Power as Organization: the Politics of Compute Networks
The above Eli the Computer Guy clip is good and important. It contains a key understanding to the politics of compute. Only twenty minutes, it’s well worth watching.
Nvidia has come up with something they’re calling “Pair,” which allows AI – “like fuck me, right?” – tasks to be performed not with one connection to a centralized server, but by clustering various compute devices through numerous connections, such as connecting together your PCs, phones, and whatever other device you have with under utilized compute. Pair connects these distributed nodes to combine them for collective compute power. Eli explains this process well, so I’ll skip more detail.
What Nvidia Pair proposes is nothing new. They call it clustering, but it’s been known long before as distributed networks or peer to peer networking. Such decentralized network design, its architecture, is imperative if we are to have any future democratic politics with the technology of compute.
The history of compute is essential to understand. Once again on this history, I cannot more highly recommend Yasha Levine’s Surveillance Valley. Compute began in the 1950s with large centralized computers, mainframes, containing all compute power. Over the next couple decades, compute became increasingly more powerful, the machines got smaller.
In the early 1980s, came the PC Revolution, so labeled by the still bad societal hangover suffered from the 60s. The idea was to put compute power on every desk at work and in every home. At work, PCs became networked together through a centralized computer server allowing communication and data sharing among the network's various PC nodes. All communication and data sharing of the nodes have to first go through the centralized server. This network architecture, again architecture simply meaning the design and configuration of the network, was called client/server. It became the default architecture for business computer networks. The important thing to understand in a client/server network is who controls the server, controls the network.
As the PC Revolution took place, the internet continued to develop. The internet started in the 1960s with the networking of the mainframes in universities, corporations, and government agencies (largely military), the large entities who could afford them. In 1991, with the establishment of the World Wide Web, the net was opened to everyone else – public and business – through the established mainframe architecture.
In sending data between servers on the internet, a system was developed called Transmission Control Protocol and Internet Protocol (TCP/IP). Data from any given computer is broken into packets to send to another computer. What’s most intriguing about TCP/IP is with any data set broken into innumerable packets, the packets sent from one computer server to another could take different routes across various backbone internet servers to reach a destination server.
This ability for this distributed routing of packets gave rise to the wrongheaded notion and wrongly promoted by far too many, that the architecture of the internet was distributed. It wasn’t. The net’s architecture was entirely dependent on centralized servers. As more and more computers connected to the internet, exponentially more data was created and transferred. In the 1990s, centralized "server farms" were created, the internet established what is basically a client/server relationship between a handful of corporations and everyone else. Today, we are in a completely insane race to centralize these processes and increase mega-corporate control to an even greater degree with data centers.
I became introduced to the internet in early 1993. Rolling Stone had an interview with Mitch Kapor, who was one of the early internet proselytizers. Mitch was an archetype of the PC Revolution, 60s kid in all that that meant, and like many others, at the right place and right time with compute. He produced the electronic spreadsheet Lotus 1, 2, 3, think Excel today. In 1983, his spreadsheet became the “killer app” of the new IBM PC, unfortunately most copies of Lotus sold came attached with the MS-DOS virus. Four years later, Kapor walked away with some hundreds of millions of dollars in 1980s money.
At that time, having just spent nine months of my life on a 1992 presidential campaign alerting the American people for the desperate need for political reform, being introduced to the internet was beyond intriguing. It didn’t take much imagination to think this could help initiate a badly needed democratic communication revolution catalyzing needed political reform.
In the RS interview Mitch said several things that struck a cord. He was just as worried about centralized corporate power as government power, with compute “architecture was politics,” and he mentioned, clearly without much understanding, Jefferson. Unchecked corporate power and Jefferson were my wheelhouse and I knew more than most about compute and took it upon myself to learn a lot more.
Funnily, I dug up Kapor’s office phone number in Cambridge, called it, and he answered. We talked about 20 minutes. After, I went straight away and got my first internet connection from a guy selling them from a garage in El Cajon. The first work I did with Mitch was compiling some of Jefferson's thought and how it might apply to this new era.
We communicated for the next 15 years, including a stint in the last three in his office south of Market, where he paid me a little to develop a political reform program, but he never would fund it. So, I left. Like most in the Valley, Mitch isn’t well politically wired. And by politically wired, I don't mean the shit show we call politics today, but that much deeper instinct of this social animal defining who we are. It doesn't help anyone that our present political culture in no way fosters for anybody our better political instincts those "better angels of our nature." After I left, I came to realize all the years I talked about political organizing, he didn’t really understand what it was. To all our detriment, it is a lost art in America.
I have an email I sent to Mitch from the late 1990s, where I had figured out the architecture politics of compute networks. It was clear a client/server architecture could not be democratic. At the time, the decentralized myths of the net still pervaded and money was beginning to be made by the boatload. Politics had zero value for all those developing the technology.
Centralized server data centers would start springing up across the land and companies like Google, Microsoft, Facebook, Amazon et al. would grow to control and profit from the majority of interactions across the internet by creating centralized architectures that were basically client/server. Controlling servers meant profit, democracy valued not at all.
Today, with the latest generation of compute marketed as AI, trillions of dollars are being spent on centralized data centers. Make no mistake, every one of these data centers represent a loss of power for everyone except those who control them. Democracy will not only not be gained through centralized compute, it can't be.
This quick history of compute by no means meant suggests Nvidia Pair is on your side, it's not. Again, talking about distributed networks has been around a long time in corners of the industry. A few will remember in the late 90s Napster, a wildly popular distributed architecture the industry crushed as soon as possible. To refresh my memory, I looked up Napster. Google AI responded, “Mass piracy: Napster popularized free, illegal downloading.” How fucking funny is that? As you read this, Google’s and the rest’s AI bots scrub the net for every bit they can loot. Don’t worry, they’ll happily sell it back to you, even what they stole from you.
A real value of this latest generation of compute will be automation, whether of information, communication, or physical acts. Automation is about repetition not change. The idea of automating many current societal practices, designs, and institutions through massive, centrally controlled processes, when so many of them first need to be changed should be abhorrent to all. Its promoters should not be celebrated but condemned.
An essential point to understand and it is true of every technology, adopted technologies homogenize. Once in place, changing them becomes difficult. The more powerful the technology, the greater the values instilled are those of the technology itself not humanity. The alternative to technologically imposed values are human values. The only way to keep human values dominant is through politics. Technological architectures must all be designed with dynamic human elements.
Presently, the only decisive human value in all technological development is what's accounted as profit. A value overwhelmingly imposed and defined by an oligarchic corporate order. They seek to present profit as beyond political design, as some sort of divine order. What we do call politics are the decrepit processes centered around our centralized DC cesspool and bad reality TV. No one cares much for it as the processes and institutions are perpetually degraded by those claiming to save it.
Our cultural, political, and government organizations need to evolve. Societal adoption of new technologies impacts human culture similarly to the environmental forces of natural selection upon a species, however technologies homogenize more than diversify. To utilize distributed network architectures, we need to simultaneously develop distributed political, government, and cultural order, that is democratic organization. Creating both distributed compute architecture and democratic political organization can only be accomplished as an act of mass political will, beginning by talking face to face.
Correction: In last piece sent, I mistakenly attributed to Illich what was actually from Vaclav Smil’s Energy and Civilization. A nice related thought of Illich's is "Beyond a certain point, more energy means less equity."