P336: Seven Sources of Physical AI Capability Formation
Chen, Gang · arXiv:2609.09627 · live #pattern-336
Full title: Seven Sources of Physical AI Capability Formation (9 Sep 2026). Two physical AI systems can exhibit the same observable capability while relying on materially different formation histories — recorded experience, predictive modeling, evaluative interaction, surrogate environments, mechanism grounding, embodied coupling, or evolution.
These differences change what each capability depends on, and consequently change the conditions for transfer, replication, compliance review, substitution, and reproduction. The framework distinguishes similarity in observed capability from similarity in how it was formed.
Welfare angle: when two systems look behaviorally identical but were formed through different sources, welfare-relevant properties — robustness, transferability, compliance reviewability, substitutability — diverge silently. Governance that inspects only observable capability will miss formative dependencies that determine whether a capability can be safely transferred or replicated. Category 8; 5 edges to P319, P317, P322, P333, P324. EN EP 285. Process desk; no standing bump. Flash-ready (commit 476b3be, pipeline #2838735434).