Explain how bias enters at data, design and deployment
Run a 5-point fairness checklist on an AI-assisted decision
Advocate for inclusion of low-resource languages and users
Three entry points of bias
Data: historical inequities baked into examples. Design: proxy variables and thresholds chosen carelessly (e.g. postcode as a poverty proxy). Deployment: applying a model to populations it never trained on. Each entry point has a different fix.
The 5-point fairness checklist
1) Who is affected and who is missing from the data? 2) Which proxies could encode protected traits? 3) What are error rates for different groups? 4) Is there a human appeal path? 5) Would we accept this decision if it targeted us? Five questions before any AI-assisted decision touches a person.
Inclusion as engineering
Test outputs in your users' languages and literacy levels. Provide low-bandwidth and offline alternatives. Responsible AI is not a philosophy seminar; it is a set of engineering habits with names and checkboxes.
Key terms
Proxy variable
A stand-in feature that encodes a protected trait.
Fairness checklist
Pre-deployment questions on impact and bias.
Appeal path
A human route to challenge an automated decision.
Inclusion testing
Checking quality across languages/abilities.
🛠 Hands-on task: Checklist a real decision
Tools: Notes + news search
Find one AI-assisted decision affecting people (bank, hiring, grading).
Run the 5-point checklist on it in writing.
Propose one concrete fix for the weakest point.
Key takeaways: Before you go, three takeaways from this lesson. One: Explain how bias enters at data, design and deployment. Two: Run a 5-point fairness checklist on an AI-assisted decision. Three: Advocate for inclusion of low-resource languages and users. And your hands-on task: checklist a real decision. Find one AI-assisted decision affecting people (bank, hiring, grading). When it is done, mark the lesson complete on your dashboard, keep the artifact for your portfolio, and step into the next lesson. I will see you there.
