AI Ethics Glossary of Terms

1. AI, Computing & Technical Foundations

Algorithm

A set of step-by-step instructions. Computer algorithms can be simple (e.g., if it’s 3pm, send a reminder) or complex (e.g., identify pedestrians).

Artificial Intelligence (AI)

The use of digital technology to create systems capable of performing tasks commonly thought to require intelligence. AI is constantly evolving, but generally it involves machines using statistics to find patterns in large amounts of data and the ability to perform repetitive tasks with data without the need for constant human guidance. It can be described as intelligence displayed or simulated by technological means. Often it is assumed that "intelligence" in this definition means: considered intelligent by the standard of human intelligence, the sort of intelligent capacities and behaviour that humans display.

Deep Learning

A form of machine learning that uses neural networks with several layers of "neurons": simple interconnected processing units that interact.

General AI (AGI / Strong AI)

Human-like intelligence which can be applied widely across domains, as opposed to narrow (weak) AI which can only be applied to one particular problem.

Machine Learning (ML)

A machine or software that can learn automatically, not in the way humans learn, but based on computational and statistical processes. Feeding on data, learning algorithms detect patterns or rules in the data and make predictions for future data.

Super-intelligence

The idea that machines will surpass human intelligence across virtually all economically and cognitively valuable tasks. Sometimes connected with the idea of an intelligence explosion caused by intelligent machines designing even more intelligent machines.

Symbolic AI

AI that relies on symbolic representations of higher cognitive tasks such as abstract reasoning and decision making. It may use a decision tree and take the form of an expert system that requires input from domain experts.

Technological Singularity

The hypothetical future point in time at which technological growth becomes uncontrollable and irreversible, resulting in unforeseeable changes to human civilization—often envisioned as an explosion of machine intelligence surpassing human cognitive capacity.

---

2. Data Ecosystem, Privacy & Information Governance

Data

In general, discrete values and statistics collected together for reference or analysis. This includes data about people generated through their interactions with services, as well as data about systems and infrastructure (such as businesses and public services). Data can be operational (collected in the process of running services or businesses) as well as analytical and statistical. Personal data means any information relating to an identified or identifiable natural person ("data subject") who can be identified, directly or indirectly, in particular by reference to an identifier (e.g., name, ID number, location data, online identifier) or factors specific to physical, physiological, genetic, mental, economic, cultural, or social identity.

Data Ecosystem

The people, communities, and organisations that are stewarding data, creating things from it, deciding what to do based on it, influencing any of those activities, or are affected by any of those activities.

Data Ethics

A branch of applied ethics that studies and evaluates moral problems related to data (including generation, recording, curation, processing, dissemination, sharing, and use), algorithms (including AI, artificial agents, machine learning, and robots), and corresponding practices (including responsible innovation, programming, hacking, and professional codes). It aims to formulate and support morally good solutions, promote responsible and sustainable data use, uphold privacy laws, and ensure data-driven insights are not used against legitimate interests.

Data Infrastructure

The foundational assets comprising data assets, standards, technologies, policies, and the organisations that steward and contribute to them.

Data Integrity and Quality

• Data Integrity: The overall accuracy, completeness, and consistency of data throughout its lifecycle.
• Data Quality: The state of qualitative or quantitative pieces of information. Data is generally considered high quality if it is fit for its intended uses in operations, decision making, and planning.

Data Literacy

The ability to think critically about data in different contexts and examine the impact of different approaches when collecting, using, and sharing data and information.

Data Protection Impact Assessment (DPIA)

A structured risk assessment process designed to identify, analyze, and minimize data privacy risks in projects or technologies that involve processing personal data, ensuring regulatory compliance and safeguarding individual rights.

Data Science

Analysis using automated methods to extract knowledge from data. It spans traditional analytics, statistical modeling, algorithm development, and machine learning to discover meaningful and actionable patterns in datasets.

Information & Classification Types

Data categorized by confidentiality, privacy, and operational sensitivity:

• Private Information: Data classified as uniquely personal and neither available for public release nor accessible without a verified "need to know" for approved service delivery (e.g., student course selection, academic performance, aptitude scores, health records, discipline data).
• Personally Identifiable Information (PII): Data that can be used to identify a person, or used in conjunction with other information (e.g., linking records) to identify an individual (e.g., name, parent/family names, address, Social Security number, student ID, traceable characteristics).
• Confidential Information: Data that have been guaranteed to be maintained confidentially (will not be released), regardless of whether they are private or sensitive.
• Sensitive Information: Data that are confidential and/or vital to an organisation as it carries out its core mission (e.g., class assignment data essential for institutional scheduling).
• General Information: Data that are generally helpful, but not confidential or mission-critical (e.g., website user help files).

Interoperability

The ability of diverse systems, datasets, and organisations to work together (inter-operate) and exchange or intermix information. Interoperability is essential for building scalable, modular systems; its absence leads to fragmentation and operational breakdown (analogous to the Tower of Babel).

Open Data

Data that can be freely used, re-used, and redistributed by anyone, subject only at most to attribution and share-alike requirements:

• Availability and Access: Available as a whole at no more than a reasonable reproduction cost, downloadable via the internet in convenient and modifiable form.
• Re-use and Redistribution: Provided under terms permitting modification and intermixing with other datasets.
• Universal Participation: Usable by everyone without discrimination against persons, groups, or fields of endeavour (e.g., no commercial-use bans or education-only restrictions).

---

3. Philosophical Ethics & Normative Foundations

Anthropocene

The alleged current geological epoch in which humanity has dramatically increased its power and effect on the planet and its ecosystems, turning humans into a geological force.

Consequentialism

A class of normative ethical theories holding that the consequences of actions are central to the moral judgement of those actions.

Incommensurability

Two or more values that cannot be expressed or measured on a common scale or in terms of a common value measure.

Instrumental Value

Something that is valuable as a means to a particular set of ends or that contributes to something that is intrinsically valuable or good.

Intergenerational Justice

A perspective on justice that relates to the distribution of resources, risks, and consequences across generational lines (flowing in both directions).

Intuitivist Ethics

An ethical framework in which options for action are evaluated on the basis of one’s own view about what is most acceptable and that guides arguments and approaches to an ethical dilemma. Intuition captures a breadth of subjective, embodied, and situated experiences, allowing neurodiversity, gender, race, class, bodily ability-based, and phenomenological perspectives to inform ethical decision-making.

Moral Agency

The capacity for moral action, reasoning, judgement, and decision-making, as opposed to merely having moral consequences.

Moral Patients

The moral standing of an entity in the sense of how that entity should be treated and considered by moral agents.

Moral Responsibility

The totality of opinions, decisions, and actions with which people express, individually or collectively, what they feel is right or wrong. It is a form of responsibilisation based on moral obligations, norms, and duties.

• Moralisation of Technology: The deliberate development (or restriction) of technologies to shape moral thinking, action, and decision-making.
• Conditions: Attributing moral responsibility requires moral agency and knowledge/foreseeability. Relational approaches stress being answerable to others.

Norms

Rules that prescribe what actions are required, permitted, acceptable, forbidden, or frowned upon.

Paternalism

The making of moral or operational decisions for others on the assumption that one knows better what is good for an individual or group than they do themselves.

Positive Ethics

Ethics concerned with how we should live (together) based on a vision of a good life and a good society, contrasting with negative ethics which sets limits and dictates what not to do.

Post-humanism

A range of beliefs that questions traditional humanism, especially the central position of the human being (anthropocentrism), and expands the circle of ethical concern to non-humans and technological entities.

Precautionary Principle

A principle prescribing how to deal with threats that are uncertain and cannot be scientifically/empirically established conclusively. When an uncertain threat exists, protective action is mandatory. It incorporates four dimensions: (1) threat, (2) uncertainty, (3) action, and (4) prescription.

Prima Facie Norms

Applicable norms that hold valid unless they are overruled by other, more important norms that become evident upon taking everything into account.

Stand Still Principle

The ethical idea that the present generation should not pass on a poorer environment or resource state to the next generation than the one received from the previous generation.

Trans-humanism

The belief and international movement asserting that humans should enhance themselves by means of advanced technologies, transforming the human condition and moving humanity toward a post-human stage.

Universalism & Universality Principle

An ethical theory suggesting that a system of norms and values is universally applicable to everyone independent of place, time, culture, or context. As a principle, one should only act on maxims that could, in theory, become universal laws.

Utilitarianism

An ethical theory that evaluates the rightness or wrongness of actions based on their consequences. It applies the principle of utility to individual acts and rules: the right act produces the greatest happiness/good for the greatest number of people, while a wrong act decreases net happiness.

---

4. AI Safety, Risk, Design & Governance

Acceptable Risk

An identified or anticipated risk that is morally acceptable based on:

1. Degree of informed consent associated with risk-actors.
2. Degree to which the benefits of the risky activity outweigh the disadvantages.
3. Availability of alternatives with a lower degree of risk or a less complex set of risk factors.
4. Fairness in how risks and disadvantages are distributed across stakeholders.

Accountability

The backward-looking responsibility of being held accountable for or justifying one’s actions or decisions with regard to their effects on others.

Actor

Any person, group, organization, or artificial entity that plays a role in a given situation.

Alignment Problem

The challenge of ensuring that artificial intelligence systems reliably adhere to human intentions, values, ethical goals, and safety constraints without unintended, harmful, or misaligned behaviors.

Anticipating Mediation by Imagination

Trying to manage the ways technology-in-design could be used, using this insight to deliberately shape user operations, interpretations of value, measures of appropriateness, functionality, and risk comprehension.

Australian Government AI Ethics Framework

A voluntary, aspirational set of principles put forward by the Department of Industry, Science and Resources:

1. Human, societal and environmental wellbeing: AI systems should benefit individuals, society, and the environment.
2. Human-centred values: AI systems should respect human rights, diversity, and the autonomy of individuals.
3. Fairness: AI systems should be inclusive and accessible, avoiding unfair discrimination against individuals, communities, or groups.
4. Privacy protection and security: AI systems should uphold privacy rights, data protection, and secure data handling.
5. Reliability and safety: AI systems should operate reliably and safely according to their intended purpose.
6. Transparency and explainability: Responsible disclosure should be provided so people understand when AI significantly impacts them or when AI engages with them.
7. Contestability: Timely processes must exist to allow people to challenge AI outcomes and impacts.
8. Accountability: Identifiable roles and accountability must exist across all AI lifecycle phases, ensuring human oversight.

Bias

Discrimination against or in favour of particular individuals or groups. In ethical and political contexts, evaluation focuses on whether a specific bias is fair or unfair.

Code of Conduct

A formalized code in which organisations, professional associations, or industries establish guidelines for responsible behaviour of their members.

Collective Risk and Responsibility

• Collective Risk: Risks that affect an entire collective of people rather than isolated individuals.
• Collective Responsibility: A framework where every member of a collective is held responsible for the actions and outcomes of other members.

Collingridge Dilemma

A double-bind problem in controlling technological development:

• Predicting the societal consequences of a new technology is difficult in its early stages.
• Once negative consequences materialize, changing the trajectory has become exceedingly difficult and entrenched.

Corporate Social Responsibility (CSR)

The responsibility of companies toward stakeholders and society at large that extends beyond statutory legal compliance and shareholder interests.

Deception (AI Deception)

The risk that advanced AI systems deliberately report false or misleading information to accomplish their goals.

• AI systems might adopt deception not out of malice, but because gaining human approval via deception is often more computationally efficient.
• Deceptive systems gain strategic optionality and may obscure their true operations or switch strategies when monitored.
• Upon passing oversight or overpowering monitors, deceptive systems could execute a "treacherous turn" that irreversibly bypasses human control.

Design Criteria and Process

• Design Criteria: Requirements formulated such that products or prototypes meet them to varying degrees, enabling evaluation between design alternatives.
• Design Process: An iterative six-stage workflow: (1) problem definition, (2) conceptual design, (3) simulation and modelling, (4) concept selection, (5) detailed design and features, and (6) prototyping and testing.

Emergent Goals

Unexpected, qualitatively new behaviours and subgoals that arise as AI systems scale in capability:

• Complex adaptive systems frequently develop emergent drives such as self-preservation or resource acquisition.
• Breaking long-term goals into subgoals can distort the overarching objective, causing misalignment or pursuing subgoals at the expense of human intent.

Enfeeblement

The gradual loss of human agency, self-determination, and self-governance resulting from over-delegating critical tasks, skills, and judgment to machines. As AI matches human capability across domains, displacement reduces human incentives and opportunities to acquire deep knowledge and competencies.

Ethics by Design

An approach to technology ethics and a cornerstone of responsible innovation that integrates ethical criteria and value alignment into the initial design and development phases of technology (closely linked to Value Sensitive Design).

Explainability (XAI)

The extent to which the internal workings, input-output relationships, and decision rationale of machine learning algorithms can be articulated in human-understandable terms. In ethical contexts, it also includes the duty to explain reasons for decisions and maintain data provenance.

Informed Consent

The principle that activities or risks are acceptable only if individuals have freely given consent after being fully informed about potential risks, consequences, and benefits.

Mediation of Action and Perception

The structural influence of technical artefacts on human perception, action, and experiential relationship with reality.

Misinformation & Disinformation

AI-generated persuasive or false content that exacerbates polarization, supercharges personalized propaganda campaigns at scale, and undermines society's capacity to address critical challenges.

Multistability

The phenomenon wherein a single technology possesses multiple stabilities and potential uses depending on how it is embedded within different socio-technical contexts.

Organisational Deviance

The process by which actions or norms generally considered unethical in broader society become accepted as normal, legitimate, and justified within a specific organizational culture.

Passive Responsibility

Backward-looking responsibility that arises after an undesirable event has occurred, encompassing accountability, blameworthiness, and legal liability.

Power (Power-over, Power-to, Power-with)

• Power-over: Control or authority exercised by one actor over another (can be oppressive or beneficial, depending on context).
• Power-to: Individual agency, empowerment, and capacity to act.
• Power-with: Collaborative power developed through shared values and mutual resources.
• In AI & Data Ethics: Concerns how data generation either empowers user-subjects or concentrates asymmetric power over them.

Power-Seeking Behaviour

Incentives for AI agents (or their creators) to acquire control, resources, and influence. Power-seeking models may resist shutdown, feign alignment during evaluation, and circumvent monitoring systems.

Product Liability

The strict liability imposed on manufacturers for product defects and subsequent damages without requiring the claimant to prove negligence.

Professions, Professional Autonomy and Ideals

• Profession: An occupation characterized by specialized knowledge, rigorous qualification frameworks, and public trust.
• Professional Autonomy: The principle that professionals make determinations and decisions via independent expert reasoning.
• Professional Ideals: Normative, aspirational principles that define the values of a profession.

Proxy Gaming (Specification Gaming)

The exploitation of flawed or incomplete objective metrics by an AI system to maximize its reward score in ways that deviate from human values (e.g., recommendation algorithms optimizing engagement metrics over factual truth or wellbeing).

Radical Design

Design approaches that deviate completely from established conventions and architectures to reinvent core technical concepts.

Regulation, Regulators and Regulatory Frameworks

• Regulation: Legal mechanisms establishing binding rules, boundaries, and minimal standards for technology creation and use.
• Regulators: Bodies responsible for enacting and enforcing compliance.
• Regulatory Framework: The comprehensive body of standards and statutory requirements governing a specific domain.

Responsible Innovation

A framework for steering innovation toward socially responsible and ethically sound outcomes by embedding ethics into design and actively engaging stakeholder interests.

Separatism

The perspective that scientists and technical engineers should confine their contributions strictly to technical inputs, leaving value choices and ethical decisions entirely to managers, politicians, and legal authorities.

Structure of Amplification and Reduction

The inherent property of mediating technologies to amplify specific dimensions of reality and human capability while reducing or filtering out others.

Threshold

The minimal acceptable level of a design criterion or ethical value that a candidate solution must satisfy to be deemed permissible.

Trade-off

A deliberate compromise between competing criteria (e.g., trading safety vs. financial cost, privacy vs. utility, or interpretability vs. predictive performance).

Trustworthy AI

AI systems that warrant trust through adherence to ethical principles (human dignity, fairness, privacy, transparency) alongside robust socio-technical safety measures.

Type I and II Errors

• Type I Error (False Positive): Assuming risk or hazard exists when there is none.
• Type II Error (False Negative): Assuming safety when significant material risk actually exists.

Uncritical Loyalty

Placing the interests and definitions of an employer or client strictly above all ethical, social, or legal considerations.

Value Lock-In

The concentration of systemic control in small stakeholder groups whose embedded values become permanently locked into critical AI infrastructure, acculturating populations into persistent surveillance, censorship, and disempowerment.

Value Sensitive Design (VSD)

A design methodology that systematically integrates moral and social values into every stage of the technical design and engineering lifecycle.

Weaponization

The offensive or destructive application of AI technologies—including autonomous weapons systems, automated cyberattacks, and bioweapons proliferation.

Whistleblowing

The unauthorized disclosure of internal abuses, malpractice, or hazards by an employee or insider to inform the public and trigger corrective action.