Have you ever really stopped to think how much of your time online is actually spent interacting with, well not humans, beat? The answer, if you really look into it, it might genuinely surprise you. And it's leading a lot of people to ask this pretty profound question. Is the internet that we knew, you know, that big messy human place, is it maybe? Fading. Welcome to the Deep Dive. Today, we're really going to get into a concept that's quietly gaining some serious traction.
The dead internet theory. This isn't just some fringe idea from obscure forums anymore. No, it's being talked about, even worried about, by industry folks, data scientists. So our job today is to try and separate the panic from the actual reality. We'll use the facts, the data we have, to see what's really going on. Exactly. Our mission here is to explore what the data is actually telling us about AI's, well, kind
of silent takeover. We're going to look at the numbers, quantify this bot tsunami, then we'll unpack why spotting AI content isn't as easy as you might think. We'll look at the economic shifts, how the internet's business model is changing, and then, crucially, look at the proposed solutions. And yeah, there are tricky downsides. It's sort of a shortcut to understanding why the internet feels different lately and what
that really means for you day to day. OK, let's unpack this bot tsunami then, these new AI models like ChatGPT. What was it that made them such a huge game changer? Well, it really comes down to how well they can mimic human chat. These models, especially the ones everyone can access, the large language models, they basically just smashed the Turing test. Ah, the Turing test. That's the classic one, right? Where an AI tries to fool you into thinking it's human. Exactly.
Can a machine imitate a person so well you just can't tell the difference? That's the test, in a nutshell. And before these big LLMs, that was a pretty high bar. Now, it's almost normal. So they crossed some kind of threshold. They got really good at sounding human. And that's what kicked off this machine invasion. Pretty much. Yeah. And the scale of it. I mean, this is where it gets really interesting. The numbers are quite stark. Get this. The Imperva bad bot report for
2025. It found that 51 % of all internet traffic in 2024 was automated. Wait, over half? Over half. Just machines talking to other machines, basically. And maybe more worrying, 37 % of that automated traffic was malicious. That's a record high. So it's a lot of automated stuff, and a big chunk of it isn't friendly. 51%. Wow. That's... That's incredible. It's like there's this whole other internet running underneath ours, just bots chatting. It kind of feels like that sometimes,
yeah. And the speed it's growing is just wild. Look at Cloudflare's data. Perplexity bot traffic. It surged by something like 157 ,490%, almost unbelievable. 157 ,000%. Yeah. And the bots that LLMs use to grab real -time info, those retrieval bots, they grew by 49 % in just the first quarter of 2025. Just Q1. It's an explosion. That's not just growth. That's... Exponential isn't even the word. Absolutely. And think about the sheer
power behind some of these. TechRadar said, open AI's crawlers alone send out over a billion requests a month. A billion? From one company's bots? Yep. And fastly found that those open AI bots made up, get this, 98 % of all the global Fetcher bot traffic in 2025. 98%. Whoa. Just... Imagine scaling to a billion requests a month. Just one AI. That really is a silent invasion, isn't it? But massive. It's like this unseen digital city humming away beneath us. These aren't just numbers,
you know? It paints this picture of the internet becoming, well, mostly a machine conversation. That's a really good way to put it. And it's not just like... reading web pages. Over a third of all web traffic back in May 2025 came from APIs. APIs, remind us. That's like how different computer programs talk to each other, right?
Exactly. That's the short version. So a third of traffic from APIs in these autonomous agents, not people clicking around, and all this activity, it puts a massive strain on the infrastructure. Wikimedia, for instance, they said bots are only 35 % of their page views, but they eat up 65 % of the resources for the complex stuff. So they're disproportionately resource -hungry. Hugely. Wikimedia saw their bandwidth use jump 50 % in early 2025 just from AI scraping. Yeah.
And there was an ARCIDIS study found that for many small servers like 80, 95 % of their traffic is now AI crawlers. 80 to 95%. Yeah. It basically makes it impossible or at least prohibitively expensive for them to run. chokes out the little human run sites. So, okay, with all this machine interaction, this overwhelming bot traffic, what does that actually mean for our experience, you know, when we just go online day -to -day? Well, fundamentally, you'll find less unique human
stuff, just a lot more digital noise. Right. Now, I think a lot of us feel like, oh, I can spot AI content, you know. We look for those slightly weird phrases, maybe delve or tapestry or those super complex sentences. That chart showing the word delve just shooting up in research papers after ChatGPT came out, I mean, that's hard to ignore. But is that confidence really right? Or are we maybe fooling ourselves a bit? That's such a key point. Yeah. Because here's
the twist, right? You're nacked for detecting AI. It isn't really some kind of superpower you developed. It's more that you've gotten really good at spotting one specific AI persona, the chat GPT persona. Ah, okay. So it's not all AI, just the one we see most often. Exactly. See, these models get trained on, like, the whole internet, but then they get fine -tuned. They're fed a smaller set of stuff the creators think
is good writing. Yeah. And that process, it created a very specific, very recognizable voice or persona. So we're just recognizing that style. Precisely. You haven't magically become an AI detector. You've become a chat GPT detector. Other models, like Kimi for moonshot AI, huge in China. They have a totally different way of sounding. And they're probably just flying under our radar completely. And I gotta admit, even I still wrestle
with prompt drift sometimes. Prompt drift. That's when the AI's output kind of wanders away from what you first asked it. Yeah, exactly. Its output subtly changes from your original intention. It's subtle, but yeah. Sometimes I feel like I'm writing to the AI, trying to guess its style, instead of just using it as a tool. It's a weird feeling. That raises a really important question then. What happens when the AI starts learning
mostly from, well, from other AI content? I've heard this called model collapse, or sometimes the feedback loop of mediocrity. What does that actually mean for the quality of stuff online? That is exactly the worry. Model collapse is basically what happens when AI systems get trained on data that's full of their own outputs, stuff they generated before. It leads to this degradation. The quality, the diversity, the originality just goes down over time. So everything starts sounding
the same? Pretty much. We risk creating this closed loop, this information echo chamber, where everything's just a rehash of a rehash. We lose that human spark, that unpredictability. Defined as... AI quality degrades when it learns from its own recycled outputs. About right. That's a good way to put it. Yeah. So thinking about human creativity, original thought, what's the real danger of this kind of uniform AI voice taking over? Well, the danger is that true human
creativity just gets drowned out. Originality gets diluted. OK, so it's not just the content. The actual business model of the internet seems like it's under threat too. Yeah. We're heading towards this future where AI agents don't just write stuff. They do stuff for us. They execute tasks. Yeah. And what's really fascinating or maybe scary, depending on your view, is that optimizing your software just for humans, that might actually be a losing strategy soon. Really?
How so? Because companies will increasingly need to build for AI agents. Think about advertising. AI agents, they're totally immune to all the psychological tricks marketers use. Like, an AI agent doesn't care at all if Sydney Sweeney wears those jeans in the ad. Huh. Right. No celebrity influence. None. It just optimizes for logic. Show me skinny jeans, under $30, locally sourced, rated 4 .5 stars or higher, best durability score, pure data. Marketers use celebrities because
we humans tend to emulate people we admire. Right? And AI. Yeah, ignores that completely. So ads aimed at agents will just be spec sheets, basically. Essentially, yeah. Optimized for machine readable stuff. Right. Price, return policy clarity, shipping speed guarantees, super detailed product specs. So the whole attention economy grabbing human eyeballs that's giving way to this agentic economy, it's about efficiency, verifiable data. And who wings in that world? Well, obviously, the company's
making the best base AI agents. Your open AI is Google's Entropix. But also businesses that have really clean, well -documented APIs. Because that makes it easy for the agents to find and use their products or services. APIs, again, the way programs talk. So having a good robot doorway becomes critical. Exactly. A good robot doorway. I like that. So this shift to an agentic economy, how does that fundamentally change how companies even reach us? Or maybe reach our agents
effectively. Marketing totally shifts away from emotion towards pure, verifiable, machine -readable data. Mid -roll sponsor read. Okay, so the genie's out. We can't just, you know... turn off AI, that's not happening. But if the internet is filling up with bots, we definitely need ways to figure out who's human online. What are the main ideas being thrown around for making that distinction? How do we verify humanness? Well, one big concept is proof of humanhood. P -O -H,
they call it. And there are a couple of main flavors. One is based on blockchain. People like OpenAI, Microsoft, Oxford University have proposed this. Blockchain. Like Bitcoin sort of yeah using that underlying tech for decentralized anonymous personhood IDs It relies on something called zero knowledge proofs zero knowledge proofs. Okay, what's that in plain English? It's basically a way to prove something is true Like your unique human without revealing the actual sensitive
data about who you are. Ah Okay, so privacy is built in that's the idea it creates this kind of proof of human layer over the internet potentially interesting What about other approaches things based on? us, our bodies, biometrics. Yeah, that's definitely on the table too. WorldCoin is probably the most famous example. That's the Sam Altman project, right, from OpenAI. The very same. Their approach involves scanning people's irises. Unique identifier. Ooh, scanning eyeballs. That sounds
intense. It does, yeah. And it raises huge privacy red flags for a lot of people. The idea of a massive global database of everyone's iris scans, potential for mass surveillance. It's understandable why people are worried. It's a big trade -off. OK, so that's proving you're human. But what about flipping it? Instead of humans proving they're human, what if the bots had to prove they're bots? That's the other side of the coin. Proof of bot. And yeah, it's a really interesting
alternative. Cloudflare, they handle a huge chunk of internet traffic. They actually rolled out a system for AI agents to basically declare themselves, hi, I'm a bot. Oh, really? How's that going down? Well, one agent company, Prowzer Base, they already support it. So some are playing ball. But it's super controversial. Gary Tan, the CEO of Y Combinator, he like flat out called it an axis of evil. Wow. Strong words. Why such a strong reaction? The main worry is that it breaks the whole spirit
of the open internet. It could create these powerful gatekeepers, right? Cloudflare, whoever runs these systems, gets to decide which bots are good and which are bad. Right. Who makes that call? That's a lot of centralized power. Exactly. It's a major sticking point. So if the main public internet gets... Well, noisier, more fragmented, full of bots. Where do you think people, actual humans are gonna go? Where will we find those
real human connections? My feeling is that humans will naturally seek out, let's call them digital oases. More private, maybe curated spaces. Like gated communities online. Sort of, yeah. Social media might fragment more you could see more paid forums private discord slack servers Maybe even a comeback for personal newsletters and blogs, you know quality over quantity platforms like medium or sub stack They seem perfectly set up for this places offering high quality.
Maybe longer form stuff away from all the noise It's like moving from the chaotic town square into smaller quieter cafes where you know the regulars. Mm -hmm So looking at all these paths proving we're human making bots identify themselves, retreating into these digital oases. Is there one that seems, I don't know, most aligned with keeping that genuine human connection alive online? Honestly, each one has pretty big trade -offs. It's always going to be a balance between open
access, privacy, and who holds the control. There's no easy answer there. OK, so let's try and bring this all together. What's the big picture for us, the people using this network every day, this dead internet theory? It sounds grim, but it feels like it's not just a future prediction anymore. In many ways, it kind of feels spot on right now. The internet does feel different, sometimes broken. Less about finding unique voices, more about wading through an echo chamber. I
think that's fair. Yeah. We've seen the numbers and the insane bot traffic. We've talked about how content is becoming more uniform because of how AI learns. And we've touched on how the basic economics of the online world are shifting under our feet. The real struggle I think is how do we preserve I don't know, the soul of the internet. How do we make sure we can still find and value that real human connection if it's all the machine noise? It's definitely not
about killing AI. That's impossible. It's about building new systems, finding new norms so humans can thrive alongside these incredibly powerful new tools. That leaves us with a pretty big thought, doesn't it? If our digital world is increasingly becoming a conversation between machines, what actually happens to our voices? are authentic human voices? How do we make sure they don't
just get lost in the noise? How do we ensure they're heard, and maybe more importantly, valued as something distinct from all the synthetic stuff? It's a huge question. And honestly, it's something for all of us to really think about as we figure out how to navigate this evolving landscape. Well, thank you for joining us on this really fascinating deep dive today. My pleasure. Until next time, keep exploring, and definitely keep questioning. Outtifro Music.