> Small-scale methods may be easy to analyze, but they lack the richness and complexity that makes intelligence interesting.

I don't disagree.

But what a scientist would do after having strong evidence that huge scaling might help is attempting to understand what part of the much larger complexity leads to this qualitative change.