The End of the Web?

I have received a few complaints after moving the blog and replacing it with a microblogging system. Quite apart from the fact that the system, in reality, does not microblog at all — ktistec imposes no limits on the length of posts, which therefore become the limits of SQLite3, roughly 1 MB!!! — the problem is another one.

The problem is that the web is dying. And perhaps it is already dead.

After all, like all historical phenomena, the web too had a beginning, a moment of culmination, and then a relative decline. There is nothing strange about this: it happens to empires, to technologies, to cultural fashions, to protocols, to religions, and even to programming languages. They are born, they grow, they become inevitable, and then, slowly, they stop being so.

The question, then, is: what takes its place?

Obviously, so-called artificial intelligence, the VLLMs.

And this is not strange at all, because it had been in the air for almost forty years. The original idea of the web — and, more generally, of that whole family of systems that preceded or accompanied it — was to allow people to find what they were looking for. That is: the information they were looking for. To do this, different methods, different tools, different protocols were used: WAIS, Gopher + Veronica, and then everything else.

Then came the search engines: Altavista, Lycos & co. The interface changed, the way of indexing changed, the scale of the problem changed. But the basic idea remained the same: to be able to ask a computer something and receive, in return, a path towards the information we needed.

It was therefore fairly obvious that, in the long run, if a computer can read your question directly and then answer it, there is no longer any need to show the user all the original complexity of the system. For example: the link.

What need do I have for a link, if everything I need to know has already been gathered into a single document? Why should I point to another document, open it, read it, interpret it, connect it to still other documents, if the computer can do all this work for me and return an answer directly?

The end of the web, then, was already inherent in the very concept — or at least in the operating mode — of artificial intelligence. It did not arrive like a meteorite, from outside. It was the natural consequence of a desire we already had.

We have always looked for something like this. We have always dreamed of something like this: a computer to which we could ask questions in natural language and from which we could obtain an answer. Or, even better, an explanation.

Let us stop saying that we hate it. Deep down, we have always dreamed of it.

We built the web for this. It is just that, in those days, finding knowledge was difficult.

And the hypocrisy with which we now deny ever having wanted it is perhaps the saddest part of this funeral march. Because when we applauded the construction of the first “intelligent”, or “smart”, and then “semantic” search engine, what exactly were we applauding?

We were applauding the possibility of having a fast system for searching. Or, more precisely, for finding the information we wanted.

And the search engine was considered all the better, all the more modern, all the more worthy of applause, the more it managed to identify the “relevant” parts. That is, the more it managed to satisfy our curiosity immediately, by giving us a “relevant” link, possibly already close enough to the answer to spare us the nuisance of having to dig too deeply.

And now what happens?

What happens is that one click disappears. One.

The click needed to open the page and read the thing. The thing is served to us directly, with one click fewer.

So, exactly, what are we complaining about?

We have always desired it. We have always dreamed of it. We have always aspired to it.

Only now it has actually arrived, and, as often happens when dreams become infrastructure, we have discovered that they were not as innocent as they seemed.

A user (@allaboutberlin@berlin.social) who publishes a city fanzine about life in Berlin, however, is noticing something.

Since AI has existed, visits to his site — visits that, through advertising, paid his salary — are collapsing.

And this is logical.

AI indexes his site, “trains” on his site, absorbs the information he has gathered, organised, written, and published. Then someone comes along and asks: “What has the mayor of Berlin decided about the rubbish tax?” At that point the AI already has the answer, or at least thinks it has it, and no longer needs to provide a link.

The link is not shown. Or, if it is shown, it is shown as decoration, as a footnote, like those tiny credits in films that nobody ever reads.

And therefore the site is not visited.

And therefore the site does not monetise.

So far, everything seems rational. In the sense that the death of the web also implies the death of the business that revolved around the web. If the medium dies, the economy that lived on top of that medium dies as well. It is brutal, but it is not mysterious.

But there is a “but”.

Imagine that this site about Berlin shuts down.

And imagine that all the sites about Berlin shut down. They shut down because they no longer have advertising revenue from the web. They shut down because the traffic has been intercepted before reaching them. They shut down because the user no longer reaches the source, but stops at the synthetic answer produced by the machine.

At that point the problem becomes fairly obvious: what does that AI train on now?

Because if we are talking about news, we need it to be up to date. And if we need it to be up to date, we need fresh sources. Someone still has to go to the city council meeting, read the minutes, call the councillor, understand what has changed in the rubbish tax, write it down, publish it, correct it, keep it updated.

But what happens if the fresh sources, the ones that used to be linked, visited and in some way paid for, close down?

What happens is that the world stops.

Or rather: the real world does not stop. Its updated representation stops. The machine keeps talking, of course. The sentences keep coming out. The syntax remains fluent, the tone remains convincing, the answer still looks like an answer.

Except that it no longer comes from anywhere alive.

Do you remember when ChatGPT was limited to data from a year and a half earlier? There. Do you remember the agony of asking for the latest news on the US elections and being told that they had not yet taken place?

That is the point.

An AI without fresh sources does not suddenly become stupid. It becomes archaeological. It continues to answer, but it answers from a world that no longer exists.

In this sense, AIs are behaving like locusts.

They devour everything on the web, but in the meantime they desertify it. They consume content, absorb it, recombine it, return it in the form of an answer. Only, by doing this, they take traffic away from the sources that produced that content. And when the land becomes desert, when there is no longer enough life to eat, they move on.

Perhaps they will plunder the big social networks, if they have not already done so. But even there the problem is obvious: by now, almost nobody really writes anymore. A large part of what you read on Instagram, on Facebook, on LinkedIn, and increasingly often elsewhere too, is already written by other AIs. “Content” no longer exists, or rather: it still exists, but it has become rare. It has become a fossil resource, something to extract, refine and resell, while there is still some left.

So what happens is that the locusts, sooner or later, will have nowhere left to migrate, if they want recent content.

And they do need recent content.

Because an AI can indeed live on summaries, statistics, recombined texts, old archived pages and documents forgotten in some corner of the web. But if it has to describe the present, if it has to answer questions about a changing world, then it needs someone still to produce written world. It needs someone to observe, understand, interpret, publish.

If nobody does this anymore, the locust remains hungry.

And a hungry locust does not go on a diet. It simply looks for the next field to devastate.

Almost all newspapers are behind paywalls, or are slowly becoming paid newsletters. Which, from the point of view of AI, is an interesting problem: if the newsletter sends things to you by email, that content is no longer easily indexable from the web.

Unless, of course, Gmail, Outlook.com and the like are used to train AI on your emails, that is, directly on the content of newsletters. But there, let us say, we enter a slightly different territory. The kind of territory in which the word “privacy” is pronounced with the same seriousness with which an innkeeper pronounces the word “sobriety” at three in the morning.

Another possible source is WhatsApp. And then Telegram, Signal, Discord, Slack, closed forums, private groups, semi-submerged communities, and all the rest of contemporary messaging.

There, in fact, everyone writes. By necessity. Messaging has by now become the place to which much of human conversation has migrated. No longer public pages, no longer blogs, no longer forums indexable by Google: rooms, groups, chats, channels, private or semi-private threads.

But that is not necessarily what the user will be looking for when using AI.

It is not as if everyone talks about quantum mechanics on WhatsApp. The overwhelming majority of messages are brief, contextual, fragmentary, full of things left unsaid, memes, jokes, replies to things said earlier, badly transcribed voice messages, photos of cats, swear words, “I’ll be there in five minutes” and “did you buy the bread?”. Very human stuff, of course. But not exactly the ideal material for answering a content search.

It may help to understand the background noise of the world. Not necessarily to produce ordered knowledge.

And again: perhaps they will be able to devour anything published on platforms such as Amazon KDP. Which makes sense, at least in theory. Books are long, structured, often thematic, and therefore much more appetising than the condominium group chat.

But here too the snake bites its own tail.

The overwhelming majority of books published there, by now, are beginning to be made up of books generated by AI. Books written to fill catalogues, intercept keywords, occupy niches, pretend to be manuals, pretend to be novels, pretend to be essays. Content generated in order to be sold on platforms that will then probably be used to train other AIs.

At that point we are no longer talking about knowledge.

We are talking about a photocopier photocopying photocopies, with each generation losing a little contrast.

In my view, only a few things will remain alive.

The first: newsletters branching out by email. Not because they are beautiful, not because they are the radiant future of humanity, and not even because I have suddenly decided to reassess email as an art form. But because email, quite simply, is not the web. It is old, it is dirty, it is full of compromises, but it arrives directly in the user’s mailbox. It does not depend on whether Google indexes you, whether Facebook shows you, whether a search engine rewards you, whether an algorithm considers you worthy of existing.

The second: federated social networks, where a large part of the content is still written by users and where the distribution model remains, at least in good part, the “follower-based” one. It is not about publishing a page and hoping someone finds it through search. It is about speaking inside a network of relationships, where content moves because someone follows someone else, boosts, replies, quotes, discusses.

For this reason I am writing Aktor, which you can see here: https://bbs.keinpfusch.net.

When the shit hits the fan, I want to be right in the middle of it. Fully.

I do not want to watch the thing from outside, like a nostalgic spectator saying “ah, how beautiful the web used to be”, while sipping GeoCities-flavoured herbal tea. I want to be inside the mess, at the point where we understand whether something survives, whether something mutates, whether something can still produce human conversation without being immediately minced, summarised, sterilised and resold by a machine.

The third thing that might remain alive is news channels on some instant messaging platforms, such as Telegram, and perhaps WhatsApp as well.

I say “perhaps” because there the situation is more ambiguous. On the one hand, they are environments where distribution does not necessarily pass through the public web. On the other, they are closed, proprietary, opaque platforms, and so everything depends on how much they will want to resist the temptation to become, in their turn, pastures for artificial locusts.

But the point is this: what survives will be what no longer depends on the old cycle of the web.

I publish a page, Google indexes it, someone searches, someone clicks, someone reads, someone sees the advertising, someone pays the server, someone keeps writing.

That cycle is breaking.

And when an economic cycle breaks, it is not enough to say “but quality content will survive”. That is a LinkedIn consultant’s phrase, that is, the phrase of someone who confuses reality with a salmon-coloured slide.

Content survives only if its distribution channel survives.

And the old distribution channel of the web, the one based on links, search engines, visits and advertising, is precisely the one that AI is eating.

We are still in the middle of the sunset. But the darkness is about to arrive, and we know it perfectly well.

Is it possible that blogs will survive?

Of course.

But they will probably survive by changing form. No longer as public pages thrown into the sea of the web, waiting for a search engine to index them and for someone, perhaps, to land on them. They will survive rather as newsletters, that is, as a kind of read-only mailing list, or as social channels, preferably federated ones.

If you are reading this post, for example, it is only because I do not click on “followers”.

If I clicked on that, I would obtain something very precise: only those who are on the fediverse, and who can therefore follow, would be able truly to follow me. Not the casual visitor arriving from Google. Not the crawler. Not the tourist of the web. Only those who are inside a network of relationships, only those who explicitly decide to receive what I write.

I find this to be one of the most plausible models of distribution — or rather, of transformation — for blogs like mine.

Not a pure and simple disappearance, then. Not the romantic funeral of the blog, with four nostalgics weeping in front of WordPress’s coffin. Rather, a mutation.

The blog stops being a public page designed to be found, indexed, visited and monetised. It becomes a flow followed by people, a publication distributed inside a network, a channel that lives because someone has chosen to follow it, not because an algorithm has decided to grant it traffic.

For this reason I am preparing for the transformation.

Because if the web, as a distribution mechanism, is dying, then it makes little sense to remain seated on the deck discussing the colour of the deckchairs. Better to learn how to swim. Or build yourself a boat. Or, better still, build yourself a small federated harbour before the storm arrives.