Terminology

CAHDD-Related Terms — Initial Master List

Algorithmic Assistance
Technology that uses defined computational rules, scripts, or algorithms to assist a human creator without generative AI creating expressive content. Closely associated with CAHDD Stage 2.

Artificial Intelligence (AI)
Computer systems capable of performing tasks associated with aspects of human intelligence. Within CAHDD, the important distinction is not simply whether AI was present, but what role it played in the creative process.

Authorship
The human relationship to a work involving creative intent, judgment, direction, decision-making, and responsibility. Authorship is one of CAHDD’s central concepts.

Automation
Technology performing a task or sequence of tasks with reduced human intervention. Automation is not necessarily artificial intelligence.

CAHD™ — Computer Aided Human Designed
The original concept from which CAHDD developed. CAHD describes work in which computers or technology assist while human design, intent, judgment, and authorship remain central.

CAHDD™ — Computer Aided Human Designed & Developed
A human-centered framework for visibly communicating the relationship between human creative contribution and technological assistance.

CAHDD Disclosure
A visible indication of how technology participated in the creation or development of a work, commonly communicated through a CAHDD Stage, symbol, statement, mark, or combination of methods.

CAHDD Stage
One of the defined CAHDD classifications used to describe the relationship between human creation and technological assistance in a particular work or workflow.

Computer Aided
The use of computers or digital technology to assist a human activity. Within CAHD and CAHDD, aided is important: technology may support the process without necessarily becoming the author or originator of the work.

Content Credentials
Digital provenance information that can accompany a file and provide information about its origin and editing history. CAHDD views embedded provenance and visible human-readable disclosure as complementary rather than mutually exclusive.

Creative Agency
The meaningful ability of a human creator to originate, direct, evaluate, modify, select, and make decisions about creative work.

Creative Intent
The purpose, idea, direction, or desired outcome established by a human creator. Creative intent helps distinguish human authorship from simply initiating an automated process.

Creator
A human individual, group, studio, or organization responsible for meaningful creative intent, direction, judgment, and the work being claimed.

Creator Choice
The CAHDD principle that creators should retain a meaningful voice in how their creative process and use of technology are visibly communicated.

Creator’s Mark™
A visible human claim of authorship and responsibility. It carries forward the long tradition of creators signing or marking their work to communicate: I made this. I stand behind it.

Digital Provenance
Information used to document the origin, history, or modifications of digital content. Provenance can help establish where content came from, but does not necessarily establish human authorship by itself.

Digitally Assisted Creation
Creation performed by a human using digital tools under direct human control without those tools generating expressive creative content or making substantive creative decisions. CAHDD Stage 1.

Disclosure
The act of communicating meaningful information about how a work was created, including the role of digital tools, automation, algorithms, or artificial intelligence.

Expressive Content
Creative material that contributes meaningfully to the perceptible work itself, such as imagery, writing, music, voice, video, design elements, or other creative expression.

Fully Autonomous AI Creation
A workflow in which AI produces all or almost all expressive content with little or no meaningful human creative contribution beyond configuration, initiation, triggering, observation, or distribution. CAHDD Stage 5.

Generative AI
Artificial intelligence capable of generating new content such as text, images, music, audio, video, code, or other material. Within CAHDD, generative AI is distinguished from conventional digital tools, algorithms, and automation.

Human Agency
A person’s ability to make meaningful choices, exercise judgment, direct outcomes, and retain control over decisions affecting their work.

Human Authorship
Authorship attributable to meaningful human creative activity, including intent, judgment, direction, decision-making, and responsibility.

Human-Centered
An approach that places human needs, values, abilities, agency, and well-being at the center of technological development and use.

Human–AI Hybrid Creation
Human-directed creation in which AI assists with research, writing, grammar, refinement, enhancement, exploration, or other creative or support tasks while meaningful human authorship and direction remain central. CAHDD Stage 3.

Human + Machine Integrity
The CAHDD principle that neither human contribution nor technological contribution should be intentionally misrepresented.

Humanocentricus™
CAHDD’s broader human-centered philosophy that technology should develop and operate around human needs, values, abilities, agency, and well-being rather than requiring humanity to reorganize itself around technology.

Human-Readable Disclosure
Information about a creative process presented so ordinary people can understand it directly, rather than requiring access to metadata, specialized software, or technical provenance systems.

Human Responsibility
The principle that responsibility for claimed creative work ultimately belongs to the person or organization standing behind it, rather than to the tools used to produce it.

Meaningful Human Contribution
Human participation that materially influences a work through activities such as conception, creation, direction, judgment, editing, refinement, selection, or decision-making rather than merely initiating an automated process.

Metadata
Information stored with or about a digital file. CAHDD distinguishes metadata-based information from visible disclosure that audiences can directly see and understand.

Procedural Assistance
Rule-based or parameter-driven technology used to assist creation, such as procedural modeling, parametric systems, simulations, particle systems, or procedural textures. Procedural assistance does not inherently constitute AI.

Process Transparency
Providing meaningful information about the methods and technologies involved in creating a work.

Provenance
Information describing the origin and history of something. In digital media, provenance can help document where a file originated and what happened to it, but provenance and authorship are not necessarily the same thing.

Stage 0 — Traditional Human Craft
Entirely human-created through physical or analog methods, without digital tools, automation, or AI.

Stage 1 — Digitally Assisted Creation
Digital tools are used under direct human control and do not generate creative content or make substantive creative decisions.

Stage 2 — Algorithmic / Procedural Assistance
Scripts, filters, simulations, parametric systems, procedural systems, and task-level automation assist execution without generative AI creating expressive content.

Stage 3 — Human–AI Hybrid Creation
Human-directed creation in which AI assists with research, writing, grammar, refinement, enhancement, or other creative or support tasks while meaningful human authorship and direction remain central.

Stage 4 — AI-Generated, Human-Curated
AI generates expressive content while a human directs, iterates, selects, evaluates, curates, and approves the result.

Stage 5 — Fully Autonomous AI Creation
AI produces all or almost all expressive content with little or no meaningful human creative contribution beyond configuration, initiation, triggering, observation, or distribution.

Stage X — Hybrid or Multi-Phase Workflow
A complex workflow combining substantially different processes, stages, contributors, or phases that cannot be accurately represented by a single CAHDD Stage.

Stewardship
The responsibility to shape technological adoption thoughtfully, considering human agency, consequences, opportunities, and the future relationship between people and technology.

TechRatio™
A CAHDD-related concept for describing the relationship between human contribution and technological assistance in greater detail when a single CAHDD Stage does not tell enough of the creative story.

Technological Assistance
The use of digital tools, algorithms, automation, AI, or other technologies to support some portion of a human creative or developmental process.

Traditional Human Craft
Work created entirely through human physical or analog methods without digital tools, automation, or artificial intelligence. CAHDD Stage 0.

Transparency
The principle of communicating meaningful information about how a work was created and the respective roles of human creators and technology.

Transparency Without Judgment
The CAHDD principle that disclosure should provide useful information rather than establish a hierarchy in which one creative process is automatically considered better or worse than another.

U.N.O. — Unless Noted Otherwise
A CAHDD disclosure convention allowing a creator or organization to establish a default CAHDD Stage for a body of work while identifying exceptions separately.

Visible Authorship
A clear, human-readable indication of the person, group, studio, or organization claiming authorship and responsibility for a work.

Visible Disclosure
Creative-process information intentionally presented where people can see and understand it, rather than existing exclusively as embedded technical data.

Workflow
The sequence of methods, tools, decisions, and processes used to create or develop a work. A workflow may remain within one CAHDD Stage or cross multiple stages.

General Technology & AI Terms — Master Working List

• 3D Model — A digital representation of an object, space, character, or environment created using three-dimensional geometry.
• Agentic AI — AI designed to pursue goals through multiple steps, often deciding what actions or tools to use along the way rather than responding only once to a prompt.
• AI Agent — An AI-based system capable of taking actions, using tools, gathering information, and carrying out multiple steps toward a goal with varying degrees of human supervision.
• AI Assistant — An AI system primarily designed to help a person perform tasks, answer questions, generate material, or work with information under human direction.
• AI Detector — Software that attempts to estimate whether text, imagery, audio, or other content was created by AI. Such systems generally provide probabilistic judgments rather than definitive proof.
• AI-Generated — Content substantially produced by an artificial intelligence system rather than directly created by a human using conventional tools.
• AI-Assisted — Work in which artificial intelligence helps a human perform some portion of a task while meaningful human participation remains part of the process.
• AI Model — A computational system trained or constructed to recognize patterns, make predictions, generate content, or perform other tasks.
• AI Safety — The field concerned with reducing unintended, harmful, unreliable, or dangerous behavior from artificial intelligence systems.
• AI Slop — An informal and usually negative term for large amounts of low-quality, repetitive, misleading, or minimally curated AI-generated content.
• AI Washing — Marketing something as being powered by or based on artificial intelligence when its actual AI component is minor, conventional, exaggerated, or sometimes nonexistent.
• Algorithm — A defined set of instructions or rules used to solve a problem or perform a task. Algorithms existed long before modern AI, and an algorithm is not automatically artificial intelligence.
• Algorithmic — Performed, controlled, or determined according to an algorithm or defined computational rules.
• Alignment — Efforts to make an AI system behave in ways consistent with intended human goals, instructions, preferences, or safety requirements.
• Anthropomorphism — Attributing human thoughts, emotions, motives, understanding, or personality to something nonhuman, including an AI system.
• Application Programming Interface (API) — A defined way for one software system to communicate with or use features from another software system.
• Artificial General Intelligence (AGI) — A proposed form of AI capable of performing or learning across a broad range of intellectual tasks rather than being limited to particular functions. There is no universally agreed threshold for when AGI would have been achieved.
• Artificial Intelligence (AI) — A broad term for machine-based systems capable of performing tasks associated with abilities such as learning, perception, prediction, reasoning, language, planning, or decision-making. AI is a larger category than generative AI.
• Artificial Neural Network — A computational system made from interconnected processing units inspired loosely by the organization of biological neurons. Neural networks form the basis of much modern machine learning.
• Artificial Superintelligence (ASI) — A hypothetical artificial intelligence whose capabilities greatly exceed human capabilities across most or all important intellectual domains.
• Attention — A mechanism used in many neural networks that helps a model determine which parts of its input are most relevant when processing or generating information.
• Audio Generation — The use of generative technology to create sounds, music, speech, effects, or other audio.
• Augmented Reality (AR) — Technology that overlays digital objects or information onto a view of the physical world.
• Augmentation — Using technology to extend or enhance human abilities rather than completely replacing the human performing the task.
• Autonomous — Able to perform actions or make certain decisions with little or no immediate human direction.
• Autoregressive Model — A model that generates a sequence incrementally by predicting what should come next based on what has already been produced.
• Automation — Technology performing tasks or sequences of tasks with reduced human intervention. Automation can be completely rule-based and does not necessarily involve AI.
• Automation Bias — The human tendency to trust or defer to the output of an automated system even when there are reasons to question it.
• Avatar — A digital representation of a person, character, or identity, ranging from a simple icon to a realistic animated human likeness.
• Benchmark — A standardized test or dataset used to compare the performance of AI models or other technologies.
• Bias — A systematic tendency that can influence the results of a system. AI bias may arise from training data, model design, evaluation methods, deployment conditions, or human assumptions.
• Binary — A system based on two states, usually represented as 0 and 1, forming the underlying representation used by digital computers.
• Black Box — A system whose internal reasoning or operations are difficult for users, and sometimes even its developers, to interpret or explain.
• Bot — Software designed to perform automated tasks. A bot may or may not use artificial intelligence.
• C2PA — Coalition for Content Provenance and Authenticity — The organization and technical standard behind a system for attaching verifiable provenance information to digital content.
• CAD — Computer-Aided Design — Software used by people to create, modify, analyze, or document designs digitally.
• Chatbot — Software designed to communicate conversationally through text or voice. Modern chatbots frequently use large language models, but chatbots existed long before current generative AI.
• Checkpoint — A saved version of a machine-learning model at a particular point in its development or training.
• Classification — Assigning data to categories, such as identifying an image as containing a dog, determining whether email is spam, or classifying sentiment as positive or negative.
• Cloud Computing — Computing services provided through remote servers accessed over a network rather than performed entirely on a user’s local computer.
• Computer-Generated Imagery (CGI) — Imagery created using computers, particularly 2D or 3D graphics. CGI is not inherently AI-generated.
• Computer Vision — Technology that allows computers to analyze and interpret visual information such as photographs, video, objects, faces, or scenes.
• Conditional Generation — Generating content according to supplied information or constraints, such as a text prompt, reference image, sketch, depth map, or other input.
• Content Authenticity — The ability to establish reliable information about digital content and its history. Authenticity does not necessarily mean that the content itself is factually true.
• Content Credentials — The preferred public-facing term used by C2PA for provenance information cryptographically associated with digital content. It can describe information about an asset’s origin and modifications but should not be confused with proof of human authorship.
• Context — Information available to an AI system that helps it interpret a request or determine an appropriate response.
• Context Window — The amount of information a language or multimodal model can consider at one time during an interaction.
• Copyright — A legal form of protection for certain original creative works. Copyright law and eligibility can differ substantially depending on jurisdiction and the extent of human authorship.
• Corpus — A collection of text, images, audio, or other material assembled for analysis, training, research, or language processing.
• CPU — Central Processing Unit — The main general-purpose processor in a computer.
• Creative Automation — Automation applied to creative production tasks such as layout generation, resizing, versioning, animation, editing, or media production. Creative automation does not necessarily involve generative AI.
• DAW — Digital Audio Workstation — Software used to record, edit, mix, arrange, and produce music or other audio.
• Data — Information represented in a form that computers can store, process, analyze, or transmit.
• Data Augmentation — Creating modified variations of existing training examples to increase the quantity or diversity of data available for machine learning.
• Data Labeling — Adding categories, descriptions, annotations, or other identifying information to data so it can be used for training or evaluating machine-learning systems.
• Dataset — An organized collection of data used for analysis, training, testing, evaluation, or other purposes.
• Deep Learning — A branch of machine learning that uses neural networks with many layers to learn complex patterns from data.
• Deepfake — Synthetic or manipulated media designed to realistically depict a person saying, doing, or appearing as something that did not actually occur.
• Deterministic — Producing the same result when given the same inputs and conditions.
• Diffusion Model — A type of generative model commonly used for image, video, and other media generation. It learns to construct content through a process related to removing noise from data.
• Digital Asset — A digitally stored item such as an image, video, recording, document, 3D model, or other file.
• Digital Signature — Cryptographic information used to verify that digital information came from a particular signing source and has not been altered in certain ways after signing.
• Digital Twin — A digital representation of a physical object, environment, process, or system that may incorporate real-world data and simulation.
• Digital Watermark — Information embedded within digital content to help identify its source, ownership, status, or history. Watermarks may be visible or invisible.
• Distillation — A technique for training a smaller or more efficient AI model to reproduce useful behavior from a larger model.
• DPO — Direct Preference Optimization — A technique used to tune AI models based on preferred versus less-preferred responses without using the same reinforcement-learning process employed by methods such as RLHF.
• Embedding — A numerical representation that places information such as words, images, sounds, or concepts into a mathematical space where related items can be located near one another.
• Emergent Behavior — Capabilities or behaviors that appear in a complex system without having been explicitly programmed as individual rules.
• Encoder — A component that converts input information into an internal representation that another part of a model can process.
• Encryption — Converting information into a protected form so unauthorized people or systems cannot easily read it.
• EXIF Data — Metadata commonly stored in image files that can include camera information, dates, exposure settings, location information, and other details.
• Expert System — An older form of AI that uses programmed knowledge and rules to make decisions or provide recommendations within a specialized domain.
• Explainable AI (XAI) — Methods intended to make an AI system’s outputs, reasoning factors, or behavior more understandable to humans.
• Face Recognition — Technology that analyzes facial characteristics to identify or verify individuals.
• Face Swap — Digitally replacing one person’s face with another in an image or video, using conventional visual-effects methods, AI, or a combination.
• Feature — A measurable property or characteristic used by a machine-learning system when analyzing data.
• Fine-Tuning — Additional training performed on an already-trained model to adapt it to particular tasks, domains, styles, behaviors, or datasets.
• Foundation Model — A broadly trained AI model that can be adapted or used for many different downstream tasks.
• Frontier Model — A loosely defined term generally referring to highly capable, state-of-the-art AI models near the leading edge of current development.
• GAN — Generative Adversarial Network — A type of generative neural network in which one network produces content while another evaluates it, allowing the generator to improve. GANs became particularly important in early modern synthetic-image generation.
• Generative AI (GenAI) — AI designed to generate new synthetic content such as text, images, video, audio, code, or other digital material based on patterns learned from data.
• Generative Fill — A tool that uses generative AI to create, replace, remove, or extend portions of an image based on surrounding content and/or instructions.
• Generative Model — A model designed to produce new data or content resembling patterns found in the material from which it learned.
• GPU — Graphics Processing Unit — A processor originally developed primarily for graphics but now widely used for AI because it can perform many calculations simultaneously.
• Grounding — Connecting an AI system’s response to supplied information, documents, databases, tools, or real-world sources rather than relying exclusively on information encoded in the model.
• Guardrails — Technical or policy mechanisms intended to limit undesirable, unsafe, unauthorized, or inappropriate behavior by an AI system.
• Hallucination — An AI-generated statement or output that appears plausible but is unsupported, incorrect, invented, or inconsistent with available evidence.
• Hash — A compact digital value calculated from data that can be used to detect changes or identify matching information.
• Human-in-the-Loop (HITL) — A process in which humans remain involved in reviewing, correcting, approving, directing, or otherwise influencing decisions made by automated or AI systems.
• Image Generation — The creation of new images using generative algorithms or AI models.
• Image-to-Image — A generative process that uses an existing image as input and produces a modified or transformed image.
• Inference — The stage at which a trained AI model uses what it has learned to produce a prediction, classification, response, or generated result.
• Inpainting — Generating or reconstructing content inside a selected portion of an existing image, video frame, or other media.
• Input — Information supplied to a computer system for processing. In generative AI this may include text, images, sound, video, documents, or other data.
• Instruction Tuning — Additional model training intended to improve how well an AI follows human-written instructions.
• Intellectual Property (IP) — A broad category of legally recognized rights associated with creations, inventions, brands, designs, confidential information, and other intangible assets.
• Jailbreak — An attempt to make an AI system ignore or circumvent restrictions, safeguards, policies, or instructions that were intended to constrain its behavior.
• Knowledge Cutoff — The approximate point after which information was not included in a model’s original training. Connected search, retrieval, and tools can sometimes provide information newer than the model’s training data.
• Knowledge Graph — A structured representation of entities and the relationships between them.
• Large Language Model (LLM) — A large AI model trained primarily to recognize and generate patterns in language, usually by predicting sequences of tokens.
• Latent Space — An internal mathematical representation in which a generative model encodes patterns, characteristics, or relationships learned from data.
• LoRA — Low-Rank Adaptation — An efficient method for adapting a larger AI model by training relatively small additional components rather than retraining the entire model.
• Loss Function — A mathematical measurement used during machine learning to indicate how far a model’s output is from the desired result.
• Machine Learning (ML) — A branch of AI in which computer systems learn patterns or relationships from data rather than relying solely on explicitly programmed rules.
• Manifest — A structured record containing information about another digital object. In C2PA, a manifest contains provenance-related information and digitally signed assertions associated with an asset.
• Metadata — Information describing other data or content, such as author, date, camera settings, file type, location, editing history, or software used.
• MIDI — Musical Instrument Digital Interface — A digital standard used to communicate musical information such as notes, timing, velocity, and instrument controls. MIDI contains performance instructions rather than recorded sound itself.
• Model — A computational representation trained or constructed to perform a task such as prediction, classification, generation, recognition, or decision-making.
• Model Collapse — Degradation that can occur when AI models are repeatedly trained on low-quality or recursively generated synthetic data, causing diversity or accuracy to deteriorate.
• Model Drift — A decline or change in model performance as real-world data, conditions, or behavior shift from what the model encountered during development.
• Model Parameters — Adjustable numerical values learned during model training that influence how the model processes inputs and produces outputs.
• Model Weights — The learned numerical values inside a neural network. The terms weights and parameters are often used loosely in similar ways.
• Mixture of Experts (MoE) — A model architecture containing multiple specialized internal components, with only some activated for a particular input or task.
• Multimodal AI — AI capable of working with more than one form of information, such as text, images, audio, video, or other data.
• Natural Language Processing (NLP) — Technology concerned with allowing computers to analyze, understand, generate, or otherwise work with human language.
• Neural Network — A machine-learning structure made from interconnected computational units that learn relationships and patterns from data.
• NPU — Neural Processing Unit — A processor designed specifically to accelerate neural-network and AI workloads, increasingly included in personal computers and mobile devices.
• OCR — Optical Character Recognition — Technology that converts text visible in images, scans, or photographs into machine-readable text.
• Open Source — Software whose source code is made available under a license allowing specified forms of inspection, modification, and redistribution.
• Open-Weight Model — An AI model whose trained model weights are made available for others to download or use. Open weights do not necessarily mean that the training data, source code, or entire development process is open.
• Output — The result produced by a computational or AI system after processing an input.
• Outpainting — Generating additional imagery beyond the original boundaries of an existing image.
• Overfitting — When a machine-learning model learns its training data too specifically and performs poorly on unfamiliar data.
• Parameter — A numerical value within a model that affects its behavior. Modern AI models may contain billions or more learned parameters.
• Parametric Design — A design method in which relationships, dimensions, geometry, or other properties are controlled through parameters and rules. Parametric design is not inherently AI.
• Perceptual Hash — A digital fingerprint designed so visually or perceptually similar media produces related values, allowing systems to identify modified versions of content.
• Photogrammetry — Creating measurements, maps, or 3D models from multiple photographs of real-world objects or environments.
• Predictive AI — AI primarily used to estimate classifications, probabilities, future outcomes, recommendations, or other predictions rather than generate expressive content.
• Pretraining — The initial large-scale training process through which a model learns general patterns before possible additional specialization or fine-tuning.
• Procedural — Produced or controlled according to defined rules, parameters, or computational procedures rather than manually constructing every element.
• Procedural Generation — Creating content automatically from rules or algorithms, such as terrains, textures, vegetation, building layouts, game levels, or patterns. Procedural generation is not necessarily AI.
• Procedural Texture — A texture generated mathematically or algorithmically rather than being based entirely on a stored image.
• Prompt — An instruction, request, example, image, or other input supplied to a generative AI system to influence its response.
• Prompt Engineering — The practice of designing and structuring prompts to obtain more useful, predictable, or controlled results from AI systems.
• Prompt Injection — Instructions embedded in data or content that attempt to manipulate an AI system into overriding its intended instructions or behavior.
• Prompt Weighting — Giving portions of a prompt greater or lesser influence over a generative result.
• Provenance — Information about the origin and history of an asset, including how it was created or modified. C2PA specifically describes provenance as understanding the history of an asset and its interactions with actors and other assets.
• Quantization — Reducing the numerical precision used by an AI model so it requires less memory or computing power, sometimes with a modest change in performance.
• Raster Image — A digital image made from a grid of pixels, such as most photographs.
• Ray Tracing — A rendering method that simulates the paths of light rays to create realistic lighting, reflections, shadows, and other optical effects.
• Real-Time Rendering — Generating visual imagery quickly enough to respond interactively to user input or scene changes.
• Reasoning Model — A broad product and research term for AI models optimized to perform more involved multi-step problem solving before producing an answer.
• Recommendation System — Software that predicts which content, products, music, videos, people, or other items a user may find relevant.
• Reinforcement Learning (RL) — A machine-learning approach in which a system learns behavior through feedback associated with actions and outcomes.
• Reinforcement Learning from Human Feedback (RLHF) — A method for improving AI behavior using human judgments or preferences to help guide model training.
• Rendering — The process of calculating and producing an image, animation, or other visual output from digital scene information. Rendering itself is not artificial intelligence.
• Retrieval-Augmented Generation (RAG) — A method that allows a generative AI model to retrieve relevant information from external documents or databases and use it when producing an answer.
• Robot — A physical machine capable of performing actions in the real world. Robots may use AI, conventional automation, remote control, or combinations of these systems.
• Robotics — The field concerned with designing, constructing, controlling, and using robots.
• Sampler — In generative image and diffusion systems, a method that influences how a model moves from noise or an initial state toward a generated result.
• Sampling — Selecting among possible model outputs according to probabilities rather than always choosing the single most likely result.
• Seed — A numerical starting value used by many generative systems to initialize random processes. Reusing a seed can sometimes help reproduce or vary a result.
• Semantic Search — Searching based on meaning and conceptual similarity rather than relying only on exact matching words.
• Sentiment Analysis — Software analysis intended to determine whether language expresses attitudes such as positive, negative, or neutral sentiment.
• Simulation — A computational model of a real or hypothetical system used to study behavior, test possibilities, or predict outcomes. Simulation does not inherently involve AI.
• Speech Recognition — Technology that converts spoken language into text or interpretable commands.
• Speech Synthesis — Technology that produces artificial spoken voice from text or other data.
• Stochastic — Involving probability or randomness, meaning the same input may not always produce exactly the same result.
• Style Transfer — Applying visual or other stylistic characteristics from one source or learned representation to different content.
• Supervised Learning — Machine learning performed using training data containing known labels or desired outputs.
• Synthetic Data — Artificially created data used instead of or alongside data collected from the real world.
• Synthetic Media — Images, video, audio, text, or other media generated or substantially altered using computational techniques, particularly generative AI.
• System Prompt — Instructions supplied to an AI model by the system or application operating it that help establish behavior, priorities, context, or restrictions before the user’s own prompt is processed.
• Temperature — A setting in many generative models that influences the variability of outputs. Higher values generally allow more variation, while lower values tend toward more predictable output.
• Text-to-3D — Generating 3D geometry or representations from written instructions.
• Text-to-Audio — Generating sounds, music, speech, or other audio from written instructions.
• Text-to-Image — Generating an image from a written description or prompt.
• Text-to-Speech (TTS) — Converting written text into synthesized spoken audio.
• Text-to-Video — Generating moving imagery or video from written instructions.
• Token — A unit into which AI language systems divide text for processing. A token may represent an entire word, part of a word, punctuation, or another text fragment.
• Tokenization — Dividing text or other information into smaller units that a model can process.
• Tool Use — The ability of an AI system to interact with external tools such as calculators, search systems, software, databases, APIs, browsers, or other applications.
• Training — The process through which a machine-learning model adjusts its internal parameters by learning patterns from data.
• Training Data — Data used during the training of a machine-learning model.
• Transformer — A neural-network architecture built around attention mechanisms that became the foundation of most modern large language models and many multimodal AI systems.
• Turing Test — A test proposed by Alan Turing in which a machine’s conversational behavior is evaluated according to whether a human can reliably distinguish it from another human. Passing such a test does not establish consciousness or human-like understanding.
• Uncanny Valley — The discomfort some people experience when an artificial human representation appears very realistic but still has noticeable nonhuman characteristics.
• Unstructured Data — Information without a fixed tabular structure, such as documents, photographs, recordings, video, and natural-language text.
• Unsupervised Learning — Machine learning in which the system attempts to discover patterns or structures in data without being supplied predetermined labels for every example.
• Upscaling — Increasing the resolution or apparent detail of an image or video. Traditional upscaling uses mathematical interpolation; AI upscaling may infer or generate additional detail.
• VAE — Variational Autoencoder — A type of neural network that learns compressed representations of data and can generate or reconstruct content from those representations.
• Vector Database — A database designed to store and search numerical embeddings, allowing information to be retrieved according to similarity or meaning.
• Vector Embedding — A numerical representation of content used to compare relationships and similarity between items such as text, images, or sounds.
• Vector Graphic — Digital artwork represented through mathematical shapes, lines, curves, and coordinates rather than a fixed grid of pixels.
• Virtual Reality (VR) — Technology that immerses a user in a computer-generated environment, typically using a headset.
• Vision-Language Model (VLM) — An AI model capable of processing both visual information and language, allowing it to describe, analyze, answer questions about, or otherwise work with images.
• Voice Clone — A synthetic voice designed to reproduce the recognizable vocal characteristics of a particular person.
• Voice Synthesis — Artificial generation of human-like speech, whether based on a generic synthetic voice or a particular person’s voice characteristics.
• Watermark — A visible or hidden mark associated with content to convey information such as ownership, origin, authenticity, or status.
• Weights — Learned numerical values within a neural network that influence how strongly information affects subsequent calculations and outputs.
• Workflow — The sequence of tools, actions, decisions, and processes used to complete a task or create something.
• Zero-Shot Learning / Zero-Shot Prompting — Asking a model to perform a task without first providing specific examples of how the task should be done.