Skip to main content

Ethical frameworks for AI development

Instruction and application
Complete

How do we integrate ethics into AI decision-making? A practical way is to use established ethical frameworks that provide structured approaches for evaluating risks and benefits.

Integrating ethics​

No single framework is universally perfect. The best approach is often selecting and combining frameworks based on context, impact and stakeholder needs.

Common ethical frameworks​

Deontology (rule-based ethics)​

This framework follows strict ethical rules, regardless of consequences.

Principle

  • AI should always follow clear, predefined ethical rules to ensure consistency in decision-making.

What it might look like

  • Always allowing users to opt out of data collection in a recommendation system, even if this reduces model effectiveness.

Utilitarianism (consequentialism)​

This approach evaluates AI based on balancing benefits and harms. For additional context, see the MIT Moral Machine example on autonomous driving trade-offs.

Principle

  • Decisions should maximise overall benefit while minimising overall harm.

What it might look like

  • A healthcare system prioritising diagnostic accuracy to benefit the largest number of patients, even if some predictions are less interpretable.

Human-centered AI​

This framework prioritises human well-being, agency and oversight in AI development.

Principle

  • AI should include safeguards that keep humans in control of high-impact decisions.

What it might look like

  • A self-driving capable vehicle that still allows a human operator to override AI decisions in critical moments.

Each framework provides guidance for resolving ethical dilemmas in AI. Used thoughtfully, these approaches help organisations design systems that are trustworthy, accountable and aligned with human values.

Action item: Pause and think​

Take a few minutes to consider the case study below and reflect on which approach best balances business and ethical priorities.

Case study: FirstDerm skin cancer detection​

FirstDerm is a US-based online dermatology firm that built an algorithm to detect skin cancer.

The dataset used to train the model included only around 10% ethnic minority individuals. Although those groups may have lower average incidence rates, they can face higher risk of severe or late-stage outcomes when disease is missed.

Consider what route the designers should take, and capture your thinking below.

Reflection
Which approach is most appropriate when balancing business needs with this ethical dilemma?
Your reflection here...
Considering business needs and ethical risk, what informed your decision?
Your reflection here...