Glossary

Search results

Egocentric Spatial Reasoning(also: First-Person Spatial Understanding, User-Relative Spatial Reasoning)

The ability of a system to understand and describe the spatial positions of objects relative to the user's body and perspective, rather than from a bird's-eye or absolute reference frame. For AI systems assisting blind travelers, egocentric spatial reasoning is critical —…

Embodied Conversational Agent(also: ECA, Virtual Agent, Animated Agent)

A computer-generated animated character designed to interact with human users using multiple simultaneous communication channels — typically speech, eye gaze, facial expression, head and body posture, and hand gestures. ECAs are used in tutoring systems, customer-service agents,…

Emotion Recognition(also: Facial Emotion Recognition, FER, Affect Recognition)

AI technology that attempts to identify human emotional states from facial expressions, voice patterns, body language, or physiological signals. Emotion recognition systems have been widely criticized for poor accuracy, cultural bias, and particular harm to people with…

Engagement Detection(also: Engagement Monitoring, Engagement Recognition)

The use of sensors, computer vision, or other technologies to automatically assess whether a person is actively engaged with a task, device, or activity. Engagement detection systems typically monitor observable behaviours such as gaze direction, touch interaction patterns,…

Epistemic Barrier(also: Knowledge Barrier, Epistemic Divide)

A barrier to collaboration or understanding that arises from fundamental differences in knowledge systems, expertise, values, and ways of knowing between groups. In the context of sign language AI development, epistemic barriers exist between machine learning practitioners (who…

Ethics Washing(also: Ethics-Washing)

The practice of creating the illusion of high ethical standards through superficial transparency efforts, ethics committees, or principles documentation while actual practices do not reflect these stated values. In technology contexts, ethics washing may involve publishing AI…

Fabrication(also: AI Fabrication, Confabulation)

An AI error where the model generates content that does not exist in the input, such as describing objects not present in an image, inventing text that does not appear in a document, or creating details that are entirely fictional. Fabrication is distinct from misinterpretation…

Face Detection(also: Face Recognition, Facial Detection)

A computer vision technology that identifies and locates human faces within digital images or video frames, typically providing bounding box coordinates around each detected face. Face detection serves as the foundation for more advanced tasks like face recognition (identifying…

Face Recognition(also: Facial Recognition, Face Detection)

A technology that uses computer vision and machine learning to identify or verify a person by analysing their facial features from images or video. In accessibility contexts, face recognition has significant potential as an assistive tool for blind and deafblind people, enabling…

Facial Expression Recognition(also: FER, Facial Action Recognition)

Computer vision technology that detects and classifies facial expressions from images or video. In sign language contexts, facial expression recognition is essential for capturing non-manual signs — the facial movements that carry grammatical meaning in ASL, such as raised…

False Negative(also: Type II Error)

An error in which a system fails to identify something that is actually present or true. In privacy and obfuscation contexts, a false negative occurs when an AI system fails to detect private content that should be obfuscated, potentially exposing sensitive information like…

False Positive(also: Type I Error)

An error in which a system incorrectly identifies something as present or true when it is not. In privacy and obfuscation contexts, a false positive occurs when an AI system incorrectly flags non-private content as private and applies obfuscation unnecessarily, potentially…

Few-Shot Learning(also: N-Shot Learning, Low-Shot Learning)

Few-shot learning is a machine learning approach that enables AI models to learn new concepts from only a small number of examples — typically 1 to 10 — rather than the hundreds or thousands traditionally required. This is achieved through techniques like meta-learning, where…

Few-Shot Prompting(also: In-Context Learning, Few-Shot Learning)

A technique for guiding large language models by providing a small number of examples within the input prompt to demonstrate the desired task or output format. In accessibility applications, few-shot prompting can help AI systems perform context-specific tasks like correcting…

Fuzzy Logic(also: Fuzzy Inference)

Fuzzy logic is a form of many-valued logic that deals with approximate reasoning, where truth values range continuously between 0 and 1 rather than being strictly true or false. In assistive technology and signal processing, fuzzy logic systems are used to handle imprecise or…

GenAI Accessibility(also: Generative AI Accessibility)

The design and implementation of generative AI tools—including large language model chatbots, AI image describers, and multimodal AI systems—so they can be fully and equitably used by people with disabilities. While text-based GenAI interfaces appear superficially accessible,…

Gender recognition(also: Automatic gender recognition, AGR, Gender classification)

AI technology that attempts to infer a person's gender from visual features such as facial characteristics, body shape, or voice. Gender recognition systems are controversial in accessibility contexts because they typically enforce binary gender classifications, frequently…

Generative AI(also: GenAI, Generative Artificial Intelligence)

Artificial intelligence systems that can create new content — including text, images, audio, video, and code — based on patterns learned from training data. Generative AI tools like ChatGPT, DALL-E, Midjourney, and Stable Diffusion have significant implications for…

Generative Adversarial Network(also: GAN, Adversarial Network)

A type of deep learning architecture consisting of two neural networks — a generator and a discriminator — that are trained in competition with each other. The generator creates synthetic data (such as images) while the discriminator tries to distinguish between real and…

Gloss(also: Sign Gloss, Gloss Notation)

A form of transliteration used in sign language research where written words from a spoken language (typically the dominant spoken language of the region, such as English) are used as labels to represent individual signs. Glosses are written in capital letters by convention…

Grad-CAM(also: Gradient-weighted Class Activation Mapping)

A widely used explainable AI technique, introduced by Selvaraju et al. in 2017, that produces a class-discriminative heat map over an input image by weighting convolutional feature maps by the gradient of the target class score. Grad-CAM and its variants (SmoothGrad-CAM,…

Guide-by-Pointing(also: Point-and-Ask, Hand-Guided Visual Query)

A prompting technique for multimodal AI assistants where a user extends their hand into the camera's field of view and asks the AI to identify what they are pointing at, or to provide spatial directions for moving their hand toward a specific item. This technique enables blind…

Hallucination(also: AI Hallucination, Confabulation)

In the context of AI and large language models, the generation of content that is plausible-sounding but not grounded in the input data or factual reality. Hallucinations pose a particular risk in accessibility applications such as captioning, audio description, or alt text…

Head Pose Estimation(also: Head Orientation Detection, Gaze Direction Estimation)

A computer vision technique that determines the orientation or direction a person's head is facing, typically classifying whether someone is looking towards or away from the camera. In accessibility contexts, head pose estimation can help blind users determine whether a passerby…

Hidden Markov Model(also: HMM)

A statistical model used extensively in pattern recognition where the system being modeled is assumed to follow a Markov process with hidden (unobserved) states. HMMs have been foundational in both automatic speech recognition and sign language recognition, as they can model…

High-Stakes Scenarios(also: Safety-Critical Scenarios)

Situations where errors in AI-generated information could lead to significant safety, health, financial, or social consequences. In the context of visual access technology for BLV users, high-stakes scenarios include medication identification (where misreading a dosage could be…

Human Computation

A computing paradigm in which humans perform tasks that computers cannot yet do reliably, often embedded within systems that combine human and machine capabilities. The classic example is reCAPTCHA, which used human text recognition to digitise books while verifying users were…

Human-AI Alignment(also: AI Alignment, Value Alignment)

The design and training of AI systems to exhibit behaviours consistent with human values, intentions, and goals. In accessibility, human-AI alignment requires that AI systems accurately represent and respond to the diverse values and experiences of disabled and neurodivergent…

Human-AI Collaboration(also: Human-AI Teaming, AI-Assisted Authoring)

An interaction paradigm where humans and artificial intelligence systems work together, each contributing their complementary strengths to achieve outcomes neither could produce as effectively alone. In accessibility contexts, human-AI collaboration combines AI efficiency in…

Human-AI Interaction(also: HAI, Human-AI Collaboration, AI Interaction Design)

The study and design of how people interact with artificial intelligence systems, including how AI communicates its outputs, uncertainty, and limitations to users. Key principles include making AI behavior transparent, supporting user correction of errors, acknowledging…

Human-in-the-Loop(also: HITL)

An approach to AI system design and evaluation that incorporates human judgment, feedback, and oversight at critical points in automated processes. In accessibility contexts, human-in-the-loop methodologies involve people with disabilities and other affected communities in…

Human-like trust in AI(also: Anthropomorphic trust)

The phenomenon where users develop trust in AI systems based on their human-like qualities — such as natural voice, conversational style, emotional expressiveness, and social behaviors — rather than the system's actual functional reliability. In accessibility contexts, this…

Hybrid Captioning(also: AI-Augmented Captioning, Blended Captioning)

A captioning approach that combines human-generated captions with AI-powered correction or enhancement to achieve higher accuracy than either method alone. Hybrid systems leverage the reliability and contextual awareness of trained human captioners while using automatic speech…

Ideational Convergence(also: Creative Homogenization)

A phenomenon where the use of generative AI tools leads to a narrowing of creative diversity, as multiple users producing content with the same AI system tend to converge on similar outputs. In audio description, ideational convergence risks flattening the variety of descriptive…

Image Captioning(also: Automatic Image Description, AI Image Description)

A computer vision task in which an AI model generates a natural language description of the content of an image. In accessibility contexts, image captioning technology enables visually impaired users to understand visual content by converting images into text that can be read…

Image Classification(also: Visual Classification, Photo Classification)

A computer vision task where a machine learning model assigns a category label to an input image based on its visual content. Image classifiers are trained on labeled example images and learn to recognize patterns associated with each category. In accessibility applications,…

Image Description(also: Image Caption, Visual Description)

A textual representation of the content of an image, providing information about objects, people, scenes, text, colors, spatial relationships, and other visual elements. Image descriptions serve as a primary means for blind and low vision users to access visual content. They can…

Image Recognition(also: Image Classification, Computer Vision Recognition)

The use of computer vision algorithms to identify and classify objects, text, faces, scenes, and other visual content in images or video. In accessibility applications, image recognition enables tools that describe visual content to blind and low vision users, such as smartphone…

Image-to-3D Generation(also: Image-to-3D Conversion, 2D-to-3D Generation)

An AI-powered process that converts two-dimensional images into three-dimensional digital models suitable for manipulation, rendering, or physical fabrication via 3D printing. In assistive technology design workflows, image-to-3D generation serves as the second stage of an…

Inclusive AI(also: Accessible AI, Disability-Inclusive Artificial Intelligence)

The design and development of artificial intelligence systems that account for the needs, experiences, and data of people with disabilities and other marginalized groups. Inclusive AI requires representative training datasets, accessible interfaces for AI-powered tools, and…

Information Verification(also: AI Output Verification, Fact-Checking AI)

The process of assessing the accuracy, reliability, and trustworthiness of information generated by AI systems before acting on it. For blind users of generative AI, information verification is uniquely challenging because visual cross-referencing is unavailable, switching…

Instance Segmentation

A computer vision technique that identifies and delineates individual objects within an image at the pixel level, distinguishing separate instances even when they belong to the same category. In accessibility contexts, instance segmentation enables assistive tools to provide…

Intelligent Tutoring System(also: ITS, AI Tutor)

An AI-powered educational system that provides personalized instruction, feedback, and scaffolding adapted to individual learners' needs, knowledge levels, and learning patterns. Modern intelligent tutoring systems increasingly use generative AI and large language models to…

Intelligent Virtual Assistant(also: IVA, Virtual Assistant, AI Assistant)

A software-based agent that uses artificial intelligence, natural language processing, and speech recognition to understand and respond to human voice or text commands. Intelligent Virtual Assistants such as Amazon Alexa, Google Assistant, and Apple Siri are embedded in smart…

Just-in-Time Prompting(also: Trigger-Based Prompting, On-Demand Visual Check)

A prompting technique for voice and video-capable AI models where the user pre-configures the AI with a role and task description, then uses a trigger phrase (such as "check now") to initiate an on-demand visual analysis of the current camera view. Developed as a workaround for…

LIME(also: Local Interpretable Model-agnostic Explanations)

An explainable AI technique, introduced by Ribeiro et al. in 2016, that approximates any black-box model's behaviour around a single prediction by fitting a simple interpretable model (usually sparse linear regression) to perturbed versions of the input. The resulting feature…

LLM Agent(also: Generative Agent, Task-Executing Agent)

A software system built around a large language model that autonomously perceives state, plans actions, executes them against an environment (a web page, a mobile app, a shell, a UI), and reflects on outcomes to make progress toward a goal. In accessibility work, LLM agents are…

LLM Disability Bias(also: AI Ableism, Language Model Disability Bias)

Systematic prejudice against people with disabilities embedded in large language models due to biased training data and development processes that underrepresent disabled communities. Research has documented multiple forms of this bias: pretrained language models associate…

LLM Hallucination(also: AI Hallucination, Model Hallucination)

The phenomenon where large language models generate content that is factually incorrect, fabricated, or not grounded in the input provided, while presenting it with apparent confidence. In bias research, hallucinations are particularly concerning because models may invent…

LLM-Based Auto Correction(also: AI-Powered Text Correction, LLM Autocorrect)

The use of large language models to automatically detect and correct common text errors without requiring manual user intervention. In accessibility contexts, LLM-based auto correction can reduce the editing burden for users with disabilities by fixing predictable errors…