By John in computers — 19 Aug 2025

Assessing the dead Internet

I called it the Web in the post because that's what the W3C calls it.

Over the past few days I have seen several posts on social media and various blogs about how the Web is dead, for some definition of dead. Given that I have no regular readers on this blog, I feel a good thing to do would be to respond to these claims.

An analysis of the claim that the Web is dead raises two questions: What is the Web, and what does it mean for the Web to die? After all, it's difficult to say something is dead if we don't know what it is. The World Wide Web Consortium (W3C) provides us with a brief definition in Architecture of the World Wide Web, hereafter referred to as Architecture^[1]:

The World Wide Web (WWW, or simply Web) is an information space in which the items of interest, referred to as resources, are identified by global identifiers called Uniform Resource Identifiers (URI).

The key pieces here, then, are that there are resources and identifiers arranged in some space. As the term implies, a URI acts as a reference to some abstract resource. A proper URI specification exists^[2], but the informal explanation is that these are the links to resources users interact with every day to do things, such as https://treats.now or mailto:john@doe.com. The Web, then, is a system for organizing resources that connect to other resources using URIs. This system in turn uses the Internet as its means of communication.

While an identifier has its own well-defined semantics and specifications, the concept of a resource needs defining beyond "item of interest" for us to make use of it. Architecture, in a bit of circular reasoning, stipulates that one uses the term resource "in a general sense for whatever might be identified by a URI". Thus, a resource is whatever is identified by a URI, and a URI is whatever can identify a resource. Beyond this description of a resource, Architecture additionally defines an information resource as a resource whose "essential characteristics" can be conveyed via a message, or a unit of communication between two agents.

Put differently, an information resource is a message, and one with a consistent representation accessible via URI. An HTML document according to the Architecture is an information resource because it is made of bits, unlike a dog or car. Of course, one might ask whether this distinction is meaningful. Where does the boundary lie between the essential characteristics of an HTML document and that of an animal? According to Norbert Wiener, such a boundary does not exist. The distinction between the two kinds of resources is entirely artificial^[3]:

When one cell divides into two, or when one of the genes which carries our corporeal and mental birthright is split in order to make ready for a reduction division of a germ cell, we have a separation in matter which is conditioned by the power of a pattern of living tissue to duplicate itself. Since this is so, there is no absolute distinction between the types of transmission which we can use for sending a telegram from country to country and the types of transmission which at least are theoretically possible for transmitting a living organism such as a human being.

We will revisit the implications of this broader idea of information later.

Other descriptions of resources exist in early Web literature. In Web Architecture from 50,000 Feet, Tim Berners-Lee describes an alternative system of the Web as "socially aware state propagation" where this state is propagated using URIs to identify endpoints for calling procedures on remote servers^[4]. The URI scheme of choice is left unspecified. Other documents from W3C and Xerox PARC point to a reimagining of the Web as a combination of such remote procedures and distributed object storage as early as 1998^[5].

The tension between how systems communicate and what they communicate, then, has existed since the earliest days of the Web. A resource is information unless it isn't, in which case it's probably still information anyway. A URI points to a resource, but a resource could represent an endpoint for invoking a process on a system rather than some message. Distinguishing between these within a single URI scheme is discouraged because URIs are intended to be opaque^[6].

The result of this ambiguity is that many systems in practice choose their own interpretations of what a URI represents. Some tend towards the notion of a URI as synonymous with an information resource. Links in HTML and the visual language of browsers, for example, reinforce a concept of the Web as a set of interrelated pieces of information. Bookmark features assert the claim that a URI is a place you can go to, much like an actual bookmark in a book.

Other systems leverage the idea of a URI as a communication endpoint. RSS and GraphQL are two such technologies. The former is a framework for distributing news via a feed designated by some URI, where the contents of the feed are dynamic. The latter is a richer system for interacting with objects that are related to one another using a domain-specific query language. Both of these share the common trait of using a single URI for doing all the work, though to different degrees. While RSS at least requires feed elements contain a URI to some HTML document, GraphQL entirely abandons the notion of anything location-like, directing users to interact with a single URI to do all work.

If a URI isn't a location, then what is it? I like to think of it as a signifier for a communication channel in the information theoretic and cybernetic sense, where all parties have some agreed-upon understanding of the semantics of what might travel over that channel. A URI for a JPEG image doesn't point to a resource at all, but rather to a channel where the distribution of possible messages is such that both parties believe it will only ever contain one message. A URI for a GraphQL endpoint signifies a broader system of communication for retrieving and manipulating data, conversely, with a wide variety of things that can be communicated. From this perspective, a URI is much less like "go to this place" and much more like "talk to this person".

Given this, I would like to offer a slightly different definition of the Web from the W3C's: The Web is a communication space in which channels of interest are identified by global identifiers called Uniform Resource Identifiers (URI).

This distinction between item and channel matters because the foundations of the Web very much rest on this notion of a URI as a location. The W3C's definition of the Web describes it as a collection of items with identifiers. Search engines like Google aim to "index" the Web, expending vast computing resources in an attempt to chase down the latest output from every URI they can find using Web crawlers. Smartphones allow us to share links to whatever might be of interest in just a couple of taps on a screen. Any departure from this has extensive implications for all of the Web's technologies, and it is this exact departure which we are seeing play out now. This, I believe, is what marks the "death" of the Web.

Over the past decade or so, the overall flow of information on the Web has changed. Metaphors of the Internet as a collection of places no longer hold. One example of this in action is the single-page web app, a popular method of serving content currently. Such apps do not work by fetching resources and presenting them to users as static messages on a screen. Rather, they load a mix of source code and nominally static assets like images before running the code itself in the browser. A URI to such an app has little meaning; the entire application must run up to some point before it even begins to superficially resemble an information object of the early Web. The story is similar for actual compiled mobile apps and desktop applications, which may do away with the notion of a Web entirely in favor of interacting with a single graph API endpoint at a single URI. Such systems are not readily indexed by search engines, nor do they lend themselves to ideas like consistent representation of resources. For many applications like streaming media apps, it is not clear what indexing would even mean.

When the spatial metaphor of the Web is no longer valid, the Web as originally envisioned breaks down. A library of communication channels looks very different from a library of objects. Attempting to approximate the latter using the former merely results in systems that at most can only provide statistical approximations of what may have been communicated at some point in time with respect to some broader context, with all the associated caveats. It is entirely possible that the shift away from the Web as a collection of items is a major driver behind the apparent popularity of tools like LLMs.

This situation presents us with plenty of gloomy present and future states for the Web. We find ourselves drowning in AI slop that Google indexes in order to generate its own AI slop, the ultimate outcome of which may be some sort of epistemic collapse. Facebook and other social platforms make linking to information within their environments all but impossible, forcing users to live only within their app environments if they want anything approaching a coherent experience online. Algorithms distort our ability to communicate on the very channels we use to try to talk to one another. These phenomena threaten both the vision of the Web as both an information space and communication system.

Here I think a food analogy (courtesy of a friend, K) works well. McDonald's is the largest restaurant chain in the United States. It is also emblematic of the typical food swamp, where one has access to cheap fast food and little else. For many, McDonald's is cheap and easy to find, even if no one would accuse it of being too healthy. As an institution, it exists where other food providers either cannot or will not participate.

The giants of the Web perform the same function, often to a far greater degree than a fast food chain ever could. It would be absurd to imagine a world where government officials have sold their offices and hold press conferences at McDonald's, and yet with Facebook today that is largely the case; the death toll for the 2025 Kerr County floods in Texas was likely higher than it would have been otherwise because the county issued flood warnings over Facebook^[7]. This is partly because Facebook provides its own addictive hooks and feedback mechanisms like McDonald's, but also partly because coordinating communication - much like finding food in a food swamp - is hard. In a similar vein, Google killed its Google Reader RSS feed service in 2013^[8], shortly before pushing users to Google-hosted news pages and now LLM content. The common thread across these companies is that they actively exploit their respective domains to drive the production of high volumes of low-quality goods.

In light of this, it can be tempting to lay blame at the feet of large tech platforms and say they need to be broken up and people need to stop using them. Such a response is not entirely unreasonable, but it fails to take into account the inherent complexities of communication. Agreeing on where and how to communicate are difficult tasks, and the resultant challenges lend themselves to people using a few large systems rather than many small ones. Cities and Facebook both exist because they facilitate transmission of large quantities of information. Discord fills a niche that not too long ago was the domain of Microsoft, Yahoo, and AOL. Even IRC^[9] is not immune to this phenomenon of consolidation. A proper response to such platforms requires both understanding the harm they cause and accepting that, for better or worse, the history of telecommunications is also largely a history of monopolies.

I will leave this post on a somewhat positive note. Despite the problems it may pose in the moment, a change in how things work is not necessarily bad. There are many possible directions the Web can go in beyond those outlined above, some of which are very much in the spirit of earlier ideas of the Web that focused on its power as a decentralized system of communication. In social media, we can already see glimpses of this in systems like ActivityPub^[10] (i.e., Mastodon) and AT Protocol^[11] (i.e., Bluesky). Longstanding platforms like the Internet Archive's Wayback Machine attempt to index not just the most current response a given HTTP URI provides, but the responses as they have changed over time. RSS feeds still exist for many news sites. Your local librarians are probably digitizing and filing away volumes of information as you read this post, all of which is available to you if you want to look. The future is what we make of it; our use of the Web is no exception to this.