Module 2: Core Ethical Principles
AI Ethics and Safety
The Five Principles That Matter
1. Beneficence — Do Good
AI should actively benefit people. This sounds obvious, but in practice it's not. Facebook's recommendation algorithm was designed to maximise engagement. It succeeded brilliantly — and in doing so, amplified misinformation, radicalised users, and worsened mental health outcomes among teenagers. The system was doing what it was designed to do. The design goal was wrong.
The principle of beneficence asks: beneficial to whom? An AI that's good for the company's revenue but bad for users' wellbeing fails this test. Designing for genuine benefit requires defining "benefit" honestly.
2. Non-Maleficence — Don't Harm
The medical principle "first, do no harm" applies directly to AI. But harm in AI is often indirect and delayed:
- A hiring AI that subtly disadvantages candidates from certain postcodes
- A content recommendation system that gradually narrows someone's worldview
- A predictive policing tool that sends more officers to already over-policed neighbourhoods
The challenge: harm is often invisible to the people building the system. It shows up in the lives of people they never meet. Non-maleficence requires actively looking for harm, not just avoiding obvious damage.
3. Autonomy — Respect Human Agency
People should understand when AI is influencing their decisions and be able to opt out. This principle is violated constantly:
- Dark patterns that use AI to manipulate purchasing decisions
- News feeds that use AI to influence what you believe is important
- AI systems that make decisions about you without explanation
The EU AI Act requires that high-risk AI systems provide meaningful human oversight. This isn't bureaucracy — it's the principle of autonomy in legal form.
4. Justice — Distribute Benefits and Harms Fairly
AI shouldn't systematically benefit one group while harming another. But it often does:
- Facial recognition works best on white male faces and worst on dark-skinned women (Joy Buolamwini's research at MIT showed error rates of 34.7% for dark-skinned women vs. 0.8% for light-skinned men)
- AI-driven healthcare tools trained on data from wealthy countries may not work for patients in developing nations
- Generative AI benefits companies that can afford to deploy it while displacing workers who can't
Justice requires asking: who benefits, who's harmed, and is that distribution acceptable?
5. Transparency — Be Open About How AI Works
People affected by AI decisions deserve to understand how those decisions are made. This principle underpins:
- The right to explanation (GDPR Article 22)
- Requirements for AI labelling on generated content
- Open-source AI development
Transparency isn't just about code access. It's about meaningful communication: can an ordinary person understand why an AI system made a particular decision about them?
Match each ethical principle to its core question:
How These Principles Conflict
Here's what most ethics discussions leave out: these principles regularly contradict each other.
- Beneficence vs. Privacy: AI health diagnostics could save lives, but require access to private medical data
- Transparency vs. Security: Explaining how a fraud detection system works helps fraudsters evade it
- Autonomy vs. Beneficence: Should AI override a patient's treatment preferences if it predicts better outcomes?
- Justice vs. Efficiency: Fair AI systems may be less accurate than biased ones (because real-world data reflects real-world inequality)
There are no clean answers. Ethics isn't a checklist — it's a continuous process of weighing competing values. The companies and individuals who handle AI ethics well are the ones who acknowledge these tensions rather than pretending they don't exist.
Which principle conflict is illustrated by the example of AI health diagnostics that could save lives but require private medical data?
Ethical Principle Analysis
My company is building an AI system that analyses employee emails to predict burnout risk. The goal is early intervention — offering support before employees hit crisis point. Evaluate this against the five core AI ethics principles (beneficence, non-maleficence, autonomy, justice, transparency). For each principle, give me: how the system could satisfy it, how it could violate it, and what specific design choice would determine which way it goes.
Principle Conflict Resolution
I'm designing an AI-powered personalised learning platform for children aged 8-14. I've identified a conflict: - BENEFICENCE: The AI could provide much better learning by tracking detailed behavioral data (attention patterns, emotional responses, learning speed) - AUTONOMY/PRIVACY: Children can't meaningfully consent to this level of surveillance, and the data could be misused Help me think through this tension. What are the possible approaches, what do other companies do, and what would you recommend? Don't give me a vague "balance both" answer — give me a specific design recommendation with tradeoffs acknowledged.
Everyday Ethics Check
I want to evaluate three AI products I use daily against core ethical principles: 1. ChatGPT (OpenAI) — I use it for work writing 2. TikTok — AI-driven content recommendations 3. My car insurance app — uses AI telematics to set my premium For each: which ethical principles does it handle well? Which does it violate? Give me specific evidence, not generalities. Include the principle of justice — who benefits and who's disadvantaged by each system?
1. Pick one AI system you interact with regularly
2. Score it 1-5 on each of the five principles (1 = clearly violates, 5 = exemplary)
3. For any score below 3, write one specific thing that would improve it
4. Share your scorecard with someone and see if they agree — disagreements reveal interesting assumptions
The point isn't to get the "right" scores — it's to practice systematic ethical thinking about AI systems rather than vague unease.
---
- 1Five core principles: beneficence (do good), non-maleficence (don't harm), autonomy (respect agency), justice (be fair), transparency (be open)
- 2These principles regularly conflict with each other — ethics is about navigating tensions, not following a checklist
- 3Real-world examples show every principle being violated by major companies — this isn't theoretical
- 4The EU AI Act puts these principles into law for the first time — understanding them is now a legal requirement
- 5Practice applying these principles to AI systems you use daily — it builds the ethical muscle memory you'll need