Embodied AI places artificial intelligence in a physical body that senses and interacts with the real world. Sensors such as cameras and microphones provide grounding that text-only language models lack. section 10.2.3
By comparing language-model assumptions with sensory evidence, an embodied system could detect contradictions, revise faulty common-sense rules, and learn from mistakes. Physical consequences make errors harder to ignore than online text errors. section 10.2.3
Counting apples
An embodied system could connect the statement “one apple plus another apple equals two apples” with objects it actually senses, grounding abstract reasoning in physical reality. section 10.2.3
