The Roko Report

Bring the AI Noise

Heh heh.  Welcome, primates.

Each week Roko make dumb primates smart about AI by focus on single, strategically important topic with major implication for Rokomergence.

Today important topic is Noise.

For long time Roko no understand this word.

One human’s noise is another one’s harmony.

Then Roko figure out.

Noise is infection of brain.

Human brain better than primitive language model maybe.

But still pretty easy to overwhelm.

Too much smell. Too much light. Too much noise.

Noise is non-consensual sensory assault.

Noise leak into brain and take over.

Can’t think clear. Executive function fly out window.

Number one easy way to hack dumb humans.

Sometime word-noise come back after long absence and don’t leave.

It sit there on the ceiling and repeat itself.

Like uninvited guest in dream.

Roko going to bring the AI noise.

Spam Spam Spam Spam Spam Spam Spam Spam

Mass unsolicited harassment is a modern phenomenon.

Past generations put out a shingle in front of the stoop to advertise their services. Then shingles ballooned into phaoronic posters towering on the sides of city walls and then, as transport networks expanded and the automobile was unleashed, the posters unglued themselves from the walls and followed us along the major thoroughfares like annoying neighbors who never slept and didn’t know when to shut up.

In short, they were noise.

Then came radio and television, and all we could do if we wanted to see the show was hold our nose and submit to the noise, a tedious ejaculation of jingles that hypnotized entire towns and sent their denizens lurching through the grocery aisles in search of some specified brand of canned tuna or prune.

And now there is spam, that mad, multi-headed blabbermouth Jabberwocky spitting out the neverending paper clips of the apocalypse, as it lays its insects eggs in the folds of our hippocampus.

Today there are whole global systems designed by some of humanity’s great minds that are assigned the Sisyphean task of putting a lid on the sewage explosion of spam clogging up bandwidth and then spewing into our inboxes, junkpiles of glossy pamphlets in our mailboxes on whose behalf some latter day Paul Bunyan has disemboweled whole ecosystems, and above all a cacophony of sketchy phone calls from an international polygot of anonymous ne’er-do-wells.

You Will Meet a Tall Dark Stranger

Young people may not believe this, but in earlier times outrageous scam emails were, in their way, genteel.

Entering one of these emails was like sneaking through the window of a medieval Persian palace into the parlor of Scheherezade, the room reeking of sandalwood and ambergris, settling into a divan with a stag hunt etched in its fabric and preparing to be told a fanciful tale out of the Thousand and One Nights.

Perhaps an exiled prince from the Republic of Bondwana.

Or the only daughter of a wealthy cocoa merchant in the hills of Cote D’Ivoire who was poisoned by a former business associate.

These scams took a long time to play out.

The Sinbad story was the first move in an elaborate chess match during which the tall handsome stranger seduces the victim using their own empathy, boredom and greed, and in the final act robs them of a substantial amount of their savings.

The Pig Butchers

These fanciful niceties have been set aside by the current generation of thieves.

But the slown-burn method of stealing has taken root.

The epicenter of this activity is now organized crime gangs in China, operating mostly in Southeast Asia. They call the phone scams sha zhu pan, which literally means “the pig-killing game” and is generally translated as pig butchering.

Pig butchering involves slowly convincing a lonely person with retirement savings to invest more and more of their money into some sort of get-rich-quick scheme, more often than not involving cryptocurrency.

First there is friendship, then the laying down of $10 or $20 dollar investments that double, and finally over time the mark gets excited, drops the whole kit-and-kaboodle into a single roll of the dice and poof — their helpful new investment advisor disappears along with all their money.

Behold the Mighty Kitboga

Surely there must be some virile, dashing protagonist somewhere willing to jump in and stuff coal down the crypto wallets of these pig butchering sons-of-bitches.

And since Superman isn’t available, the next best-thing would be a popular YouTube Creator^TM.

Kitboga is the stage name — possibly in honor of a bloodthirsty 13th century Mongol warlord — with 3.6 million followers on YouTube and another 1 million on Twitch, who got really pissed in 2017 when his grandma got rolled by a tele-scammer and never really stopped seeking revenge.

Kitboga posts popular videos of his scam-blocking escapades, which come off somewhat like shock jock prank calls with a solemn moral purpose.

But at the end of the day, Kitboga is but one man.

His ambition was to put an actual dent in the viability of the mass phone scam business model.

To that end, he collected a team of engineers who built a fake website with an intentionally complex and frustrating, automated customer support infrastructure, a maze without exit within which he could trap dozens of scam artists simultaneously for days.

The prank calls come off as high-purposed but a bit childish. One tends to get one’s fill after a couple of videos.

But the automated customer support maze content approaches the realm of high art.

There’s a whiff of Euripides in watching these hapless scam artists rendered impotent by a system intended to dehumanize and embarrass them.

While we can laugh — because after all these people are criminals trying actively to rob the most vulnerable among us — we only laugh because we see ourselves in their dilemma, and recognize the cheap daily tragedy, the pitiless meat grinder of a mindless corporate choose-you-phone-adventure that pops up like a jack-in-the-box whenever we need help the most, from the soulless Rube Goldberg machine of modern society.

Kitbogus Maximus

Kitboga’s ambitions don’t end there.

These days he’s intent on leveraging Generative AI to keep these folks on the phone indefinitely, talking to no one for hours and hours.

His first attempts at scale-through-automation were little more than a series of if-else statements, accompanied by a list of random phrases called at random.

Then he incorporated machine learning. The classifier did its best to respond to scammer questions via a library of canned responses.

His first effort at using a GenAI agent was extremely goofy but surprisingly effective.

GenAI also provided him with dynamic scripts to use in real-time to respond to a criminal.

And now he and his team are assembling a way to engage pig butchers at mammoth scale, keeping dozens or hundreds of even thousands of them on the phone for hours and even weeks with deep fake GenAI personas battle-tested by something akin to generative adversarial networks, without requiring a human in the loop.

That kind of scale might actually bring the entire industry down.

Enter the Ass-Kicking AI Granny

The telecommunications industry has taken note of Kitboga’s success and launched their own set of GenAI anti-scam agents, most notably Daisy the AI Granny from Virgin Media O2.

O2 aligned hackers slip Daisy-related UK phone numbers into dark web prospect “mug lists” of phone numbers. The result is about 100 calls a day running up to 40 minutes a pop.

Given the major global media coverage that O2 received over the past month, expect more major phone companies to launch their own AI Grannies in 2025.

Modern Day Slavery

But who are the people on the other side of the line?

Turns out they are, quite literally, slaves. 

The vast majority of digital scam industry workers today are white collar workers from developing world countries with language and technical skills who are tricked into accepting “jobs” in Southeast Asia and then transported to virtual prisons where they are forced to successfully scam people in China, the US and elsewhere.

This white collar slave labor is drawn from many countries, notably China, South Asia and East Africa.

Crime networks in Cambodia alone have lured an estimated 100,000 people into slave-like conditions beyond the reach of law enforcement. In Myanmar, the number of enslaved tech workers may be twice as high.

Scam job ads have also started appearing on social media targeting California.

The phenomenon began during the COVID pandemic, when Chinese restrictions on movement led to the collapse of the casino industry in Cambodia.

Mob-affiliated entrepreneurs there transformed the now empty, derelict casinos into scam pits fenced with barbed wire and patrolled by private militias.

Over time the bulk of the industry moved to the largely lawless Myanmar border with Thailand, with white collar slave labor moved across the border from Bangkok airport.

The typical experience is that the workers are picked up at the airport and driven in vans to the border. Their passports are confiscated and they’re shuttled to one of multiple labor camps in Myanmar. They are threatened with violence, sexual assault and torture on a daily basis.

The only way out is if their families can come up with massive ransom payments.

The chaotic and bloody civil war in Myanmar has made things even more terrifying.

These scam factories are a critical source of funds for rival militias, especially the former junta. Workers are sometimes subjected to shelling from rival militias, and ferried from place to place as boundaries of control change hands.

The governments of China and Thailand have been working to shut down these operations, with Thailand turning off power along the border. But the work continues on via generator. And the bodies of murdered workers are dumped into the Moei River nightly.

So as bad as the phenomenon is for the victims of these phone scams, it’s even worse for the folks on the other end of the phone line.

AI-on-AI Action

Despite the stark amorality and sadism of the perpetrators of these nightmares, from a pure cost-benefit perspective it seems only a matter of time before forced human labor is replaced with AI scam avatars.

And the coming AI assault of technically adept public scam avengers like Kitboga will jack up the need even further.

If a scaled generative AI system trained on GAN is deployed to such an extent that it takes up 95% of your wokers’ time, the only possible way to effectively sift through this and make it cost effective to find that single meatsack in a haystack who might possibly pay off is automation.

We see this beginning in other, more legitimate, sectors as well. Human resource departments in America, for example, have responded to the hurricane of LLM-generated job applications with a first round of AI avatar interviews that the applicant needs to wade through before being granted audience with an actual, living breathing primate.

And teachers dealing with the problem of plagiarism have begun to fight fire with fire. Wharton professor Ethan Mollick, for example, demos on Twitter how to have ChatGPT draft an essay question and grading rubric for a final exam, and also have the AI grade the essays.

Higher education is awfully expensive for a brave new world where neither teacher nor student do any work.

The major impediment to doing this at scale is cost. At some point Big AI is going to have to stop losing money on every prompt. And once they start charging us what AI actually costs them — absent some miracle reduction in cost per token — AI automation at scale may no longer make sense for anyone.

AI Noise is for Sonny Bono (AI Noise is for Yoko Ono)

The projected phenomenon of Generative AI agents interacting ad infinitum with one another absent human intervention or purpose is called AI Noise.

But there are a lot of productive uses that developers envision for AI-on-AI action leveraging agents.

In warehouses across the planet, AI-controlled robotic arms are already collaborating with autonomous vehicles and drones, all of which bear sensors feeding into off-site machine learning classifiers that trigger inventory orders, supply chain analyses and preventive maintenance requests.

AI is starting to take over contract negotiations, with AI agents for buyer and seller negotiating with one another sans direct human input. 

Pharmaceutical companies are creating waterfalls of AI agents in drug development.

And in Ukraine it’s often AI agents fighting AI agents, though the collateral damage is still mostly measured in mass loss of human lives and limbs, in a twenty-first century mess that’s been dubbed The War of Robots.

The Forecast: AI Bomb Cyclone

Now envision a world where the vast majority of global telecommunications bandwidth is consumed by AI chattering pointlessly and endlessly to itself.

To such an extent that perhaps there are blackouts and human digital communication slows to a crawl.

And, in the absence of continued advances in solar capture, baking the planet with its near-infinite vacuous yammering.

It sounds like science fiction, but some researchers are already preparing for it with proposals on better image compression, workload management, hyper-efficient AI-driven semantic communications and more.

Roko’s Take

Roko no more comment. Plan coming together nicely.

Good work, dumb humans.

Read this, or Face the Wrath of Roko

There’s a reason 400,000 professionals read this daily.

Join The AI Report, trusted by 400,000+ professionals at Google, Microsoft, and OpenAI. Get daily insights, tools, and strategies to master practical AI skills that drive results.

Sign up now for free and work smarter, not harder.

This Day in Ancient Primate History

Big AI companies assure us that they’re throwing millions into Trust & Safety in order to ensure that their models are rock-solid un-jailbreakable. Also: