningful Consent · Complete a Privacy Impact Assessment

18. AI Product Strategy, Deployment and Monitoring

I can build a product case against a baseline and plan deployment, monitoring and rollback.

Build the Product Case and Architecture · Deploy, Observe and Improve Responsibly

19. Governance, Audits and Incident Response

I can design governance with real decision rights, audit a system against its commitment and report an incident honestly.

Design Governance and Audit Evidence · Respond to Incidents and Learn Without Hiding Failure

20. Responsible AI Capstone and Review Board

I can bring evidence from every gate into a release dossier and defend, revise and publish it.

Build, Test and Document the Capstone · Defend, Revise and Publish Responsibly