MODULE 16 — AI SAFETY & ETHICS

7

16.1 Introduction to AI Safety & Ethics

AI is increasingly used in areas such as:

• Healthcare
• Education
• Banking
• Transportation
• Business
• Government

Because AI can influence important decisions, it must be designed and used in a way that is safe, fair, transparent, responsible, and accountable.

AI Safety

AI Safety focuses on preventing harmful behaviour from AI systems and making them secure and dependable.

AI Ethics

AI Ethics refers to the moral principles and guidelines used to develop and use AI responsibly.

Core Goal

AI Safety + AI Ethics = Trustworthy AI

16.2 Objectives of AI Safety

7

AI systems should be designed to:

Prevent Harm

Protect users and society from harmful AI behaviour.

Avoid Dangerous Actions

Prevent AI from taking unsafe or inappropriate actions.

Reduce Incorrect Outputs

Minimise false or harmful responses.

Protect Data

Keep sensitive information secure.

Maintain Human Control

Ensure humans can supervise and override important AI decisions.

16.3 Key Principles of AI Ethics

1. Fairness

AI should treat users equally and avoid discrimination.

Example:
A hiring AI should evaluate candidates using consistent criteria rather than unfairly favouring a demographic group.

2. Transparency

Users should be able to understand how AI decisions are made.

Explainable AI (XAI) can help improve trust and accountability.

3. Accountability

Developers and organisations must take responsibility for AI decisions and errors.

16.4 Privacy, Safety & Human Control

7

Privacy

AI systems should protect personal information. Data collection should be minimal, purposeful, and based on appropriate user consent.

Safety & Reliability

AI should be tested regularly across different situations, including edge cases.

Human Control

Humans should retain oversight and the ability to override critical AI decisions, particularly in areas such as:

• Healthcare
• Law
• National security

16.5 AI Bias

What is AI Bias?

AI bias occurs when an AI system produces unfair or prejudiced results.

Bias can come from:

• Poor training data
• Incomplete data
• Human prejudice in datasets
• Lack of diversity
• Imbalanced class representation

Examples

• Facial-recognition systems performing differently across skin tones
• Hiring systems favouring certain demographics
• Biased recommendation systems

Impact

Unchecked bias can lead to:

• Unfair hiring or lending
• Discrimination at scale
• Loss of public trust
• Legal and regulatory consequences

16.6 How to Reduce AI Bias

5

Use Diverse Datasets

Include varied demographics, cultures and geographic regions.

Regular Testing & Audits

Evaluate AI outputs across different user groups.

Monitor Outputs

Look for unusual patterns or unequal outcomes after deployment.

Ethical Reviews

Use diverse teams and ethical review processes throughout development.

16.7 AI Hallucinations

An AI hallucination occurs when an AI system generates false information, fabricated facts, or incorrect answers with apparent confidence.

Examples

An AI might:

• Invent a research paper
• Give an incorrect historical date
• Provide incorrect medical information

Causes

• Outdated information
• Vague prompts
• Insufficient context
• Over-reliance on pattern matching

Reduction Strategies

• RAG
• Context injection
• Fact-checking
• Reliable databases

RAG can retrieve information from a knowledge base before generating an answer.

16.8 Privacy & Data Protection

7

AI systems may process sensitive information such as:

• Personal details
• Financial information
• Medical records
• Conversations

Major Risks

Data Leaks — Unauthorised exposure of information.

Unauthorised Access — Attackers gaining access to AI systems.

Identity Theft — Stolen information used to impersonate users.

Surveillance Misuse — AI being used for inappropriate monitoring.

Protection Methods

• Data encryption
• Secure storage
• Access controls
• User consent systems

16.9 AI Security Risks

Important security risks include:

Hacking Attacks

Attackers compromise AI infrastructure.

Data Poisoning

Malicious data is inserted into training datasets.

Prompt Injection

Specially crafted inputs attempt to manipulate AI behaviour or bypass safety controls.

Malware Threats

AI systems or outputs may be misused to spread malware.

Security Solutions

• Strong authentication
• Secure AI architecture
• Monitoring
• Regular security updates

16.10 Responsible AI & Governance

6

Responsible AI

Responsible AI should be:

Ethical + Fair + Safe + Transparent + Human-Centred

AI Governance

AI governance consists of rules and policies controlling how AI is developed and used.

It can include:

• Ethical standards
• Government regulations
• Organisational policies
• Compliance systems

16.11 Real-World Ethical Challenges

These examples show why AI systems need appropriate human oversight, testing, governance and ethical safeguards.

16.12 Future of AI Ethics

Future developments include:

• Stronger AI regulations
• Explainable AI
• Global AI policies
• Safer AI systems
• Greater human-AI collaboration

Module 16 Activity

Ethical AI Case Study

Choose an AI application such as hiring, healthcare or education.

Identify:

1. Possible benefits
2. Possible bias
3. Privacy risks
4. Safety risks
5. Required human oversight
6. Appropriate ethical safeguards