AI Glossary / Governance, ethics and safety / Intermediate

Ethics by design

Definition

Ethics by design is an approach that incorporates ethical considerations into the purpose, requirements, development, deployment, and ongoing management of a system. For AI, it means translating concerns about people and consequences into concrete design decisions and evidence. It is a continuing practice, not a special algorithm or a certificate that a system is automatically ethical.

What it means

A project makes value choices long before its model is trained. It chooses which problem to solve, whose needs count, what data to collect, which errors to tolerate, and who can challenge an outcome. Ethics by design makes those choices visible and reviewable. The approach asks whether the proposed system should be built in its intended form, as well as how to build it responsibly. A list of broad principles is only a starting point; the work lies in connecting them to requirements, responsibilities, tests, and decisions throughout the system's life.

How it works

Teams identify affected people and relevant values, investigate possible benefits and harms, and translate findings into design requirements. They compare alternatives, document tradeoffs, and define ways to test whether requirements are met. For example, a concern about meaningful user choice can lead to a clearly explained optional feature and a usable alternative route. A concern about accountability can lead to records, review authority, and a correction process. Monitoring after deployment checks whether assumptions still hold. Standards and frameworks can structure the work, but they do not remove disagreement about values or decide every difficult tradeoff automatically.

Three practical examples

• A learning assistant collects only the information needed for its stated purpose and tests whether learners can understand and control optional data use.
• An AI-supported service decision includes a practical route to request review, with responsibility assigned for considering the request and correcting mistakes.
• A team evaluates accessibility with intended users before making an AI interface the only way to obtain a service. Each example connects a value concern to observable product behavior.

Think of it this way

Think of including accessibility in a building's plans while entrances and circulation are still being designed. It is easier to consider how people will use the building when those choices are open than after every wall is fixed. The analogy is limited because AI systems and social expectations can change continuously. Ethics by design therefore also requires maintenance, feedback, and reconsideration after launch.

Worked example

Suppose a fictional team is considering three designs for an AI help service. Design A requires a full account profile before answering any question. Design B requires only an account identifier and the question. Design C allows general questions without an account but requires authentication for personal records. The team identifies two distinct purposes: general information and account-specific help. Rather than assigning an unsupported numerical ethics score, it maps each data request to those purposes. C avoids account collection for general questions while retaining access checks for personal information, but introduces a routing requirement to prevent accidental disclosure. The resulting test checks that general queries remain usable and that personal-record requests require appropriate authentication. This example illustrates a traceable design choice, not proof that C is always the best or sufficient solution.

Illustrative case study

Fictional case: Beech Learning plans an assistant that recommends training opportunities. Staff representatives raise concerns that managers may treat recommendations as performance judgments. The team clarifies the intended purpose, changes how outputs are described, restricts inappropriate reuse within the planned workflow, and defines a way to challenge inaccurate information. It evaluates the revised design with users and records unresolved concerns. The case demonstrates ethical questions shaping design decisions rather than a claim that consultation guarantees acceptance or eliminates all potential harm.

Common misconceptions and mistakes

Ethics by design is not a final checklist applied after technical development. It also is not the same as legal compliance, although legal requirements are part of the context. A model with an explanation feature is not automatically accountable, and a privacy notice is not automatically meaningful choice. Values can conflict, so a design process should explain tradeoffs rather than imply that all principles have been maximized at once. Another mistake is consulting only purchasers while overlooking people affected by the system's outputs. Participation should inform decisions, not merely decorate the project record.

Business use

Begin with the actual purpose and affected population. Assign owners for ethical requirements and connect those requirements to product decisions, evaluation evidence, and release criteria. Include relevant domain expertise and meaningful input from affected people. Review alternatives, including simpler processes where they may meet the need with fewer burdens. Keep records of unresolved concerns and who has authority to accept, mitigate, or stop a proposed use. After launch, investigate complaints and changing conditions, and verify whether promised controls work in practice. The value comes from decisions and follow-through, not the label attached to the project.

People also ask

Does ethics by design require a single universal checklist?

No. Shared principles can guide the process, but concrete requirements depend on the system, context, and affected people.

Can it guarantee that no harm occurs?

No. It provides a structured way to identify and address concerns, while uncertainty and changing conditions remain.

Is it only for developers?

No. Product owners, managers, domain specialists, users, and other affected stakeholders shape purposes, constraints, and consequences that technical teams cannot decide alone.

Check your understanding

A project states that users should be able to challenge AI-supported decisions, but nobody owns the review process and no route exists in the product. Has the principle been implemented?

Answer: No. It needs concrete procedures, authority, user access, and evidence that review works in practice.

Related terms

• Responsible AI
• Value sensitive design
• Impact assessment
• Accountability
• Privacy-by-design
• stakeholder participation
• accessibility
• human oversight
• governance

Learn next

Choose one ethical principle in an AI project and translate it into a specific user-facing requirement. Name the responsible owner, a test that provides evidence, and a route for handling failures. Record any tradeoff that the requirement does not resolve.

More in Governance, ethics and safety

• Equality of opportunity - In binary machine-learning fairness, equality of opportunity usually means that the true-positive rate is the same across the groups being compared.
• Equalized odds - Equalized odds is a statistical fairness criterion requiring a model's predictions to be independent of group membership once the actual target outcome is...
• Error propagation - Error propagation is the way an error or uncertainty in one part of a process affects later calculations, predictions, or decisions.
• Ethics of artificial intelligence - Ethics of artificial intelligence is the study and practice of evaluating what ought to be done when AI is designed, developed, deployed, and used.
• Evaluation Bias - Evaluation bias arises when the data, benchmarks, measurements, or assessment procedures used to judge an AI system do not support an accurate picture of...
• Evaluation machine learning pipeline - An evaluation machine learning pipeline, in the ITU telecommunications context, connects ML pipeline functions with simulated network functions and served...

Sources

IEEE Standards Association, IEEE 7000 model process for ethical concerns during system design: https://standards.ieee.org/news/ieee-7000/ IEEE 7000-2021 official standard abstract: https://ieeexplore.ieee.org/document/9536679 UNESCO, Recommendation on the Ethics of Artificial Intelligence: https://www.unesco.org/en/articles/recommendation-ethics-artificial-intelligence

Previous: Eta squared ( h 2 ) Next: Ethics of artificial intelligence