AGI Requires Embodiment
Description
Artificial General Intelligence cannot emerge from purely digital systems processing symbolic information. True intelligence requires sensorimotor experience in a physical environment to develop grounded understanding of concepts and causality.
Falsification Criteria
This conjecture would be falsified if: (1) A purely digital AI system demonstrates general intelligence across at least 5 distinct domains (scoring 90%+ on standardized tests for each) by 2033; (2) The system can pass specialized Turing tests focused on physical reasoning without having been explicitly trained on embodied data; (3) The system demonstrates the ability to learn new physical tasks from descriptions alone with 85% accuracy when tested in simulation; (4) At least 3 independent research institutions verify these capabilities using standardized benchmarks; and (5) The system architecture and training methodology is fully documented to prove no embodied data or embodied simulation was used in its development.
Bounty
Contribute to the bounty for anyone who can successfully refute this conjecture
You must be signed in to contribute to the bounty.
Sign inRefutations
Rational criticism and counterarguments to this conjecture
Recent language models demonstrate understanding across domains without embodiment. GPT-4o can reason about physical situations, social dynamics, and abstract concepts despite never having a body and pass the Turing test. This suggests that sufficient training on human-generated text can substitute for direct embodied experience. Source: https://arxiv.org/abs/2503.23674
You must be signed in to submit a refutation.
Sign in