Comprehensive Study Notes: Moral Principles in AI, Typology of Bias, Real-World Case Studies, Responsible AI Frameworks & Complete Solutions
Ethics are the moral principles guiding individuals to act fairly, honestly, and responsibly. Artificial Intelligence is not solely a computational tool; it is a profound socio-technical system directly impacting human lives in critical sectors such as healthcare, recruitment, banking, education, and public safety.
Because AI algorithms learn directly from historical patterns, they risk replicating and amplifying historical societal prejudices based on gender, caste, creed, ethnicity, race, or disability. Ethical governance enforces neutrality.
Society cannot trust automated tools that function as opaque "black boxes". When AI architectures are open, explainable, and truthful about their reasoning, users gain confidence in relying on them.
An algorithm cannot bear legal or moral liability. If an AI system denies a loan or misdiagnoses a patient, the human developers, corporate executives, and deploying institutions remain accountable.
AI consumes vast amounts of personal information. Ethical compliance ensures user consent is obtained, sensitive identities remain confidential, and data is protected from unauthorized harvesting.
| Law / Regulatory Body | Jurisdiction | Core Mandates & Protections |
|---|---|---|
| Digital Personal Data Protection (DPDP) Act, 2023 | India | Mandates explicit prior consent before processing personal data; grants individuals rights to access, correct, and erase their data; imposes penalties for unauthorized misuse. |
| Information Technology (IT) Act, 2000 | India | Sets the primary legal framework governing cyber law, electronic contracts, data theft, computer fraud, and cybersecurity obligations. |
| Reserve Bank of India (RBI) Lending Directives | India (Banking) | Establishes governance for automated and algorithmic digital lending platforms to guarantee transparency, auditable records, and fair financial practices. |
| GDPR and CCPA | European Union & California | International benchmarks securing consumer data privacy, algorithmic accountability, and strong institutional penalties for data breaches. |
Bias refers to an unfair preference for or against an individual, category, or demographic group without balanced, objective evaluation.
| Type of Bias | Mechanism & Cause | Real-World Example |
|---|---|---|
| Data Bias | Training data is unrepresentative, skewed, or omits certain segments of the population. | A facial recognition model trained almost entirely on lighter-skinned individuals yields high error rates when analyzing darker-skinned faces. |
| Historical Bias | Data mirrors past systemic discrimination and socio-cultural inequalities present when records were created. | A recruitment tool trained on decade-old corporate hiring histories replicates and reinforces previous male-dominated hiring trends. |
| Measurement Bias | Flawed, proxy, or incomplete parameters are selected to quantify an attribute. | Evaluating employee performance strictly by "number of hours logged in" rather than the quality, output, or accuracy of the completed tasks. |
| Algorithmic Bias | Mathematical formulations, optimization weights, or architectural assumptions inherently favor specific groups. | An algorithm heavily weighting zip codes or neighborhood locations, inadvertently penalizing low-income loan applicants despite balanced underlying inputs. |
| Human Bias | Personal preconceptions, cultural blindspots, or design choices of human developers influence data selection. | Engineers curating test benchmarks that reflect only their personal lifestyle choices, environments, or regional cultural norms. |
A global technology enterprise deployed an internal AI tool to automate resume screening by training on resumes collected over the previous decade. Because the tech industry's historical applicant pool was overwhelmingly male, the algorithm learned to associate male-correlated terms with high performance and actively downgraded resumes mentioning women's colleges or female athletic teams. The company abandoned the system once the flaw was discovered.
Independent research evaluating commercial computer vision packages revealed that facial verification error rates were less than 1% for lighter-skinned men, but surged to 30% to 35% for darker-skinned women. The disparity stemmed directly from unrepresentative training datasets that lacked sufficient images of diverse demographics.
A clinical risk management algorithm used historical healthcare expenditure as a proxy to forecast a patient's future medical need. However, financial spending reflects economic access and insurance coverage rather than true illness severity. Impoverished populations who spent less due to lack of funds were falsely classified as healthier, denying them specialized medical care.
A digital citizen is someone who navigates digital ecosystems, connected networks, and communications safely, critically, and ethically.
Classmates created an unauthorized social media impersonation account targeting a student named Ishaan, publishing humiliating falsehoods under his name. The psychological distress caused him to stop attending school entirely. This demonstrates that digital actions carry severe real-world consequences, underlining that digital rights carry strict ethical duties to protect others' dignity.
1. Which of the following is an example of structured data?
Structured data adheres to a rigid schema organized neatly into rows and columns, such as spreadsheet databases. Videos, paragraphs, and photo libraries lack this structured tabular arrangement.
2. AI bias mainly comes from:
Machine bias results from unrepresentative, skewed datasets that produce disparate real-world outcomes, such as elevated error rates for darker skin tones.
3. Human-in-the-Loop (HITL) means:
Human-in-the-Loop architectures ensure that automated algorithms serve as decision-support systems, requiring human professionals to review and approve high-stakes actions.
4. Which principle protects personal data?
The privacy principle requires that personal information is collected solely with user consent, guarded against unauthorized exposure, and handled responsibly.
5. Which is a responsibility of a digital citizen?
Responsible digital citizenship requires fact-checking and verifying the authenticity of messages before circulating them online.
1. What is AI ethics?
Answer: AI ethics is the field of study focused on designing, deploying, and governing artificial intelligence responsibly. It establishes guidelines to ensure algorithms operate fairly, transparently, and safely without causing harm or perpetuating discrimination.
2. Give one example of historical bias.
Answer: An AI resume-screening algorithm trained on a company's past ten years of hiring records penalized female applicants because the engineering department's historical hiring decisions were predominantly male.
3. Why is transparency important in AI?
Answer: Transparency ensures that AI decision-making processes are explainable and auditable. Being open about how conclusions are reached allows humans to catch errors, detect unfair bias, and build public trust.
4. Name two rights of digital users.
Answer: (1) Right to Privacy: Protecting personal information, passwords, and images from unauthorized exposure. (2) Right to Safety: Protection from online cyberbullying, harassment, and digital fraud.
5. What is data bias?
Answer: Data bias occurs when the dataset used to train an AI model is incomplete, unrepresentative, or systematically leaves out certain groups of people or scenarios. This causes the system to perform inaccurately for those underrepresented demographics.
1. A hiring system prefers male candidates based on past data which included mostly men. What type of bias is this?
Answer: Historical Bias. The model ingested past societal inequalities preserved in historical employment records and replicated those discriminatory trends into future recommendations.
2. A health AI diagnoses illness based only on money spent on treatment. What type of bias is this?
Answer: Measurement Bias. The developers selected an inaccurate proxy metric (expenditure) to represent health status, ignoring the fact that lower spending often reflects financial hardship rather than physical health.
3. A loan application is rejected by AI. Why should a human review the decision?
Answer: A human review (Human-in-the-Loop) is required to ensure accountability, verify that proxy factors (like residential area) did not unfairly penalize the applicant, and provide clear explainability for the final decision.
4. A student shares false news online. Which responsibility is not followed?
Answer: The responsibility to "Think Before You Click or Share" (verifying information authenticity) and "Respect Others" by not spreading misinformation.
| Description | One Word Answer |
|---|---|
| Repeating past unfair patterns | Historical Bias (or Bias) |
| Missing groups in training data | Data Bias |
| Final decision taken by humans | Human-in-the-Loop (HITL) |
| Moral rules that guide decisions | Ethics |
| Responsible use of technology | Digital Citizenship |