Artificial Intelligence: Glossary

AI conversations run on a lot of specialized language. Use this glossary to look up a term you have run into, or to build a working vocabulary before you need one. Definitions follow how the terms are used in AI products and UX work; vendors and researchers sometimes use them differently. Many entries link to NN/G articles that go deeper; for a guided tour of the topic, start with our study guide on how AI works and how users think about it. Look up the term you need in the table below, search this page with your browser’s Find command, or read straight through. Terms run alphabetically down each column. Agent A system that can use a model and tools to carry out multiple steps toward a goal, evaluate its progress, and decide what to do next. The product around it controls its tools, permissions, stopping rules, and opportunities for human review. - See also: agentic AI, tool use, model - Related article: A Concrete Definition of an AI Agent Agentic AI A description of how much latitude a system has between prompts. An agentic system can choose several steps and use tools toward a goal without being told each move. Autonomy is a matter of degree, so calling a product an agent says little by itself about how much autonomy it actually has. - See also: agent, tool use - Related article: A Concrete Definition of an AI Agent AI Engineering The work of turning AI models into reliable products: supplying context, designing prompts, connecting tools, evaluating behavior, deploying, monitoring, and adding guardrails. Depending on the team, it may also extend to preparing data and adapting models. - See also: context, evals, prompt engineering, guardrails AI Slop Low-quality, mass-produced AI-generated content: text, images, or code that is plausible on the surface but careless, generic, or unverified. Guarding against it with quality bars and human review is a design and governance concern. - See also: hallucination, sycophancy AI Washing Making false, misleading, or exaggerated claims about whether a product uses AI, how it uses it, or what it can do. It can erode user trust when the capability does not match the claim. - See also: alignment, slop - Related article: “Powered By AI” Is Not a Value Proposition Algorithm A step-by-step procedure or set of rules for solving a problem or performing a task. In AI, algorithms train models, search for solutions, make predictions, retrieve information, and choose actions. - See also: machine learning, model Alignment The effort to make AI systems behave in accordance with human values, intentions, and ethical guidelines. A central design challenge is making systems that do what users want while avoiding harmful or unintended outcomes. - See also: guardrails, system prompt, RLHF - Related article: How AI Models Are Trained Anthropomorphism Attributing human qualities such as feeling, intent, or humanlike understanding to an AI system because of the way it communicates. It can make a product feel approachable, and it can just as easily lead users to trust the system further than its reliability supports. First-person language, apologies, and other humanlike cues may strengthen the effect, often called the “ELIZA effect” after a 1966 chatbot. - See also: overreliance, automation bias, sycophancy - Related article: The 4 Degrees of Anthropomorphism of Generative AI Artificial General Intelligence (AGI) A hypothetical AI with broad, flexible capability across many intellectual tasks. No current system is generally accepted as AGI, and there is no agreed definition or test for it. Today’s general-purpose models handle many tasks, but unevenly. - See also: AI Artificial Intelligence (AI) Broadly, any computer system designed to perform tasks that typically require human intelligence, such as learning from examples, recognizing patterns, understanding language, or making decisions. Also, the subfield of computer science concerned with researching and developing tools that enable computers to act “intelligently.” - See also: machine learning, model - Related article: How AI Works and How Users Think About It: Study Guide Automation Bias The tendency to favor an automated system’s output and to scrutinize it less than you would a person’s, even when it is wrong. Automation bias is one form of overreliance; overreliance is broader and can arise without automation-specific bias. Introducing friction at high-stakes moments and providing easy ways for users to verify or override a result help prevent automation bias. - See also: grounding, sycophancy, overreliance, human-in-the-loop - Related article: AI Chatbots Discourage Error Checking Bias Systematic skew in data, models, outputs, organizational processes, or human use, particularly skew that unfairly disadvantages certain people, groups, or viewpoints. Bias can enter at any stage of an AI system’s lifecycle. Testing can reveal it, though no test establishes that a system is free of it; addressing it may require changes to the data, the model, the product, or the process around them. - See also: training data, dataset, alignment, evals - Related article: How AI Models Are Trained Chain-of-Thought Prompting A prompting technique that uses an instruction or worked examples to prompt a sequence of intermediate reasoning steps before a final answer. It can improve performance on many multistep tasks, but the generated steps are not guaranteed to faithfully explain how the model arrived at its answer. - See also: prompt, reasoning model, inference Compute The amount of computation required to complete a task, often measured in operations such as floating-point operations (FLOPs). Context Everything supplied or made available to a model for a single request: the prompt, the system prompt, selected conversation history, retrieved documents, stored memory, and the results of any tools it has called. In most current assistant products, the surrounding application or service stores conversational state and supplies or references it when relevant. Context is what is available for one request; memory is retained information that may later become part of that context. Choosing deliberately what goes into the context is sometimes called context engineering. - See also: context window, prompt, system prompt, RAG, memory - Related article: Context Architecture Context Window The maximum amount of tokenized information a model can take into account at once. It is usually not just the user’s text: conversation history, retrieved documents, tool results, images, and some or all of the generated response are typically counted against it. A large window lets a model reference more, though not everything in a long window influences the answer equally. Dataset A collection of examples used to train or evaluate an AI model. Datasets are the foundation of machine learning; model quality depends heavily on the quality, size, and representativeness of its data. - See also: training data, synthetic data - Related article: How AI Models Are Trained Deep Learning An approach to machine learning that uses networks with many computational layers, and is often used for learning complex tasks. Deep learning underlies most current generative language models and many modern computer- vision systems. - See also: neural networks, machine learning Embedding A numerical vector representation of text, images, or other data that captures semantic meaning in a form that machines can process. Semantic embeddings often place related items, such as “king” and “queen,” near one another in a multidimensional space. - See also: neural networks, RAG - Related article: How Do Generative AI Systems Work? Evals (Evaluations) Repeatable tests that measure whether an AI feature does its job. An eval pairs representative inputs with explicit scoring criteria, which might be reference answers, a rubric, a human-preference judgment, or an automated metric. Evals are how a team tells an improvement from a regression. - See also: bias, hallucination, overfitting, training data - Related article: Demand Accuracy in Your AI Tools Explainability How well a system can give a person a meaningful and accurate account of how or why it produced a result. A generated explanation may not accurately reflect the system’s actual process. - See also: grounding, reasoning model, automation bias - Related article: Crafting AI Explanations for Every Role in Your Enterprise Finetuning Continuing the training of an already trained model on additional data to adapt it for a task, domain, or style. It usually takes less compute than pretraining a comparable model from scratch, and it has to be handled carefully to avoid degrading other capabilities. - See also: compute, training data, RLHF, foundation model - Related article: How AI Models Are Trained Foundation Model A model trained on broad data at scale that can be adapted to many uses. Foundation models are often large. LLMs are one kind: the same base model can be prompted, finetuned, or extended for many different uses. - See also: model, LLM, finetuning - Related article: How AI Models Are Trained Generative AI (GenAI) AI systems that generate new content (text, images, code, audio) from patterns learned during training, rather than only retrieving or classifying what already exists. - See also: language model, multimodal model - Related article: How Do Generative AI Systems Work? Generative Pretrained Transformer (GPT) A family of transformer-based language models originally developed by OpenAI. The name describes the approach: it generates, it is pretrained on broad data, and it uses transformers. - See also: LLM, transformer, foundation model - Related article: How Do Generative AI Systems Work? Generative UI (GenUI) Interfaces whose content, layout, or controls are generated or selected at run time by an AI rather than being entirely fixed in advance. Instead of navigating to a fixed screen, the user describes a need and the system composes a response, of