Definitions
Definitions change with the field. This page distinguishes established standards, common technical usage, and terms that do not yet have one settled definition.
“Artificial intelligence” names a broad field. A chatbot based on a large language model is one kind of system within that field. It is not the whole field.
Software is not a person.
KISOP names the responsible actor and the actual operation: a company trained a model; a model generated text; a system ranked an application; a sensor detected movement; a robot moved an object; a person approved the result.
The broad field
Artificial intelligence
A field concerned with machine-based systems that infer from inputs how to produce outputs such as predictions, content, recommendations, or decisions that can affect physical or virtual environments. KISOP uses “the field of artificial intelligence” when discussing the whole area.
Based on the OECD's updated AI-system definition.
AI system
A particular machine-based system, not an invisible general intelligence. The system includes the model, software, data flows, interfaces, people, operating rules, and environment needed for a use.
Machine learning
Methods that produce models from patterns in training data rather than expressing every rule as hand-written instructions.
Deep learning
A family of machine-learning methods built with multilayer neural networks. Large language models and many image, speech, and perception models use deep learning.
Language and generative systems
Generative AI system
A system that uses a model to produce synthetic text, images, audio, video, software, or other content based on patterns learned from data.
Foundation model
A broadly trained model that can be adapted or used across many tasks. Foundation models may process language, images, audio, video, or several kinds of data.
Large language model (LLM)
A model trained on large collections of text or other tokenized material to predict and generate sequences. An LLM produces language output; it is not automatically a chatbot, database, source of truth, or decision-maker.
Chatbot or conversational assistant
An interface that accepts conversational input and returns conversational output. A modern chatbot may use an LLM, retrieval system, tools, business rules, or several models. ChatGPT, Claude, and Copilot are products and interfaces, not synonyms for the entire field.
Multimodal model
A model that accepts or produces more than one kind of data, such as text, images, audio, or video.
Retrieval-augmented generation (RAG)
A generative system in which a separate retrieval system supplies selected material to a model as context for an output. Retrieval may improve grounding, but the output still requires checking.
Agentic system
Software that uses one or more models to select and carry out a sequence of actions toward a stated objective, often by calling tools or other services. The people and organizations that configure and deploy the system remain responsible for its permissions and effects.
Prediction and decision systems
Classifier
A model that assigns an input to one or more categories—for example, classifying an image or flagging a transaction for review.
Predictive model
A model that estimates a future value, probability, or risk from available data.
Recommendation or ranking system
A system that orders or selects items, people, opportunities, or content according to a calculated score.
Decision-support system
A system that supplies information, estimates, or options to a person responsible for a decision. Whether people can understand, challenge, and override the output matters as much as the model.
Automated decision system
A system that makes or materially determines a decision without meaningful human judgment. Housing, employment, credit, insurance, health, and public-benefit decisions can carry special legal duties.
World models and physical systems
World model
A model intended to represent aspects of an environment and predict how its state may change. A world model can support simulation, planning, video generation, autonomous driving, or robotics. Researchers do not yet share one universally accepted definition, so a page using this term should state what the model represents and predicts.
See the 2026 research perspective A Definition and Roadmap for World Models, which explicitly documents the lack of consensus.
Embodied AI
A research area involving systems that perceive and act through a body—usually a robot or another device with sensors and actuators—within a physical or simulated environment.
Physical AI
A broad and increasingly commercial term for model-based systems that sense, reason about, or act in the physical world. It may include robotics, autonomous vehicles, industrial systems, drones, and embodied systems. Because there is no single settled standard definition, KISOP names the specific technology whenever possible.
Robot
A physical machine that senses or acts in the world. A robot may use machine learning, ordinary control software, remote human control, or a combination. A robot is not necessarily an AI system.
Autonomous system
A system able to perform some actions without direct moment-by-moment human control. Autonomy has degrees; a page should state which actions are automated, under what conditions, and who can stop them.
Digital twin
A digital representation of a physical object, process, or environment that is updated with relevant data. A digital twin may use AI-based models, but it does not have to.
How models are built and used
Training
The process of adjusting a model's parameters using data and an objective.
Inference
Running a trained model on an input to produce an output.
Prompt
Input or instructions supplied to a generative model. A prompt may include text, images, retrieved documents, tool results, or structured data.
Output
Content, a score, a classification, a recommendation, an action request, or another result produced by a model or system.
Fabricated or unsupported output
Content generated without adequate support in the available evidence. The field often calls this a “hallucination”; KISOP prefers wording that describes the failure rather than suggesting human perception.
Local model
A model run on a device or machine chosen by the user rather than through a remote provider. Local operation may reduce data transmission, but does not by itself guarantee security, privacy, accuracy, or maintainability.
Cloud model service
A remotely operated service that receives inputs over a network and returns model outputs. Its data handling depends on the provider, account, settings, contract, and product configuration.
Open-weight model
A model whose learned parameters are available under stated terms. Available weights do not necessarily include training code, data information, or permission to modify and redistribute the whole system.
Open Source AI
A system made available with the freedoms to use, study, modify, and share it, together with the preferred form needed to make modifications. “Open weights” alone should not be presented as equivalent.
Governance words
Provenance
The record of where data, content, or an output came from, how it changed, and who or what handled it.
Human review
A named person's examination of specific work against stated criteria. Merely placing a person somewhere in a workflow does not establish meaningful review.
Consequential decision
A decision that can substantially affect a person's rights, opportunities, access, safety, care, housing, work, credit, insurance, education, or public benefits.
AI-assisted
A disclosure label meaning that one or more systems from the field of artificial intelligence contributed to a stated part of the work. The disclosure should name the system, task, reviewer, and status rather than implying that “AI” acted as an author.
Writing rule for this site
| Avoid | Prefer |
|---|---|
| AI thinks or knows | The model generated; the system retrieved; the source states. |
| AI decided | The system produced a score; the named person or organization made the decision. |
| AI wants or refuses | The system was configured to permit or block the request. |
| AI read the document | The document was supplied to the model as input. |
| AI helped | Name the system and operation: generated a draft, classified an image, retrieved records, or proposed code. |