📋 Activity Overview

Students study a real case of facial recognition misidentification in an Indian city, identify what training data caused the bias, and debate the ethics of deploying such systems in public spaces.

💡 Teacher Tip

The most powerful moment is when students realise that the algorithm is not 'wrong' in a technical sense — it performed exactly as trained. The problem is in what we chose to train it on and how we deployed it. This distinction between technical correctness and ethical soundness is central to responsible AI.

🎯 Learning Objectives

  • ✓ Understand how facial recognition algorithms learn from training data
  • ✓ Identify sources of bias in machine learning systems
  • ✓ Analyse the social impact of algorithmic errors at scale
  • ✓ Evaluate ethical trade-offs in deploying AI in public safety contexts

🗂️ Materials Needed

Case study handouts Bias analysis worksheet Debate role cards (Government, Civil Liberties, Police, Affected Community, Tech Company) Timer Whiteboard for notes Voting cards

📌 Step-by-Step Instructions

Case Study (8 min) — Read the case: A facial recognition system deployed at a railway station in India incorrectly flagged 12 passengers as persons of interest in one week. 9 of the 12 were from a single demographic group. None were actually suspects.
Technical Analysis (10 min) — Groups analyse: What training data could cause this bias? (Under-representation of certain skin tones, lighting conditions in training photos, image quality differences.) How does a false positive harm the flagged person?
Stakeholder Assignment (3 min) — Each group receives a role: Government (security focus), Civil Liberties Union (rights focus), Police (operational focus), Affected Community (justice focus), Tech Company (commercial focus).
Prepare Position (8 min) — Groups prepare a 2-minute position statement from their stakeholder's perspective. What do they want? What do they fear?
Structured Debate (12 min) — Round-robin: each group presents, others may ask one question. No interruptions during presentations.
Voting and Reflection (5 min) — Class votes: Should this system continue operating, be suspended pending audit, or be banned? Show hands, then discuss: did any argument change your mind?
CT Connection (4 min) — 'Every AI system is shaped by the data it learns from — and data reflects the world's existing inequalities. Understanding this is CT applied to social impact.'
Takeaways (5 min) — Each student writes: one technical cause of the bias, one ethical concern, and one policy recommendation.

🧠 CT Pillar Connections

Abstraction
Facial recognition abstracts a human face to a mathematical vector — but this abstraction loses information (emotional context, identity, dignity) that matters enormously for ethical decisions.
Algorithmic Thinking
Students analyse the algorithm's decision pipeline: training data → model → threshold → flag. Each stage introduces potential bias — understanding the pipeline enables targeted fixes.

💬 Discussion Questions

  • Can an algorithm be racist even if it was never programmed with racist intent?
  • Who should be held responsible when an AI makes a wrong accusation — the developer, the operator, or no one?
  • If a system is 95% accurate but the 5% errors fall disproportionately on one group, is that acceptable?
  • What regulations would you put in place before allowing facial recognition in public spaces?