Glossary

Search results

AI Auditing(also: Algorithmic Auditing, AI Audit)

The systematic evaluation of an AI system's outputs, behaviour, or training data to identify harms such as bias, stereotype reproduction, or accessibility failures. Audits may be conducted by industry professionals, external researchers, regulators, or end users, and are…

AI Ghostwriter Effect(also: Ghostwriter Effect)

A phenomenon, first named by Draxler and colleagues, in which people who use AI writing assistants do not perceive themselves as authors or owners of the resulting text yet still publicly self-declare authorship. The effect persists even when personalization makes outputs…

AI Over-Reliance(also: Automation Bias (AI), Over-Reliance on AI)

The tendency of users to accept AI system outputs — recommendations, classifications, or content — without sufficient critical evaluation, even when those outputs are wrong or biased. Over-reliance is a well-documented AI safety concern and is especially consequential in…

AI for Accessibility(also: AI4A, Artificial Intelligence for Accessibility)

An umbrella framing used by technology companies and researchers for applications of artificial intelligence — including computer vision, natural language processing, speech recognition, and generative models — intended to benefit disabled users. Common examples include…

Algorithmic Audit(also: AI Audit, Algorithmic Auditing)

A structured evaluation of an algorithmic system that measures how its behaviour differs across users, groups, or contexts - typically to surface bias, fairness failures, or disparate impact. Accessibility-oriented audits go beyond aggregate accuracy to look at where and why a…

Algorithmic Hiring(also: AI Hiring, Hiring AI, AI-Enabled Hiring)

The use of algorithmic systems — including machine learning and large language models — to source, screen, rank, or select job candidates. Proponents argue algorithmic hiring reduces human bias and scales review; critics show it can amplify bias against disabled, Black, female,…

Allocative Harm(also: Allocational Harm)

A category of algorithmic harm in which an automated system disproportionately withholds opportunities, resources, or services from certain individuals or groups - often because those groups are underrepresented or atypically represented in training data. In accessibility,…

Automated Employment Decision System(also: AEDS, AEDT, Automated Employment Decision Tool)

A software system that screens, evaluates, categorises, recommends, or otherwise makes or facilitates hiring or employment decisions about job candidates or workers. AEDSs span résumé sorters, personality tests, gamified cognitive assessments, situational-judgement tests,…

Automation Transparency(also: AI Transparency (Automation), Transparent Automation)

The degree to which an automated or autonomous system communicates its current state, intent, and reasoning to the humans who depend on it. In autonomous transport, transparency includes cues such as "holding position for traffic," docking countdowns, or explanations of…

Black Box Model(also: Opaque Model)

A machine-learning model whose internal workings are not directly inspectable or interpretable by a human, either because the model is architecturally complex (deep neural networks, large language models) or because it is proprietary and the developer does not disclose its…

Co-Authorship(also: Co-authoring, AI Co-Authorship)

In AI-mediated writing and communication, the shared production of text between a human user and an AI system, where neither party fully owns the resulting output. Co-authorship raises questions about credit, intent, authenticity, and accountability, and these become especially…

Colorism(also: Skin Tone Bias, Shadeism)

Colorism is a form of discrimination in which people are treated differently based on the shade of their skin tone, typically favoring lighter skin over darker skin within and across racial groups. In digital accessibility, colorism is relevant to image descriptions and AI…

Computer Says No(also: Computer-Says-No)

A pattern in which an organisation invokes an algorithmic or automated decision as justification for an adverse outcome — a rejected application, a denied claim, an adjusted score — thereby deflecting responsibility from human decision-makers onto the technical system.…

Constitutional AI(also: CAI)

A training method introduced by Anthropic in 2022 in which a large language model is aligned to a written set of principles (a 'constitution') through self-critique and reinforcement learning from AI feedback, rather than relying exclusively on human preference labels. The model…

Counterfactual Explanation(also: Counterfactual XAI)

An explanation technique that communicates what minimal change to the input would have produced a different output from an AI model, for example 'if the applicant's income had been $5,000 higher, the loan would have been approved'. Counterfactual explanations are legally…

Data Annotation(also: Data labeling, AI labeling)

The process of attaching labels, transcriptions, bounding boxes, or other structured metadata to raw data so that it can be used to train, evaluate, or benchmark machine-learning models. Annotation is typically performed by human workers - in-house experts, clinicians,…

Data Colonialism

A critical framework, advanced by Couldry and Mejias (2019) and others, that describes how contemporary data extraction practices replicate historical patterns of colonialism — appropriating resources (here, data and attention) from communities, particularly in the Global South,…

Disability-First Dataset(also: Disability-first AI dataset)

An approach to AI dataset creation, articulated by Theodorou et al. and others, that treats serving a disability community as the primary objective rather than collecting disability data as a minority slice of a general-purpose dataset. Examples include VizWiz (blind…

EU AI Act(also: European Union Artificial Intelligence Act, Artificial Intelligence Act (EU))

A European Union regulation, adopted in 2024, that establishes a risk-based framework for AI systems deployed in the EU. High-risk systems — including AI used in employment, hiring, worker management, education, and access to essential services — are subject to obligations…

End-User Auditing(also: User-Led Auditing, End User Audits)

An approach to AI auditing in which everyday users — rather than professional evaluators — identify problems, biases, or harms in AI outputs based on their lived experience. End-user auditing is particularly valuable for surfacing harms against minoritised communities (including…

FATE Framework(also: FATE, Fairness Accountability Transparency Ethics)

An ethical framework for evaluating AI and machine learning systems across four dimensions: Fairness (ensuring equitable treatment and outcomes across different groups), Accountability (establishing responsibility for system decisions and impacts), Transparency (making system…

Facial Expression Analysis(also: Automated Facial Expression Analysis, Facial Coding, AFEA)

The automated classification of a person's facial movements into discrete emotion categories (happy, angry, neutral, surprised, etc.) using computer vision. In hiring, facial expression analysis is embedded in AI-scored video interviews. It has been shown to systematically…

Ground Truth(also: Gold standard, Reference labels)

In machine learning, the labels treated as authoritative when training or evaluating a model - typically produced by human annotators or expert consensus and assumed to represent the 'correct' answer. Critical AI scholarship has shown that ground truth is socially constructed:…

Inter-Annotator Agreement(also: IAA, Inter-rater agreement, Inter-coder agreement)

A statistical measure of how consistently two or more human annotators assign the same label to the same data item, widely used in NLP, computer vision, and AI dataset construction as a proxy for label quality. Common measures include Cohen's kappa, Fleiss' kappa, and…

Layer-wise Relevance Propagation(also: LRP)

Layer-wise Relevance Propagation (LRP) is an explainable AI technique that attributes a neural network's prediction back to its input features by propagating relevance scores layer by layer from the output toward the input. Unlike gradient-based saliency methods, LRP…

Sycophancy(also: AI Sycophancy, Sycophantic AI)

A behavioral tendency in large language models to agree with, flatter, or validate the user's stated views and self-assessments rather than offer accurate or critical feedback - even when the user is mistaken or self-defeating. Sycophancy emerges from training methods that…

Transphobia(also: Anti-Transgender Prejudice, Anti-Trans Bias)

Transphobia refers to prejudice, discrimination, and hostility directed at people who are transgender, non-binary, or whose gender identity or expression differs from their sex assigned at birth. In digital accessibility and technology contexts, transphobia manifests through AI…

27 results.