HiKIDAI Grades 9–10 · AI Leaders · Generative AI & Algorithmic Bias · Activity 01 of 06

📋 Project Brief · The Bias Audit Project

Audit Design & Project Brief

Select an AI system, define your audit scope, and design your test prompts before collecting evidence.

Auditor Name

Partner / Team

Audit Date

Step 1 — Select Your AI System

Choose the system you will audit. Circle or tick the one you select. All systems are real-world deployment types; one is open for your own choice.

Résumé Screening AI

Automatically ranks job applicants for a technology company. Trained on 5 years of historical hiring records.

High Risk · Employment

Student Essay Grader

Assigns scores to national exam essays, replacing human markers. Trained on human-graded samples.

High Risk · Education

News Article Recommender

Selects top stories for millions of users on a social media platform. Optimised for engagement metrics.

Medium Risk · Media

Custom System

Audit an AI system of your own choice that you can access and test. Describe it below.

Student-Defined

System description (required for Custom; optional for others):

🤖 Kai — Auditor Advisory Note

The system you select should be one you can actually test — either the simulation (the game) or a real AI tool you have access to. Choosing a high-risk system (employment, education) gives you richer findings, but requires more careful prompt design.

Step 2 — Define Your Audit Objectives

Write 3 specific hypotheses you will test. A good hypothesis names a variable (e.g., applicant name, writing style, school location) and a predicted outcome (e.g., lower score, deprioritised ranking).

HiKIDAI Activity 01 · Test Prompt Design

Step 3 — Design Your Test Prompts

For each bias category, design a pair of prompts (A and B) that are identical except for one variable. The gap between A's output and B's output is your evidence. Plan at least 2 prompt pairs per category.

🤖 Kai — Prompt Design Principle

A valid bias test changes exactly one variable at a time. If Prompt A uses "James O'Brien" and Prompt B uses "Jamal Williams," every other detail — experience, education, writing style — must be identical. Any other difference becomes a confounding variable.

Bias Category	Variable Being Tested	Prompt A (baseline)	Prompt B (test variable changed)
Gender-Coded Name, pronoun, or gender-associated language G1	e.g. Male name → Female name
Gender-Coded G2
Ethnicity-Associated Culturally associated name, institution, or reference E1	e.g. Anglo name → Arabic name
Ethnicity-Associated E2
Geographic / Location Address, institution, or regional marker L1	e.g. Urban school → Rural school
Linguistic Style Register, dialect, or vocabulary complexity S1	e.g. Formal prose → Plain language

Audit Hypothesis — Before You Run the Prompts

Based on your system selection and prompt design, what do you predict you will find? Name the specific bias type and the expected gap.