The sex videos horseinternet once belonged to cats.

It was that weird period in the late nineties when everyone who was online seemed to be posting cat pictures (believe it or not, My Cat Hates You, the blog that pioneered posting photos of grumpy cats, is still online). It just seemed like the thing to do: Start a blog; post a photo of your cute feline companion; be happy when someone leaves a comment.

There were other things online besides cats, of course. But it was a time when the Internet, then spelled with a capital "I", was a big thing but not an essential thing yet, and no one was completely sure how it’ll turn out.

I was there at the very beginning. Sometime in 1994, my dad connected his Windows 3.1 PC to the internet via a U.S. Robotics dial-up modem, and said, "Son, this here is called the Internet."

You could move files around (very slowly) via FTP, you could search for documents via Gopher, and the best browser was called Cello. I used it to find some weird poetry, early William Gibson works, and lyrics to Tool songs. I knew that the world wide web was going to be huge one day, but back then it was still very empty. The internet of that era wasn’t really for anybody yet; it was a vast frontier to be filled with many things, and then explored.

After that initial era of exploration, the internet stopped being one thing and became all the things. Soon, it became harder to finish a school project or do your work without the internet. And very soon after that, it became almost unimaginable. Business, institutions, and households all went online. Facebook and Twitter came about. Some people realized they were spending too much time on the internet. Some people continued to post photos of cats.

I write this very brief history of the internet because I have a strong feeling that we’re at an end of an era again. For all the good and the bad things it brought us, the internet was always mostly ours. Yes, some of the emails we were getting were automated, Google’s bots scoured the web for info on how to rank web pages on its search results, and some of our computers were turned into botnets that mined bitcoin. But the internet was still mostly designed – or at least it felt that way – for humans to use and explore.

Step aside, human. AI is here to take over

OpenAI recently launched a new version of the AI model that powers its ChatGPT chatbot, GPT-4. It has, for the first time, the ability to understand and process information from images. This opens up some possibilities that were previously unavailable, and I’ve already seen examples of how it will change everything.

Developed by OpenAI, which has grown from a non-profit research lab to a for-profit, artificial intelligence powerhouse, ChatGPT is an AI language model that can generate surprisingly human-sounding answers to a multitude of queries. Chatbots had been around for decades, but ChatGPT was different; it could answer specific and vague questions alike; it could pretend to be someone else; it could write poems and plays in the styles of famous poets and playwrights (go here for a more comprehensive overview of ChatGPT).

The previous major version of ChatGPT, based on the GPT-3 model, broke the internet figuratively, quickly becoming the hottest tech topic and the fastest-growing app of all time.

On the other hand, GPT-4 has the potential to break the internet literally. It's not immediately apparent. When you chat with it, ChatGPT based on the new GPT-4 model doesn't sound radically different from its predecessor. But it paves the way for far more powerful applications.

AI expert Travis Fischer lists some examples of what it can do in this excellent thread. These include building an AI that can interact with all or most of the elements of a web page, as well as create a functional website just by taking input from a sketch in a real-world notebook. The changes in GPT-4 also allow it to understand longer inputs and generate longer outputs, meaning it can, for example, create fairly complex apps on its own.

This Tweet is currently unavailable. It might be loading or has been removed.

Unlike a human, who can do one or perhaps a small handful of tasks at a time, an AI bot is limited by the constraints of the hardware it runs on (and associated costs, but I'll leave those out for simplicity's sake); give it a powerful processor and lots of memory, and it can do the same thing millions or billions of times per day if needed. When we consider the implications of a tool such as GPT-4, we must consider the immense scale it can operate on.

Now that ChatGPT (and soon, similar AI bots) can both usewebpages and buildwebpages, people will start using it to browse the web without them, then to build the web without them and, finally, to build the web designed to be used without them. The first phase will happen quickly; it will take some time until we get to the last one, but I'm quite sure it's coming. At that point, the web will be a giant, impossibly tangled web of data, mostly impervious to humans – unless they use an AI assistant to access it.

Need to do something online? Ask an AI

Let's say that you need to browse through a number of real estate listings in Houston; perhaps you're looking to buy, or you need some real estate data for a work project. Good news: AdeptAILabs has built a bot that can do that instead of you. In a different example, the company's AI bot can perform complex tasks in Salesforce, requiring only a single sentence as input.

• Getting a ChatGPT at capacity error? Tips on how to get past it
• Grammarly introduces a ChatGPT-style AI tool for writing and editing
• MIT's Auxbots aren't your typical modular robots
• Microsoft brings its AI-powered Bing chatbot to a sidebar in its Edge web browser
• OpenAI announces GPT-4

This Tweet is currently unavailable. It might be loading or has been removed.

If you're studying for a law school exam, you might do well to ask ChatGPT for help. After all, it beats most law school graduates on the bar exam, according to this study. An AI-based legal assistant based on GPT-4 already exists, and while it cannot replace a human lawyer, it might soon become an indispensable tool for lawyers.

In fact, AI assistants that can do all sorts of web tasks instead of you already exist, and some are about to become more powerful thanks to GPT-4. Once they become as good as humans at performing certain tasks, doing things the old way – without the help of an AI bot – might feel like a waste of time. And once that happens, the web will probably start changing.

A different kind of web

One direction this could all go towards is a sort of meta-web in which you rarely actually browse the web; instead, you talk to a bot which goes to the web to fetch the things you need. Like a fully autonomous car that no longer needs a wheel, the web might soon start to change and adapt for a landscape that no longer needs to be nice to look at for human eyes; instead, it will need to be easily accessible to bots. In a way, this has already happened – a massive chunk of what's happening online is done by software, not humans – but at least the facade of the human-facing web has remained. In a few years, it might no longer be needed, or it might take a different form, perhaps a summary of sorts, tailored to each individual user's needs and interests.

An early example – a slightly unnerving one – of this happening can be found in captchas, the annoying puzzles that require you to prove you're a human being. Recently, I've noticed a new breed of captcha, one that uses AI-generated images. There's quite a bit of irony in a system that prompts an AI to create images of robots, then asks you, the human, to identify these images accurately, in hope that another AI won't be able to do that. In this equation, the human feels like an unnecessary middleman that will one day be removed altogether.

This is already happening. In a test of GPT-4's skills, OpenAI had ChatGPT pose as a blind person to successfully circumvent a captcha prompt. In fact, it actually convinced a human, a TaskRabbit employee, to solve the captcha instead of him, and provide the answers. The AI passed the test designed for humans alone to pass, and it did it by tricking a human into doing its bidding.

None of this means that ChatGPT is "alive," or in any way smarter than us. It's still merely a tool for humans to use. But the ease and speed with which GPT-4 can perform certain tasks foreshadows a future in which it simply doesn't make sense for humans to perform such tasks. It's easy, albeit a little dystopian, to imagine captchas that are designed not only to fend away dumb bots, but humans, too, allowing only the smart bots to enter. A web that's designed for artificial intelligence to use might have portions that aren't designed for human use; not without an AI intermediary.

There's one thing we should still be able to do on our own, though: Post photos of cats.