CDEI key terms

The branch of applied ethics that examines how data is collected, used, shared, and disposed of, weighing harms and benefits against moral principles and societal values.

The systematic study of right and wrong conduct that supplies the underlying theories, such as deontology and utilitarianism, used to justify ethical judgments about data practices.

An ethical theory holding that certain duties and rules bind us regardless of consequences, so some data uses are impermissible even if they produce net benefit.

A consequentialist theory judging actions by the overall balance of benefits and harms they produce, often used to weigh public-good data uses against individual costs.

An approach focused on the character traits and dispositions of the people handling data, asking what a honest, prudent, and fair practitioner would do.

A framework holding that social arrangements are just when they benefit the least advantaged and arise from fair procedures, applied to distribution of data benefits and burdens.

The branch of justice concerned with how goods, opportunities, and harms are allocated across groups, used to evaluate whether data-driven benefits are shared equitably.

The ethical obligation to act for the benefit of others, in data contexts meaning designing systems and studies to produce genuine welfare gains rather than harm.

The duty to avoid causing harm, requiring practitioners to identify and mitigate foreseeable harms from data collection, analysis, and deployment.

Agreement to data processing given voluntarily by a competent person after receiving clear information about purpose, risks, and alternatives, without coercion or deception.

The principle that individuals should control decisions about their own data and person, grounding rights to consent, access, and refusal.

Any information relating to an identified or identifiable natural person, including identifiers, location, and combinations that can single someone out.

Personally identifiable information: data elements that can directly or indirectly identify a specific individual, a category triggering heightened protection obligations.

Irreversibly transforming data so that no individual can be re-identified by any reasonably available means, removing it from personal-data protections.

Removing or masking direct identifiers to reduce identifiability while accepting some residual re-identification risk, a weaker protection than full anonymization.

The likelihood that anonymized or de-identified records can be linked back to specific people using auxiliary data, inference, or linkage attacks.

A privacy model requiring every record to be indistinguishable from at least k-1 others on quasi-identifiers, limiting linkage-based disclosure.

A formal guarantee that analysis outputs change only slightly when any single person's data is added or removed, bounding inference about individuals.

The principle that data collected for one stated purpose may not be repurposed for unrelated uses without fresh justification or consent.

Collecting and retaining only the data actually needed for a stated purpose, reducing exposure, breach impact, and privacy intrusion.

Systematic and repeatable errors or skewed outputs in automated systems that disadvantage particular groups, arising from data, design, or deployment choices.

Bias that enters a model because the training data reflects past discrimination or structural inequality, causing the system to reproduce those patterns.

Distortion caused when the data collected is not representative of the population the model will serve, leading to inaccurate or unfair outputs for underrepresented groups.

The study and practice of measuring and mitigating unfair disparities in automated decisions, using criteria such as equalized odds or demographic parity.

A facially neutral practice that produces disproportionately adverse outcomes for a protected group, a key legal and ethical fairness concern.

A characteristic such as race, sex, age, or disability that laws shield from discriminatory treatment, often restricted as a model input.

A feature correlated with a protected attribute, such as ZIP code standing in for race, that can reintroduce discrimination even when the protected attribute is excluded.

The accountable management of data assets to ensure quality, security, appropriate use, and compliance throughout the data lifecycle.

The policies, roles, and decision rights that determine how data is acquired, accessed, protected, and used across an organization.

The stages data passes through, from creation and collection to processing, sharing, archiving, and deletion, each carrying distinct ethical obligations.

Documentation of the origin, custody, and transformation history of a dataset, supporting verification of legality, quality, and fitness for use.

A record tracing how data flows and changes across systems and processing steps, enabling impact analysis and accountability for derived outputs.

The degree to which the reasoning behind a system's output can be understood and communicated to affected people, auditors, and regulators.

The property of a model whose internal logic is understandable to humans by design, as opposed to post-hoc approximations of complex models.

The entitlement of individuals to meaningful information about the logic and consequences of automated decisions affecting them.

The obligation of organizations and practitioners to answer for data practices, including documented decisions, remedies, and consequences for failures.

The property enabling independent review of a system's data, design, and decisions against stated standards, supported by logs and documentation.

Legal entitlements of individuals over their personal data, commonly including access, correction, deletion, portability, and objection.

The EU General Data Protection Regulation, a comprehensive privacy law imposing lawful-basis, transparency, and accountability duties on data controllers and processors.

A structured pre-deployment review that identifies and mitigates privacy risks of high-risk processing, often required by law before launch.

A process for evaluating the potential harms of an automated system to individuals and groups before and after deployment, with mitigation and monitoring.

A system whose internal decision logic is opaque to users and even developers, complicating explanation, contestability, and error diagnosis.

A design in which a person reviews or approves automated outputs before consequential action, preserving judgment and providing a route to contest decisions.

Oversight that is informed, timely, and empowered to override a system, as distinct from nominal review where the human simply rubber-stamps outputs.

A short document reporting a model's intended use, training data, performance across groups, and limitations to support informed and responsible reuse.

Standardized documentation describing a dataset's motivation, composition, collection process, and recommended uses, aiding ethical evaluation by downstream users.

Openness about what data is collected, how systems work, and what decisions are made, enabling scrutiny by affected parties and regulators.

The principle that personal data should be kept only as long as necessary for its stated purpose, after which it should be deleted or anonymized.