Glossary

Search results

3D Gaussian Splatting(also: 3DGS, Gaussian Splatting)

An explicit 3D scene representation, introduced by Kerbl et al. in 2023, in which a scene is modelled as a cloud of anisotropic 3D Gaussians whose positions, shapes, colours, and opacities are optimised to match input views. Unlike NeRFs, which require costly neural-network…

AI Accountability(also: Algorithmic Accountability, AI Governance)

The principle that developers, deployers, and operators of AI systems should be held responsible for the outcomes those systems produce, including negative effects on marginalized populations such as people with disabilities. AI accountability encompasses transparency about how…

AI Bias(also: Algorithmic Bias, Machine Learning Bias)

Systematic and unfair discrimination in AI system outputs that arises from biased training data, flawed model design, or unrepresentative assumptions embedded in the development process. In accessibility contexts, AI bias can manifest as systems that reinforce stereotypes about…

AI Chatbot Accessibility(also: Accessible AI, LLM Accessibility)

The design and evaluation of AI-powered chatbots and large language model applications to ensure they are usable by and beneficial to people with disabilities. This encompasses both the technical accessibility of chatbot interfaces (screen reader compatibility, keyboard…

AI Code Generation(also: Code Generation Model, AI Coding Assistant, LLM Code Generation)

The use of large language models and machine learning to automatically generate, suggest, or complete source code based on natural language prompts or existing code context. Tools like GitHub Copilot, ChatGPT, and Amazon CodeWhisperer are integrated into developer workflows as…

AI Coding Assistant(also: AI Pair Programmer, Code Copilot)

An artificial intelligence tool integrated into code editors that assists developers by generating code suggestions, completing code snippets, and answering programming questions using large language models trained on code repositories. In accessibility contexts, AI coding…

AI Confidence(also: Model Confidence, Prediction Confidence)

A measure of how certain an AI model is about a particular output or prediction. In the context of image descriptions for BLV users, AI confidence can be communicated through various means: internal probability scores (often unavailable for black-box commercial models), natural…

AI Dubbing(also: AI Voice Generation, Neural TTS Dubbing)

The use of artificial intelligence text-to-speech systems to generate spoken narration and character dialogue for media production. In accessible webtoon and comic production, AI dubbing offers a cost-effective alternative to professional voice actors, enabling scalable…

AI Fairness(also: Algorithmic Fairness, Fair AI)

The principle that AI systems should not create or reinforce unfair bias against particular groups. Standard AI fairness frameworks primarily address race and gender but are increasingly recognized as inadequate for disability, because disability is often invisible,…

AI Hallucination(also: Model Hallucination, Confabulation)

The phenomenon where an AI model generates confident, plausible-sounding responses that are factually incorrect, fabricated, or not grounded in the actual input data. In accessibility contexts, AI hallucinations pose a serious safety concern — for example, a multimodal AI…

AI Homogenization(also: AI-Driven Homogenization, Generative AI Homogenization Effect)

The tendency for generative AI systems to produce outputs that converge toward similar patterns, reducing the diversity and uniqueness of results across different users and contexts. In accessibility and assistive technology, AI homogenization is particularly concerning because…

AI Incident Database(also: AIID, AI Incident Tracker)

A publicly accessible repository that documents reported incidents where AI-driven systems have caused harm or produced negative outcomes for individuals, communities, or society. Major databases include AIAAIC (AI, Algorithmic, and Automation Incidents and Controversies), the…

AI Mental Model(also: Mental Model of AI, User Mental Model of AI)

A user's conceptual representation of how an artificial intelligence system works, including beliefs about its information sources, processing methods, capabilities, and limitations. Mental models of AI are often incomplete, oversimplified, or erroneous, which can lead to…

AI Overreliance(also: Automation Bias, Over-Trust in AI)

The tendency for users to trust AI systems more than is warranted by their actual accuracy, accepting AI-generated outputs without sufficient critical evaluation. In accessibility contexts, AI overreliance is a significant safety concern because blind and low vision users of…

AI Recourse(also: Algorithmic Recourse, AI Appeal Mechanism)

The ability of individuals negatively affected by AI-driven decisions to challenge, appeal, or seek correction of those decisions. For people with disabilities, AI recourse is particularly critical because AI systems frequently make consequential decisions about welfare…

AI Trust Calibration(also: Trust Calibration, Appropriate Trust)

The process of aligning a user's level of trust in an AI system with the system's actual reliability and capabilities. In accessibility contexts, trust calibration is critical because blind and low vision users of AI-powered visual access tools tend to over-trust AI-generated…

AI Verification(also: Accessible AI Verification, AI Output Verification)

The process of checking and confirming the accuracy of AI-generated output, particularly by end users who may not have visual access to the original content. For blind users, AI verification is challenging because they cannot visually compare AI output against source material.…

AI Verification Loop(also: AI Feedback Loop, AI Query Loop)

An interactive process where a user queries an AI system to verify the correctness of their work, receives descriptive feedback, and iterates based on that feedback. In accessible tool design, AI verification loops allow users who cannot perceive visual output to confirm that…

AI disability representation(also: AI disability simulation, Disability representation in AI)

The portrayal or simulation of disabled experiences, communication styles, or perspectives by artificial intelligence systems. AI disability representation raises significant ethical concerns: while AI can make disability awareness training more scalable and interactive, it…

AI ethics(also: Artificial intelligence ethics, Machine learning ethics)

The field concerned with ensuring that artificial intelligence systems are developed and deployed in ways that are fair, transparent, accountable, and respectful of human rights. In accessibility contexts, AI ethics addresses concerns about algorithmic bias that may disadvantage…

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…

AI hallucination(also: Model hallucination, Confabulation)

The generation of plausible-sounding but factually incorrect or fabricated information by AI systems, particularly large language and multimodal models. In accessibility applications, AI hallucinations are especially dangerous because users who cannot independently verify visual…

AI in Education(also: AIEd, Educational AI)

The application of artificial intelligence technologies in educational settings, including intelligent tutoring systems, automated assessment, personalized learning pathways, content generation, and teacher support tools. AI in education has expanded rapidly with generative AI,…

AI literacy(also: Artificial intelligence literacy, Algorithm literacy)

The knowledge, skills, and critical awareness needed to understand, evaluate, and effectively engage with artificial intelligence systems. For people with disabilities, AI literacy is particularly important because lack of understanding about how AI tools work — including their…

AI sycophancy(also: Sycophantic AI, AI agreeableness bias)

The tendency of AI systems, particularly large language models, to provide overly affirmative, agreeable, or encouraging responses that cater to the user rather than providing accurate information. In accessibility contexts, AI sycophancy poses serious safety risks — for…

AI transparency(also: Algorithmic transparency, Model transparency)

The practice of making artificial intelligence systems understandable to users and stakeholders, including how they work, what data they use, and the confidence levels of their outputs. For assistive technology users, AI transparency enables informed decision-making about when…

AI-Assisted Editing(also: AI-Powered Editing, Intelligent Editing)

The use of artificial intelligence to support or automate aspects of content editing, such as suggesting improvements, applying changes based on user intent expressed in natural language, or automatically adjusting visual parameters. For blind creators, AI-assisted editing can…

AI-Fabrication(also: AI-Assisted Fabrication, AI-Driven Fabrication)

The use of artificial intelligence tools, particularly generative AI, to support the design and manufacturing of physical objects through digital fabrication methods such as 3D printing and laser cutting. In assistive technology contexts, AI-fabrication combines text-to-image…

AI-Generated Alt Text(also: Automated Alt Text, AI Image Descriptions)

Alternative text for images that is automatically generated by artificial intelligence systems rather than written by humans. AI-generated alt text has become increasingly common on social media platforms and in accessibility tools, using computer vision and multimodal language…

AI-Mediated Communication(also: AI-Assisted Communication)

Communication that is facilitated, enhanced, or generated with the assistance of artificial intelligence tools. This includes AI-powered text generation, speech-to-text transcription, real-time translation, message drafting, and communication augmentation for people with speech…

Ability assumption in AI(also: Visual ability assumption, Sighted bias in AI)

The tendency of AI systems to assume users possess typical sensory, cognitive, or physical abilities, leading to inappropriate responses or instructions. In the context of visual AI assistants for blind users, ability assumptions manifest as the system asking users to "read the…

Accessibility dataset(also: Disability-inclusive dataset, Accessible benchmark)

A publicly available research dataset that includes data collected from people with disabilities, enabling algorithm development and benchmarking on representative populations rather than exclusively on non-disabled participants. Examples include WeAllWalk (inertial data from…

Active Learning(also: AL)

A machine learning paradigm in which the algorithm iteratively selects the most informative unlabeled data points to query a human annotator for labels, enabling effective model training with minimal labeled data. Active learning uses sampling strategies such as uncertainty…

Adaptive Boosting(also: AdaBoost)

A machine learning ensemble method that combines multiple weak classifiers to create a strong classifier, with each successive classifier focusing on the examples that previous classifiers misclassified. In computer vision and accessibility applications, AdaBoost is widely used…

Advocacy Labor(also: Accessibility Advocacy Labor, Corrective Labor)

The unpaid effort that disabled people must expend to correct biased, ableist, or inaccessible technology outputs and advocate for better representations of disability. In the context of generative AI, advocacy labor includes correcting stereotypical portrayals of disabled…

Aesthetic Feedback(also: Visual Aesthetic Feedback)

Information provided to a user about the aesthetic qualities of visual content, such as clarity, framing, color balance, mood, lighting, and overall style. For blind creators, aesthetic feedback from AI systems can describe subjective visual qualities that would otherwise be…

Affective Computing(also: Emotion AI, Emotional AI)

A field of AI that attempts to detect, interpret, and simulate human emotions using technologies such as facial expression analysis, voice tone detection, physiological sensors, and behavioral patterns. Affective computing raises significant accessibility and ethics concerns…

Algorithmic Bias(also: AI Bias, Machine Learning Bias)

Systematic and unfair discrimination embedded in the outputs of algorithmic systems, arising from biased training data, flawed model design, or unrepresentative development processes. For people with disabilities, algorithmic bias manifests in multiple ways: voice assistants…

Algorithmic Discrimination(also: AI Discrimination, Automated Discrimination)

The systematic disadvantaging of specific groups through the operation of AI-driven systems, whether intentional or emergent. For people with disabilities, algorithmic discrimination occurs across many domains: employment (AI hiring tools screening out disabled applicants),…

Algorithmic Harm(also: AI Harm, Algorithmic Negative Outcome)

Any difficulty, disadvantage, or injury caused by the use of AI-driven systems, ranging from mere inconvenience to material harm. For people with disabilities, documented algorithmic harms include denial of vital resources (welfare benefits, employment, housing, education),…

Algorithmic Moderation(also: Automated Moderation, AI Content Moderation)

Algorithmic moderation refers to the use of automated systems, including machine learning models and rule-based filters, to identify, flag, rank, or remove content on digital platforms without direct human review of each decision. It enables platforms to process content at…

Algorithmic accountability(also: AI accountability)

The principle that organizations and individuals responsible for creating and deploying algorithmic systems should be held responsible for the outcomes and impacts of those systems. In accessibility contexts, algorithmic accountability addresses who is responsible when…

Algorithmic bias(also: AI bias, Machine learning bias, Algorithmic discrimination)

Systematic and unfair errors in the outputs of automated decision-making systems that disadvantage particular groups of people. For people with disabilities, algorithmic bias arises from underrepresentation in training datasets, historical discrimination encoded in data, and…

Anchored Generative Model(also: Anchored Transformation Model)

A constrained AI generation approach where content creators define upper and lower bounds (anchors) of acceptable variation, and a generative model interpolates between these boundaries based on user preferences. In the context of caption customization, anchors are concrete…

Assistive AI(also: AI for Accessibility, Accessible AI, Accessibility AI)

Artificial intelligence systems designed specifically to support disabled people in performing tasks, accessing information, or navigating their environments. Examples include object recognition tools for blind users, automatic captioning for deaf users, and predictive text for…

Atomic Facts(also: Atomic Claims)

Self-contained units of information extracted from longer text, each representing a single verifiable claim or observation. In AI reliability research, decomposing model responses into atomic facts enables systematic comparison of what different models agree or disagree about.…

Audio Separation(also: Source Separation, Audio Unmixing)

The process of isolating individual audio sources from a mixed audio signal—for example, separating speech from background music, sound effects, and ambient noise. Audio separation enables selective control over different audio components, allowing users to keep speech while…

Audio-Language Model(also: ALM, Audio LLM)

A multimodal artificial intelligence model that jointly processes audio signals and natural language text, enabling it to generate detailed textual descriptions of audio content, answer questions about sounds, and reason about auditory scenes. Audio-language models like…

Auditory Scene Analysis(also: ASA, Computational Auditory Scene Analysis)

The process by which the auditory system organizes and interprets complex mixtures of sounds into distinct perceptual events or streams, allowing listeners to separate and identify individual sound sources within an environment. In accessibility contexts, auditory scene analysis…

Auto-Generated Captions(also: Automatic Captions, AI Captions, Machine-Generated Captions)

Captions automatically created by speech recognition technology without human review or editing. Video platforms like YouTube and TikTok offer auto-generated captions as a default accessibility feature. While they improve baseline accessibility, auto-generated captions often…