Ethics in AI — Beginner-Friendly Notes
1. Why Ethics Is Core (not optional)
AI decides on people's lives (health, loans, jobs, policing). Without ethics, even technical success can cause harm, unfairness, loss of trust. Responsible AI = human-centred.
2. Four Core Principles (exam framework — like medical ethics)
- Fairness / Non-discrimination: AI should not systematically disadvantage groups (links to bias next topic).
- Transparency / Explainability: Decisions should be understandable (why loan denied? — right to explanation, links to expert systems).
- Privacy / Confidentiality: Respect data consent, minimise collection, protect against leakage.
- Accountability / Responsibility: Humans remain liable — who is responsible when AI fails? Need audit trails.
Add Beneficence (do good) and Non-maleficence (do no harm) and Autonomy (preserve human agency).
3. Ethical Tensions (why hard)
- Privacy vs Safety: more data → better safety (health) but less privacy.
- Explainability vs Accuracy: deep nets accurate but opaque; simpler models explainable but weaker.
- Automation vs Employment: efficiency vs livelihood.
- Global vs Local: one model may not respect local values/languages.
4. What Ethical AI Requires in Practice (process)
- Ethics by design: consider harms at problem formulation (PEAS includes performance and harms).
- Stakeholder inclusion: involve affected communities, not just engineers.
- Audits & impact assessment: bias testing, red-teaming, model cards.
- Regulation: e.g., EU AI Act risk tiers, India's approach — voluntary guidelines + sector rules.
5. One-Line Summary
AI ethics applies principles of fairness, transparency, privacy and accountability to ensure AI does good, avoids harm and keeps humans responsible throughout design and deployment.
6. Quick Self-Check
- Name the 4 principles.
- Why can "more accurate but less explainable" be an ethical issue?