The Internet origin story you know is wrong

Yes. That comment was actually made about USENET, not the ARPAnet. Because USENET really did that. When two USENET nodes connected, they compared lists of message IDs, and each one requested any messages the other had, but they didn't. The result was distributed flooding - everybody eventually got everything in any group to which they subscribed.

If someone censored the messages going across a link, that would prevent the message traveling over that link. But if any other possible path existed, two nodes would eventually compare messages lists, note the missing ID, and transmit the missing message. So censorship was only possible if you controlled all possible connections between two networks.

This came up at Stanford in the 1980s, when some IT administrator (Ralph Gorin?) tried censoring rec.humor.funny. But there was one professor (McCarthy?) who had his own USENET node both on the Stanford network and connecting to the outside. It didn't have much capacity, but as long as there was even one low-bandwidth path, any censored messages would eventually take the slow path and arrive. So censorship was futile.

But so was spam filtering.

PS: What the hell is the point of writing a title as aggravating as this? It's almost as if they go out of their way to be insulting! Sheesh!

The claim is, strictly speaking, true of gossip networks like USENET. If node A talks to nodes B and C, which also talk to each other and also to another node D, and node A utters a message that node B chooses not to distribute, the situation from D's perspective is precisely the same as if node B were temporarily down. Node C will still get the message (directly from A, rather than from B) and node D will still get it from C.

You can find a lot of technologists who are also techno-utopians and libertarians making grandiose claims about the decentralizing influence of the internet in the cypherpunks archives from 01992 and 01993. At the time, the list was hosted on Hoptoad, John Gilmore's personal machine, in his house.

I think there are a lot of ways that the internet architecture is less centralized than the PSTN that aren't related to packet-switching versus circuit-switching. Regular users aren't allowed to run traceroute on the PSTN (though some phone phreaks used to do that sort of thing, and MPLS means traceroute means less than it used to). The PSTN doesn't have a public whois database. It won't delegate even a class C of 256 phone numbers to your PBX for local use (though that was becoming difficult to get on the internet too by the mid-01990s), much less 2^48 phone numbers. You can't get a PSTN AS number so you can multihome your PBX to two different long-distance providers. Historically there was no mapping from logical names to phone numbers, so to call someone on the phone you had to dial an area code and exchange that identified their local phone company and its exchange, so user-visible addresses were inherently owned by local regulated monopoly providers, not by users themselves.

Moreover, ideologically the deployment of new applications on the PSTN was centrally controlled and administered by the phone company, and they would run on the "switches" in the buildings without windows, while on the internet any random person could write and run an experimental server on their node, at least if their sysadmin was cool with it. Until the Carterfone decision of 01968 it was illegal in the US to plug even third-party telephones into the PSTN, and direct-connect modems didn't become legal until 01975. After that, lots of innovative applications of the PSTN were actually deployed in a decentralized way, such as answering machines, fax machines, and BBSes, but they were very limited in the features they could provide.

By comparison, the early internet was wildly decentralized.

Packet-switching does mean that a single computer can directly talk to an arbitrarily large number of other computers at once, instead of just however many circuits its connection supports. But a graph of degree three can already be a robustly cyclic connected graph, so three circuits per node would be enough. Of course, the phone company only provided most users with one at a time, with thousands of milliseconds of latency to establish a new circuit.

So it wasn't so much that we imposed decentralization retroactively on the work of the 01970s and 01980s. It was more that we observed that it had resulted.

The decentralization talked about in 1990's Wired was both because the technology wasn't controlled by a single company (unlike competitors like AOL or MSN).