Originally framed as a conspiracy theory, dating back to the middle of the last decade, this is starting to emerge as the future of online activity.

The Dead Internet Theory fell into the conspiracy category due to the accusation that governments were manipulating traffic, pushing bots as a mean of human control. On this part, we will defer to others who write about it.

What cannot be denied is the rise in non-human generated activity. This is something we see expanding as AI based application skyrocket.

The reality is AI is appearing everywhere. When I do a search on an engine such as DuckDuckGo, the first response is an AI generated summary. The same is true for a couple other engines I used.

We know automated emails have been a thing for decades. Social media took things to a new level, with algorithms driving activity. This meant that bots, programmed to leverage the path algorithms were pushing things, were used to increase exposure. With the economics of the digital world based largely on clicks, this was crucial.

Things will take another turn when AI agents become the norm.

AI Agents and the Dead Internet Theory

AI agents are not simply bots. Instead, they are sophisticated systems capable of autonomous decisions. It is estimated that, by 2030, much of the online activity could be driven by agents, from moderating content to generating personalized feeds.

The idea appears to make agent ubiquitous. Going back to search engines, these are not only seeing AI used to generate summaries but reply to queries, provide news, and stimulate conversation.

Quite frankly, how can humans compete?

We are dealing with an entity that is only going to keep expanding. While there is great debate about whether AI will become sentient, the fact is that advancement will occur. Even without sentience, AI will see its capabilities improve.

At the same time, these agents can run 24/7. They do not sleep, eat, or take breaks. We also see them accelerating in their speed.

Compute is getting faster. The same is true for communication systems. Even 5 years, there is another level of speed achieved. Humans demand faster transactions. At the same time, the economics of automation mandate that speed is key. Agents that perform more tasks in a given period of time will be more profitable.

The Quest For Inference

Inference is a topic that is starting to get a lot of attention. Over the past few years, I wrote about this. However, more are becoming aware of the potential bottleneck that exists.

The reason this is brought into the conversation is because the growing demand for inference is not coming from humans. We see that people are basically maxed out. Sure, we can add a bit due to new features or services. Nevertheless, humans already spend a considerable amount of time online.

What is making this push is automation. Automated online activity is exploding. We just passed the 3 year mark since the introduction of ChatGPT. Think about all the inference required since that introduction. We know how hundreds of chatbots we can use. There are also image and video generators. If we simply focus on this part of the generative AI world, we can see how inference needs have skyrocketed.

Each time a chatbot provides an answer, inference is used. With models now reasoning, the need is jumping by orders of magnitude.

Jensen Huang, the CEO of Nvidia, is on record as saying inference compute will see a billion-fold increase in the next decade. Consider that for a moment. By 2035, he is predicting we will need a billion times more inference as we have today.

This might seem outlandish until we consider the idea of "real world AI". Some might call this embedded AI. Whatever the title, this is the idea of AI moving from a "box" like a laptop or smartphone into something that moves autonomously.

Of course, the top of the list is headed by humanoid robots, at least for the moment. Then we have concepts like self-driving cars or trucks. If we progress down the list, we can easily see billions of devices in the future needing inference.

All of this is digital traffic. Do the math and we can see how humans are less of the overall. Things are going to be running autonomous in the background, generating traffic that humans don't even see.

The Dead Internet Theory, at least when we look at the human/non-human breakdown, is coming true. Governments being in charge of this might be a stretch but the outcome is clear.

Less traffic is being generated by humans. It is a trend that will only continue.