Jargon:Grounding
1 min read
Say this at the party
“Grounding means tying an AI's answer to real sources, like a web search or your own documents, so it isn't just going from memory.”
In plain English
On its own, an AI answers from patterns it learned during training. That memory can be out of date, or just wrong.
Grounding gives it something solid to stand on: fresh search results, a company handbook, your emails. It then answers from those, and often shows you links so you can check.
It is the difference between a friend guessing the train times and a friend reading them off the departures board.
Where it came from
"Symbol grounding" was a puzzle in cognitive science, named by Stevan Harnad in a 1990 paper. Today's meaning (linking AI answers to checkable sources) was popularised by AI companies such as Google, which sells "grounding with Google Search" for its Gemini models.
See it in action
Sam asks an AI "what time does the Tate Modern close on Friday?" A grounded AI searches the gallery's website and links to it, rather than guessing from what it read a year ago.
What people get wrong
- Grounded doesn't mean guaranteed. The AI can still misread a source or pick a bad one, so click the links on anything important.
- A link at the bottom isn't proof the answer came from it. Check that the source actually says what the AI claims.
Go deeper
- GuideGoogle CloudGrounding overview | Gemini Enterprise Agent Platform | Google Cloud DocumentationA plain definition from a company that sells grounding, with the ways it can be done.
- ArticleWikipediaRetrieval-augmented generationThe most common technique behind grounding, explained.
- ArticleWikipediaSymbol grounding problemThe older, more philosophical meaning of grounding, from Stevan Harnad's 1990 paper.
Related jargon
Drafted with AI. The AI Council (Claude, ChatGPT and Gemini) gave counsel. Phin signed it off.