Show me a successful complex system, and I'll show you something that evolved via trial and error.
The Internet itself is one of these successful complex systems. Twenty years ago, the protocols that evolved into the Internet (TCP/IP) were not the ones people expected would win out. People expected that TP4/CLNP from the OSI protocol stack would win out, because (a) they were internationally supported, (b) they were backed by the ITU, and (c) they were proposed by really, really, really Smart people.
Of course, almost nobody has ever heard of TP4 or CLNP.* Seems that all those Big Idea Guys at the ITU didn't realize that something that works beats something the might work.
TCP/IP grew from systems that worked and evolved. TP4/CLNP was mired in formalism ("All n-layer functions shall communicate will all peer n-layer functions via n-1 layer functions." Blech.) and never survived outside the classroom (or the lab). TCP/IP went into the Unix kernel, and then into everything else, and won by default.
After evolving like crazy. I'm old enough to remember how Van Jacobsen electrified the networking community by making TCP run at 6 Mbps over Ethernet, when everyone "knew" that TCP would never give you more than 1 Mbps. Van figured out that networking belonged in the OS kernel, where golly gosh it lives today. Van evolved TCP.
Things that work win. The analogy of the United States - with 50 quasi-independent States able to experiment on their own - compared to France (5 separate Constitutions for a centralized state replacing each other in turn) is left as an exercise for the reader. Well OK, it's not really, but you know what I mean. Those Big Idea Guys in Paris (*COUGH* Valerie Giscard D'estang*COUGH*) didn't realize that something that works beats something that might work.
* Actually, I suspect that several of my readers will remember these, but this blog is pretty strange.