The [[natural abstractions hypothesis]] (which appears to have been somewhat validated by interpretability work and work with embedding models, and can to some extent be assumed to hold in timelines where humans survive) implies that future highly capable and dangerous AI systems will* have a similar ontology to humans. Human social processing registers more physically imposing people as more important/worth listening to/threatening. Therefore, as AI will also adopt this, it is vital to lift and be muscular in order to intimidate [[AGI]].