I love the concept and the visualizations, but am left wondering - evolving toward what? I don't get the same feeling of "complexity from simple rules" as I do from Conway's Game of Life.

I bet you could create a very interesting version of this with simple neural networks that start with random weights and determine the (x, y) coordinates to which an organism grows - reproduction is simply a matter of randomly selecting weights from two or more identical networks, and organisms that grow inefficiently are naturally selected against. I believe you could capture an increasingly complex range of organism shapes and introduce natural lifespans with a sufficiently deep network.