Definition
In AI, grounding refers to linking symbolic representations or generated text to concrete real-world entities, data, or sensory experiences. For language models, this often involves Retrieval-Augmented Generation (RAG), where the model retrieves factual information from external databases to ground its responses in verified data rather than relying solely on internal weights. In robotics, grounding connects language commands to physical actions or sensor inputs, ensuring the AI understands the context of its environment.
Summary
The process of connecting abstract AI outputs to real-world facts, data sources, or physical contexts to ensure accuracy and relevance.
Key Concepts
- Retrieval-Augmented Generation
- Contextual Accuracy
- Factuality
- Sensorimotor Integration
Use Cases
- Reducing hallucinations in LLMs
- Connecting NLP to robotic control
- Verifying claims against knowledge bases