grounding
In plain wordsConnecting an AI's words to something real — trusted documents, a live search, images or the physical world — so that its answers rest on evidence rather than only on what it absorbed in training.
- 1
Language models
Conditioning a model's output on specified, verifiable sources — retrieved documents, search results, databases, tool outputs or user files — and constraining it to make only claims those sources support, ideally with citationscitation, n.: A pointer from something an AI says to the source that supports it — a link, document or quoted passage — so people can check the claim for themselves.. Retrieval-augmented generationretrieval-augmented generation, n.: Giving an AI an open-book exam: before answering, it looks up relevant passages in a set of documents and bases its reply on them, often citing where each point came from. is the standard technique. Groundedness (also called attribution or faithfulness) is evaluated by checking each claim against the supplied sources, often with model-based judgesLLM-as-a-judge, n.: Using one AI language model to grade or compare the answers of another, as a cheaper and faster stand-in for human reviewers.. Grounding reduces but does not eliminate hallucinationhallucination, n.: When an AI states something false or made up — a fake fact, quotation or source — fluently and confidently, as if it were true., since models can misread or over-extend sources. Several providers market 'grounding' as a product feature connecting models to web search or enterprise data.
Answers must be grounded in the policy documents; if they are silent, the bot says so.
- 2
Philosophy and cognitive science
The connection between symbols or words and what they stand for in the world, rather than only to other symbols — the symbol grounding problemsymbol grounding problem, n.: The puzzle of how words or symbols inside a machine can mean anything by themselves, rather than only pointing to more symbols — like a dictionary that defines every word using other words. posed by Stevan Harnad in 1990. Critics argue that models trained only on text lack such grounding; others hold that multimodal training, interaction or the structure of text itself can supply some of it.
- 3
Perception and robotics
Linking language to perceptual or physical referents: visual groundingvisual grounding, n.: Linking words to the exact part of an image they refer to, for example finding 'the man in the red hat on the left' and pointing to him. locates the image region a phrase describes, and grounded language learning in embodied AIembodied AI, n.: AI that has a body, usually a robot and sometimes a simulated character, and learns or acts by sensing and moving in a world rather than only processing text or images. maps instructions onto objects and actions that a robot can perceive and carry out in its environment.
- 4
Dialogue
In linguistics and psychology, the collaborative process by which conversation partners establish mutual understanding, or common ground, through acknowledgments, clarification questions and repairs — a concept from Herbert Clark and Susan Brennan (1991) that informs the design of conversational agents.
Origin
Stevan Harnad's 1990 paper 'The Symbol Grounding Problem' framed the philosophical sense, and Herbert Clark and Susan Brennan's 'Grounding in Communication' (1991) the conversational one. The language-model sense — answers tied to supplied sources — spread with retrieval-augmented generation from 2020 and became a product term in 2023–2024.
Usage note
The senses are related but not interchangeable: a model can be grounded in documents (first sense) while remaining, on some philosophical accounts, ungrounded (second sense).
Related
- retrieval-augmented generationretrieval-augmented generation, n.: Giving an AI an open-book exam: before answering, it looks up relevant passages in a set of documents and bases its reply on them, often citing where each point came from.
- citationcitation, n.: A pointer from something an AI says to the source that supports it — a link, document or quoted passage — so people can check the claim for themselves.
- hallucinationhallucination, n.: When an AI states something false or made up — a fake fact, quotation or source — fluently and confidently, as if it were true.
- factualityfactuality, n.: How far what an AI says is actually true about the world. A factual model gets facts, numbers, names and dates right and does not make things up.
- symbol grounding problemsymbol grounding problem, n.: The puzzle of how words or symbols inside a machine can mean anything by themselves, rather than only pointing to more symbols — like a dictionary that defines every word using other words.
- visual groundingvisual grounding, n.: Linking words to the exact part of an image they refer to, for example finding 'the man in the red hat on the left' and pointing to him.
- embodied AIembodied AI, n.: AI that has a body, usually a robot and sometimes a simulated character, and learns or acts by sensing and moving in a world rather than only processing text or images.
- parametric knowledgeparametric knowledge, n.: What an AI model 'knows' from training, stored in the numbers of its network rather than looked up in documents — like facts you have memorized rather than ones you check in a book.
- parametric knowledgeparametric knowledge, n.: What an AI model 'knows' from training, stored in the numbers of its network rather than looked up in documents — like facts you have memorized rather than ones you check in a book.
- hallucinationhallucination, n.: When an AI states something false or made up — a fake fact, quotation or source — fluently and confidently, as if it were true.
Sources
- Harnad, S. (1990). The Symbol Grounding Problem. Physica D 42, 335–346.
- Clark, H. H. & Brennan, S. E. (1991). Grounding in Communication. In Resnick, L. B. et al. (eds.), Perspectives on Socially Shared Cognition. American Psychological Association.
Mentioned in
hallucinationhallucination, n.: When an AI states something false or made up — a fake fact, quotation or source — fluently and confidently, as if it were true., knowledge cutoffknowledge cutoff, n.: The date after which an AI model knows nothing from its training, because its training data stops there. Without a web search tool, it cannot know about later events., retrieval-augmented generationretrieval-augmented generation, n.: Giving an AI an open-book exam: before answering, it looks up relevant passages in a set of documents and bases its reply on them, often citing where each point came from., citationcitation, n.: A pointer from something an AI says to the source that supports it — a link, document or quoted passage — so people can check the claim for themselves., confabulationconfabulation, n.: When an AI fills a gap in its knowledge with an invented but plausible-sounding answer, much as some people with memory disorders sincerely 'remember' things that never happened., embodied cognitionembodied cognition, n.: The idea that thinking is not something the brain does on its own: the body, its senses and movements, and its surroundings play a real part in how we think and understand., factualityfactuality, n.: How far what an AI says is actually true about the world. A factual model gets facts, numbers, names and dates right and does not make things up., glue on pizzaglue on pizza, n.: A notorious 2024 blunder in which Google's AI search summaries told people to add glue to pizza sauce, having taken an old joke from an online forum as genuine advice., machine understandingmachine understanding, n.: Whether, and in what sense, an AI system grasps what words, situations or problems mean — rather than just producing answers that look as though it does., symbol grounding problemsymbol grounding problem, n.: The puzzle of how words or symbols inside a machine can mean anything by themselves, rather than only pointing to more symbols — like a dictionary that defines every word using other words., WikidataWikidata, prop. n.: A free, shared collection of facts about millions of things — people, places, books, chemicals — kept by volunteers in the Wikipedia family and laid out so programs can use it as easily as people., parametric knowledgeparametric knowledge, n.: What an AI model 'knows' from training, stored in the numbers of its network rather than looked up in documents — like facts you have memorized rather than ones you check in a book., visual groundingvisual grounding, n.: Linking words to the exact part of an image they refer to, for example finding 'the man in the red hat on the left' and pointing to him.