Definition
In artificial intelligence, ‘grounded’ describes the process of linking symbolic representations, such as words or logical propositions, to their actual referents in the physical world or sensory experience. This concept is central to Grounded Language Learning, where models learn semantics by correlating text with images, audio, or robot sensor inputs. Without grounding, AI may manipulate symbols syntactically without understanding their meaning, leading to hallucinations or lack of contextual relevance. Grounding ensures that AI outputs are anchored in observable reality.
Summary
Refers to AI systems that connect abstract symbols or language to real-world sensory data or physical actions.
Key Concepts
- Symbol grounding problem
- Multimodal learning
- Embodied AI
- Semantic alignment
Use Cases
- Robotics navigation using visual landmarks
- Image captioning systems
- Visual question answering