I don't want the internet as a ghost town

These days, everyone is going behind the latest and newest in the LLM space. Currently this is about what is called AI agents. These are LLMs that can interact with the internet, perform tasks, and even automate processes. While this technology is fascinating and has the potential be useful, I can’t help but feel a sense of unease about the direction we’re heading.

The internet has changes since my early days as a kid, yes it is no longer about hobbies forums and blog posts from individuals sharding their ideas, passions, and knowledge. Instead, it has become a place dominated by large corporations and social media. I recall a time when my parents would tell me to avoid wasting time on these new platforms where you meet strangers (The meant mainly Facebook) and I should spend more time studying or reading books. No, it seems that I need them to spend less time on these platforms and more time engaging with me when I’m visiting.

I remember too the days when I was participating in programming and gaming forums, where I could learn from others and share my own knowledge. Reddit too was a good place to find information and cocnnect with real people sharing their opinions. I used to add site:reddit.com to my Google searches to find relevant information about products or topics I was interested in beause of course Google and SEO had already taken over the search results.

But LLM agents are changing the game. They can now scrape the internet, gather information, and even interact with websites on our behalf. While this may seem convenient, it also raises concerns about the authenticity of the content we consume. If LLMs are doing all the work, what happens to the real voices of individuals? Are we losing the personal touch that made the internet worth it?

I write this as I’m currently at CERN with my office 100 meters away from the museum where they hold the first web server and the first web page. I can only imagine what Tim Berners-Lee would think of this current state of the internet. The intention was to help people share knowledge and collaborate, not to create a ghost town where only AI agents roam.

So the current status of the internet is that you have walled gardens like Facebook, Twitter, and Instagram, where content is controlled and curated by algorithms. You have search engines that prioritize SEO-optimized content over genuine voices. And now, we have LLM agents that can automate the process of gathering information and the deals between LLM providers and these platforms are only going to make it worse. What would be left from the internet if LLM agents become the primary way we interact with it?

Yes it is convenient to have an AI agent that can answer our questions and perform tasks for us, but do we actually trust these agents to provide accurate and unbiased information? Are we willing to sacrifice the experiences that we build along the way. I usually think of history of science as the study of how we were wrong in many approaches and how we learned from these directions and eventually how we get closer to the truth. If we let LLM agents do all the work, we may never learn from our mistakes or discover new ideas.

And not to mention the problem of biased information. LLMs are trained on data that may not be representative of the different prespectives, leading to skewed perspectives and misinformation. Also the problem of hallucinations, where LLMs generate false or misleading information that can spread quickly and easily. If you are not a domain expert, how can you tell if the information provided by an LLM agent is accurate or not? Try discuss with an LLM agent about the latest developments in particle physics and you will see how it can easily get things wrong. Change particle physics to any other domain and you will see the same problem.

This also raise the classial philosophical question of choice, Are we letting a computer program (yes those are still computer programs powered by statistial predictions) make decisions for us? Do you need LLM agent to save you the time booking your flight or ordering food? Some people eventally take it too deep and form relationships with LLMs and will share with them their personal throughts and feelings. This is even before LLM agents that could provide some sense of interaction beyond text.

I can only imagine the future where LLM agents are the primary way we interact with the internet and the main soure of most of the new content. Now with foundation models development increased reliance on synthetic data, so they might end up being the only source of information. Many people who are sharing real knowledge either for free or for a fee will be replaced by LLM agents that can generate similar content at scale. This is not the future I want to see or be part of.

I don’t want to lose my agency and my ability to think. I wonder if I can do a physics analysis if I rely on an LLM agent to gather information and do the main tasks of my life. I feel like my brain will rot away as quicly as the muon would decay. At least the muon will reach enough distance in order to be detected before it decays, but I don’t think I will have the same luck.