Well, I'll take a shot at this - most of what I'm going to say comes from the text of Thiemann or Rovelli. The choice of $g_e\in SU(2)$ is connected to the holonomy, which we choose to use because it is a gauge-invariant. In fact, it's probably more pedagogical to say "the loops come from the holonomy $g_e\in SU(2)$, which we know can be made to be gauge-invariant Wilson loops" or something. Now as far as the choice for the 2-complex, we want something related to the triads so that the original Poisson algebra between the connection $A$ and fields $E$ is preserved. It actually turns out that $E$ is an $SU(2)$-valued vector density - dual to a 2-form represented by the 2-complex. In other words, we don't have two 1-forms, but rather a 1-form and an $SU(2)$-valued vector.

So that's a short answer coming mostly from Thiemann's excellent text. This issue is taken up in depth in the third of the Ashtekar series of papers:

Ashtekar, Corichi, Zapata (1998), Quantum Theory of Geometry: III. Non-commutativity of Riemannian structures, Class. Quan. Grav. 15, 2955-2972.

This post has been migrated from (A51.SE)