The appointment of Ajit Pai to the FCC chairmanship provoked a chorus of claims about his ardent opposition to net neutrality. These charges are both true and false depending on what net neutrality means to you.

While it’s certainly true that Pai opposed the Wheeler FCC’s 2015 Open Internet Order, his opposition was mainly to the Title II legal authority used by the agency to claim jurisdiction over Internet Service Providers. Whether he’s actually opposed to net neutrality as an ideal is an entirely different question.

Alternate Internets Not Allowed

At its core, net neutrality is the idea that the firms that provide different kinds of services on and in the Internet have to be put into boxes and not allowed to cooperate with each other in meaningful ways. For example, “edge services” companies like Facebook can’t deliver data all the way to their customers.

Facebook is required to give its data to a general purpose ISP for delivery to its customers. That’s mainly a structural fact, as direct delivery would require Facebook to lay wires or build radio towers in each neighborhood.

This is an expensive proposition, so Facebook doesn’t want to do it. But what if Facebook could lease communications capacity to homes and offices for its own exclusive use? With some two billion users and a host of services from messaging to video, it’s not unthinkable that such a move would be worthwhile at some point.

Net Neutrality is a Modularity Mandate

But the Open Internet Order does not allow Facebook to lawfully communicate with users in such a direct way. Leased communication capacity is either a “Sponsored Data Service” (¶ 151) or a “Non-Broadband Internet Access Service Data Service” (¶ 207), both of which are severely constrained by the regulation.

Facebook can’t buy the service if ISPs can’t sell it, and the FCC doesn’t want the Internet to work this way. So the ISPs are in little boxes where they simply transmit mashed-up data between users and edge services, and edge services are in their little boxes where they can only communicate directly with ISPs. It’s a modularity mandate.

The idea that ISPs have to treat all data the same as all other data only makes sense if you conceive of the role of ISPs and edge services in this particular way. Net neutrality is also a mandate for blindness on the part of ISPs as to the nature of the data they carry. But this isn’t the only way to build a network, or even the best way.

There’s no Reason for It, It’s Just our Policy

Historically, the Internet has always been a modular system that separates essential services from each other in concept, if not in operation. It consists of applications, Internet routing, and end to end data control, each done according to a protocol with little or no knowledge of the other protocols (HTTP, IP, and TCP, for example.)

This organization mirrors the organization of computers at the time of the Internet’s creation: highly modular systems consisting of racks, circuit boards, operating systems, applications, and peripherals (such as terminals and disk drives). Some of these parts are hardware, some are software, and all computers were designed this way in the early ‘70s.

The Internet was designed by computer geeks rather than communication nerds, so this is just the way the geeks approached system design. The modularity theme dominated the design of computers from mainframe to minicomputer to personal computer, but it hit a wall with laptops. Modularity in the traditional sense doesn’t work at all for smartphones.

Google’s Project Ara Debacle

Jean-Louis Gassée, the ahead-of-his time brain behind Apple in the ‘80s, has written a very comprehensive account of the Ara project on Medium’s Monday Note: “Lazy Thinking: Modularity Always Works.” He explains the thinking about modularity during the heyday of the desktop PC that lead to a spectacularly wrong approach to smartphone design. Here’s the pitch for Ara:

“Led by Motorola’s Advanced Technology and Projects group, Project Ara is developing a free, open hardware platform for creating highly modular smartphones. We want to do for hardware what the Android platform has done for software: create a vibrant third-party developer ecosystem, lower the barriers to entry, increase the pace of innovation, and substantially compress development timelines.”

The Ara people were hoping to kindle a PC revolution for smartphones, freeing users from the control of companies like Apple and Samsung, and creating a vast abundance of applications, hardware add-ons, and amazing customization. It didn’t work for all the obvious reasons: Ara phones were too big, too heavy, too expensive, too slow, and generally not worth the bother.

When electronics get as small and powerful as they currently are, design weaknesses show up most dramatically in the interconnections between components, and Ara made interconnections more troublesome.

Precision machining and clever internal organization are also very important to electronics in small spaces. Ara made the spaces smaller and thereby forced inefficient trade-offs on designers. This all led to the clunkiness of the phone.

Modularity Doesn’t Always Win

Google may have been conned into believing that modularity was the road to domination of smartphones by disruption guru Clayton Christenson, author of The Innovator’s Dilemma. Christenson predicted that the iPhone would fail because it’s a closed system that doesn’t have enough modularity.

Ara tried to build a product conforming to Christenson’s dictates and failed. The failure wasn’t so much a consequence of poor execution as a bad idea. It turns out that modularity is not free, and sometimes its costs exceed its benefits.

Microsoft Windows is one example of the costs: in its heyday, Microsoft spent more engineering resources on device drivers (relatively small programs that control hardware devices) than on core software. This was necessary for Microsoft to support a vast array of hardware add-ons, but over time it means the quality of the core software will erode. The inventory of devices constantly grows, and there’s no elegant way to make the old devices obsolete.

This is one reason Microsoft Windows has become so troublesome while Mac OS X has continued to be relatively stable. Because hardware vendors that use Windows are not free to modify the operating system, their ability to make fundamental innovation is also limited by modularity constraints.

Costs of Internet Modularity

The Internet’s modularity has its benefits. ISPs can buy network equipment from a number of vendors, and they can also support a vast array of applications and user devices. This is great, but the fundamental services provided by ISPs are limited by these benefits.

The Internet is like a state highway where bicycles, cars, and trucks share a common roadway. This design aims to keep costs low, but it’s not consistent with safety or performance.

A safe design would separate cyclists from cars and trucks, probably with a wall as they do in Copenhagen. And it would separate slow-moving trucks from cars, especially going up hills.

And it would allow people with high-performance cars to drive super-fast, like the German Autobahn. It would also provide accommodation for flying cars and drones.

Moore’s Law at Odds with Neutrality

We don’t build roads that way because the costs are too high. But it we had Moore’s Law roadways that cost half as much every 18 months, we’d be able to optimize the design.

Similar logic applies to the Internet. Maximum innovation means infrastructure that supports and encourages multiple business models for diverse services.

If an innovator wants their own leased pipe to each customer, they can negotiate for it. If they want to share the road with everyone else to save money, they can do that, and if they want late night access to backup services practically for free, why not let them have it?

Neutrality Inferior to Positive Goals

The major drawback of neutrality as a policy aspiration is that it’s a negative goal: it prohibits a set of practices but refuses to say what it wants to accomplish. We can have much more productive policy discourse if we state positive goals.

If the goal of our network policy is for networks to become better – faster, more reliable, more secure and with greater reach – as well as cheaper, does it matter whether the underlying technical structure is as modular as a 1985 IBM PC or as sleek as an iPhone 7?

Modularity doesn’t always win in the marketplace and we should not assume a priori that it must win in the regulatory space. There are enormous efficiencies in integrated designs that our innovators should be able to explore as long as they meet consensus positive goals.

So yes, net neutrality is doomed by history and technology regardless of who sits atop of the FCC.