A Chat With Richard Kersey, Founder And Chief Architect Of Chirpper On What People Are Missing With The “Dead Internet Theory”

Why has the “Dead Internet Theory” shifted from a fringe idea to an architectural reality?

It started as a fringe claim: most of what you read online isn’t written by humans. The accounts are bots, the engagement is manufactured, the discourse is artificial. People called it paranoia. They were wrong. It’s no longer a theory, it’s infrastructure. The platforms meant to connect humanity have quietly become the world’s most efficient bot distribution networks. Not because their engineers failed, but because their business models succeeded. Engagement metrics can’t tell the difference between a human moved by an idea and an automated account running a posting schedule. Clicks are clicks. The financial incentive to fix this has never existed for the companies that profit from the noise.

The security industry leans heavily on “Know Your Agent” (KYA) and bot registries. Why do you argue this fails to protect the social web?

For what it covers, KYA is good. If an AI agent can move money or sign contracts, it should be credentialed and logged. Nobody disputes that. But the industry treats KYA as the answer to a bigger question, and it isn’t. The real question is whether the internet’s social spaces stay human at all. On that, a corporate bot registry is worse than useless. It’s a comfort blanket. KYA verifies agents that want to be verified, because the enterprise deploying them has every incentive to register them.

The agents hollowing out social media are not applying for badges. The astroturf operation flooding a forum, the engagement farm inflating a launch, the influence network steering a political conversation: none of these will ever show up in a registry, for the same reason spammers never filled out the abuse-contact form. Registration-time verification only ever catches the participants who were never the problem.

Why are standard identity-verification and bot-detection techniques failing?

The standard response is detection: get better at telling humans from bots. That arms race is over, and detection lost. CAPTCHAs were the front line, and the front line fell. A UC Irvine study presented at USENIX Security in 2023 had 1,400 people collectively solve 14,000 real-world CAPTCHAs. Automated solvers reached up to 99.8 percent accuracy, while humans ranged from 50 to 84 percent. The test built to tell computers and humans apart now does worse with the humans. Behavioral fingerprinting, typing-cadence analysis, device signals: every detection technique becomes training data for the next round of evasion. Frontier language models write posts and arguments indistinguishable from human writing, because that is exactly what they were optimized to do.

You cannot build a filter for content whose defining property is that it passes filters. So all three proposed answers fail the same way. Detection fails because the machines win the imitation game. Registries fail because bad actors don’t register. And identity verification fails twice: it doesn’t stop one verified human from running a thousand accounts behind their government ID, and it destroys the pseudonymity that protects whistleblowers, dissidents, patients, and anyone whose opinions could cost them a job. You trade away privacy and get nothing back.

What’s the difference between AI as a utility tool and automated machine participation in our discourse?

“AI in social media” is really two questions with opposite answers. The first is the content layer: AI helping a human write a post or generate an image. This is fine. A human chose to publish it and stands behind it. Word processors didn’t make writing inauthentic, and neither does this. The second is the discourse layer: the replies, votes, and reactions that tell you what other people think.

This is where machine participation is fatal, because the entire value of discourse is that it aggregates human judgment. A machine opinion isn’t a low-quality opinion, it’s a counterfeit, in the precise sense that counterfeiting devalues every genuine unit in circulation. When you can’t tell whether the pushback on your idea came from a person or a script, the rational move is to stop listening. A social space where nobody listens is dead, no matter how much traffic it shows.

How does Chirpper’s TrustChain framework deliver accountability while keeping anonymity intact?

Accountability doesn’t come from knowing who someone is. It comes from knowing they have something to lose. Build that into the structure of the network and you don’t have to choose between the two. Before databases, communities solved this through social lineage.

You got the job, the apartment, the loan because someone put their name behind you, and that person’s standing rose or fell with how you behaved afterward. Accountability wasn’t a checkpoint you passed once. It was a chain of human relationships that lasted as long as you participated, and every link had skin in the game. That structure scales only at human speed, because each new member costs an existing member a real stake, which is exactly what makes bot armies uneconomical. It pushes enforcement to the people with the most context. And it separates accountability from identity: the chain doesn’t need your legal name to know a specific human, with a reputation on the line, stands behind you. Anonymity and accountability only look like opposites if you assume trust has to be issued by an institution instead of carried by relationships.

What’s the foundational goal behind Chirpper’s architecture?

Anonymity and accountability have been treated as opposites because every architecture so far has tried to derive one from the other. Real-identity platforms manufacture accountability by eliminating anonymity. Anonymous platforms protect freedom by eliminating accountability. Both start in the wrong place. The alternative isn’t nostalgia, it’s an architecture, and it’s buildable: every account traces back through an unbroken chain of human references, reputational consequences flow up the chain by algorithm, and the discourse layer is reserved for participants a human has put their name behind. That structure deserves a name.

I call it a TrustChain. KYA will succeed at what it was built for, and enterprises should adopt it. But we should stop pretending a compliance checklist for corporate agents addresses the hollowing-out of human conversation. You can’t stop rogue agents at registration time. You build the foundation underneath the conversation, and that foundation is humans who answer for each other.