Courses / Computer Science XI (HSSC-I) – FBISE Complete SLO-Based Course

5.2 Artificial Intelligence: Stakeholders, Values and Fair Design

Unit 5 – Application of Computer Science · 60 min read · 8 MCQs to check yourself

Crash Course AI #18 – algorithmic bias and fairness

1. What is an AI system?

Artificial intelligence (AI) is the ability of a computer system to perform tasks that normally need human intelligence – recognising speech and images, translating languages, making recommendations and decisions. Most modern AI uses machine learning: the system learns patterns from training data.

Positive impacts of AI (model paper):

  1. Healthcare – early detection of diseases from X-rays and scans, drug discovery.
  2. Education – personalised learning, instant feedback, AI tutors and translation.
  3. Agriculture – crop-disease detection from phone photos, yield prediction, smart irrigation.
  4. Accessibility – speech-to-text, image descriptions for visually impaired people.
  5. Safety and efficiency – fraud detection in banking, traffic management, automation of repetitive work.

2. Stakeholders of an AI system

A stakeholder is anyone who affects, or is affected by, an AI system.

Stakeholder Main interest
Developers / engineers Accuracy, performance, meeting deadlines, technical quality
Companies / owners / investors Profit, market share, low cost, competitive advantage
Users / customers Useful, easy, accurate, affordable service; privacy
People affected by decisions (applicants, patients, citizens) Fairness, no discrimination, explanation of decisions, right to appeal
Government / regulators Safety, law, privacy, national interest, accountability
Society / communities Jobs, equality, cultural values, environmental impact
Data subjects (people whose data trains the model) Consent, privacy, control of personal data

3. Conflicting interests – examples

AI system Conflict
Social-media recommendation feed Company wants maximum watch time (ads) ↔ users' wellbeing and time; society wants less misinformation
AI loan approval Bank wants low risk and profit ↔ applicants want fair treatment; a model trained on past data may discriminate against women or rural applicants
Face recognition in public places Government wants security ↔ citizens want privacy and freedom; errors are higher for some groups
Self-driving car Passengers' safety ↔ pedestrians' safety; manufacturer's cost ↔ strict testing
AI hiring tool Company wants speed ↔ candidates want fair, explainable decisions

How culture and values affect design: what counts as "fair", "private" or "appropriate" differs between cultures and communities. Language support (Urdu, Punjabi, Pashto, Sindhi), dress, names and accents in training data, religious and social norms in content moderation – designers' choices reflect whose values were considered. A voice assistant trained mostly on American English understands Pakistani accents poorly; a moderation model trained on Western data may misjudge local content.

4. Bias in AI

Algorithmic bias happens when an AI system gives systematically unfair results for certain groups. Causes: unrepresentative or historical training data, biased labels, poorly chosen features (e.g. postcode as a proxy for income), and lack of diversity in the design team.

5. Policies and practices that protect stakeholders

Bias, fairness and responsible AI explained

Short and long questions

1. Write down any three positive impacts of AI systems.

Model answer

(1) Healthcare: detects diseases early from X-rays/scans and supports doctors. (2) Education: personalised learning, instant feedback and translation. (3) Agriculture/industry: crop-disease detection, yield prediction, automation and fraud detection improve efficiency and safety. (Also accessibility tools such as speech-to-text.)

2. Who are the stakeholders of an AI system? State the interest of any three.

Model answer

Stakeholders are all people/groups who affect or are affected by an AI system: developers, companies/investors, users, people affected by decisions, governments/regulators, society, data subjects. Developers – accuracy and performance; companies – profit and market share; affected people – fairness, explanation and right to appeal; regulators – safety, privacy and law.

3. What is algorithmic bias? Give two causes.

Model answer

Algorithmic bias is when an AI system systematically produces unfair results for certain groups of people. Causes: (1) training data that is unrepresentative or reflects past discrimination; (2) biased labels or proxy features (e.g. postcode standing for income), and lack of diversity in the design team.

4. Suggest four policies that can protect the interests of different stakeholders in AI systems.

Model answer

(1) Transparency and explainability – disclose AI use and reasons for decisions. (2) Fairness testing with diverse data before deployment. (3) Data protection – consent, minimum data, anonymisation, privacy laws. (4) Human oversight and an appeal process for important decisions; clear accountability and regulation (e.g. risk-based AI laws).

5. Evaluate how different stakeholders' culture, values and conflicting interests affect the design of an AI system. Use the example of an AI-based loan approval system for a bank in Pakistan.

Model answer

Stakeholders and interests:

  • Bank owners/investors – profit, low default risk, quick automated decisions.
  • Developers – a highly accurate model delivered on time.
  • Applicants – fair, fast decisions with reasons; ability to appeal.
  • Regulators (State Bank) – financial stability, consumer protection, no discrimination.
  • Society – financial inclusion of women, rural people, small businesses. Conflicts: maximum profit may lead the model to reject anyone without a long credit history – excluding rural applicants, women and young people; training on past decisions copies past bias; collecting more personal data (social media, phone usage) improves prediction but harms privacy. Culture and values: many Pakistanis use informal income and family businesses; a model designed around salaried-employee data from other countries would misjudge them; Islamic banking customers need Shariah-compliant products; Urdu/regional language support is needed for explanations. Balanced design / policies: use representative local training data; test fairness across gender and region; allow alternative data with consent; explain decisions in simple Urdu/English; human review of rejections; follow privacy rules and State Bank regulations; regular audits. This balances the bank's profit with fairness and public trust.

This lesson ends with a quiz of 8 MCQs. Open the lesson on PrepOasis to take it and save your progress.

Previous: 5.1 Foundations of IoT, Cloud Computing and Blockchain · Next: Unit 5 Review, SLO Checklist and Unit Test

All lessons in Computer Science XI (HSSC-I) – FBISE Complete SLO-Based Course