Formative practice pilotUngraded — progress saved in this browser

125.788Big Data Decision JourneyBack to journey map

Learning Module 05 · Week 5 · Individual mission

Challenge Bias in a Financial Decision

Find where a seemingly neutral model can reproduce harm and design a reviewable safeguard.

About 8 minutes. There is no timer.Formative practice only. No assessment marks.1 Insight Brief on completionLO1LO2LO3
Mission statusNot started
  1. 1BriefingCurrent
  2. 2Worked exampleNext
  3. 3Case boardNext
  4. 4CheckpointsNext
  5. 5ReflectionNext
  6. 6CompleteNext

Start here

Briefing

Removing a protected field does not automatically remove its influence. Location, device, language, and historical decisions may act as proxies.

Trace the potential harm, compare group-level evidence, and retain a meaningful human review and correction path. Check that the human-review path meets the standard regulators expect for automated credit decisions.

Practice boundary

Your answers to each question are kept only while this page is open and reset when you refresh. Your completed-mission progress and Decision XP are saved in this browser on this device only, never sent to a server. Clearing your browser data, or using Clear my progress, removes them.

Game economy

Build Evidence Momentum without risking Decision XP

+3Correct checkpoint
−1First incorrect check in a checkpoint
100 Decision XPAlways recoverable through retry

You begin with 3 Evidence Momentum. Any held points return when you solve that checkpoint. Evidence Momentum never falls below zero and never changes assessment marks. Complete the mission to open one transparent reward draw.

What you will practise

  1. Identify a proxy, representation, or measurement pathway that can create unequal outcomes.LO3
  2. Evaluate a model using performance, impact, contestability, and cultural context.LO1, LO2

6 terms for this mission

Proxy
A variable that indirectly represents another characteristic.
Representation bias
Error caused when relevant groups or conditions are poorly represented in data.
Contestability
A person's practical ability to question and correct a decision.
Cultural context
Local meanings, practices, and power relationships that affect how data and decisions are interpreted.
Statistical discrimination
Discrimination by group averages, not personal prejudice (Phelps, unlike Becker's taste-based bias); big data revives it when proxies rebuild a protected group whose field was dropped.
Human review and oversight
GDPR Article 22 gives a person a qualified right not to be subject to a solely automated decision with legal or similarly significant effects; separately, the EU AI Act establishes human-oversight duties for high-risk systems such as AI credit scoring of individuals; Vietnam's Law on Personal Data Protection (Law 91/2025/QH15, in force 1 January 2026) provides a general personal-data-protection framework (not an equivalent automated-decision right).

Every organisation, person, figure, and dataset in this mission's scenario is invented. Real institutions, laws, and standards are named only as general context.

Sources and further reading

Every organisation, person, figure, and dataset in this mission's scenario is invented. Real institutions, laws, and standards named in this mission, including in the sources below, appear only as general context, and the sources support the concepts and methods, not the events in the scenario.

  1. Barocas, S., Hardt, M., & Narayanan, A. (2023). Fairness and Machine Learning: Limitations and Opportunities. MIT Press. ISBN 978-0-262-04861-3.

    A standard reference on how proxy variables, unrepresentative data and measurement choices produce unequal model outcomes, and on why models are evaluated group by group rather than on aggregate accuracy.

  2. Regulation (EU) 2016/679 (General Data Protection Regulation), Article 22. Official Journal of the European Union, L 119, 4.5.2016, pp. 1-88

    Article 22 sets out a qualified right not to be subject to a decision based solely on automated processing that produces legal or similarly significant effects, with safeguards of human intervention and contestability.

  3. Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 14 and Annex III, point 5(b). Official Journal of the European Union, 12.7.2024

    Annex III point 5(b) classifies AI used to evaluate creditworthiness of natural persons as high-risk, and Article 14 states the human-oversight duties that then apply.

  4. Fuster, A., Goldsmith-Pinkham, P., Ramadorai, T., & Walther, A. (2022). Predictably unequal? The effects of machine learning on credit markets. The Journal of Finance, 77(1), 5-47. doi:10.1111/jofi.13090

    Peer-reviewed evidence that more flexible machine-learning credit models can shift predicted risk unevenly across groups even when protected characteristics are not used as inputs.