Home Glossary Item Grounding
« Back to Glossary Index

Grounding refers to the idea of linking abstract, symbolic knowledge or concepts in AI systems to sensory or perceptual data. It is about how machines tie their understanding of language, learned from data, to real-world concepts. The focus is on creating a meaningful connection between symbols (like words or phrases) used by AI and the entities or concepts that these symbols represent in the real world. This helps to facilitate more effective and context-aware processing of information.


The concept of grounding is vital in natural language processing, a field of AI that enables machines to understand and interact with humans in human language. For example, the word “apple” should not be just perceived as a mere combination of letters by an AI model, but it should be understood as a representation of a fruit that can be eaten, and can be red, green or yellow. Such a level of understanding requires grounding, connecting the symbolic representation (“apple”) with its worldly referents (the concept of the fruit apple and its properties).

Achieving true grounding in AI systems has been challenging. While models have become skilled at pattern recognition and predicting sequences of words in sentences, understanding the implications of those words in the context of the real world is a complex task. Progress is being made in areas like image captioning where AI models use visual grounding to relate language to visual data. Efforts are underway to create AI models with comprehensive understanding, where symbols and language are deeply associated with complex real-world concepts through robust grounding mechanisms.

« Back to Glossary Index