Skip to main content

Ethical frameworks in AI and ML

Instruction and application
Complete

Every day, huge amounts of data are processed using AI/ML techniques to create insights for organisations. The challenge is making sure this work is done responsibly. In this lesson, you will explore the foundational frameworks and principles that guide ethical decision-making in AI projects.

What do we mean by ethics?​

In data and AI, ethics is about evaluating how systems and inputs are designed, then making value-based judgements in a personal and professional context.

Definitions from leading organisations​

  • Alan Turing Institute (2025): "Data ethics is a branch of ethics that studies and evaluates moral problems related to data, algorithms (including AI/ML) and corresponding data practices."
  • Open Data Institute (2019): "Data ethics evaluates data practices that can adversely impact people and society in collection, sharing and use."
  • UK Government (2024): "Data ethics refers to principles and moral obligations that guide how an organisation collects, shares and uses data, especially personal data, to ensure fairness and non-discrimination."

Why ethical frameworks matter​

Ethics is not just about compliance. It is about ensuring AI systems positively contribute to society while minimising harm. AI can influence critical outcomes such as hiring decisions, medical prioritisation and risk assessment.

Key point​

Without ethical oversight, AI systems can perpetuate or amplify existing bias, producing unfair or discriminatory outcomes.

For example, Amazon's AI-powered hiring tool was found to discriminate against female candidates because it learned from historical hiring data that reflected prior bias.

Like all models, a tool is only as strong as its training data. If the underlying data is biased, model outputs will be biased too. Switching to a different model alone will not fix this.

Without proactive safeguards, AI can reinforce systemic discrimination, violate privacy and reduce public trust. Ethical frameworks help teams evaluate risks, increase transparency and ensure decisions are fair and accountable.

Key ethical principles in AI​

According to the European Commission, responsible AI design and governance relies on four foundational principles:

⚖️

Fairness

AI should not disproportionately harm or benefit groups based on protected characteristics.

Requirements: Bias detection in training datasets, algorithmic audits, and representative data.

🛡️

Accountability

Developers and organisations must take responsibility for AI decision outcomes.

Requirements: Clear ownership for AI outcomes and human appeal mechanisms.

🔍

Transparency

AI systems must be inspectable and understandable to users, stakeholders, and regulators.

Requirements: Clear system documentation and user-friendly decision explanations.

💡

Explainability

AI decisions must be interpretable and justified by humans using XAI techniques.

Requirements: Prefer simpler models where practical and provide visual reasoning tools.

Action item: Quiz​

Let's check your understanding of these ethical principles.

An organisation is developing an AI tool to help shortlist candidates for job interviews based on their CVs. Applying the principles discussed in the text, what is the MOST important initial step it should take to ensure ethical use of this AI?

A company uses an AI-powered content moderation system on its social media platform. Several users have reported being unfairly flagged and having their posts removed. Based on the ethical principles discussed, what would be the MOST appropriate action for the company to take?

A financial institution is deploying an AI system to assess loan applications. To adhere to the principle of explainability, what should it prioritise providing to customers who are denied a loan by the AI?

In progress