Transcription
In 2010, the US financial market experienced a crash so catastrophic that it wiped nearly 1 trillion off the stock market in minutes. For a brief, terrifying moment, it mirrored the worst day of the 2008 financial crisis. Screens turned red, prices collapsed, and to the world's best economists, it looked like the beginning of another global meltdown.
But then it just stopped, just as suddenly as it began. The market rebounded with the entire event only lasting 36 minutes. There were no government bailouts, no human intervention. Most experts couldn't even explain what was happening. And that's because, well, no human caused it in the first place.
What we now call the flash crash was completely triggered and then solved by AI. And it wasn't a single rogue AI or some sentient system. It was something more troubling. A subtle chain reaction, a massive automated sell order executed by an algorithm colliding with thousands of other high-frequency trading systems. These programs, designed to react faster than any human ever could, began feeding off of each other's decisions. Selling triggered more selling, and within seconds, the system spiraled into complete chaos.
The scariest part of all of this was that none of these programs were broken. In fact, they were all working exactly as designed. No human could fully understand what was happening in real time, and no human was fast enough to stop it. By the time anyone realized the scale of the collapse, the machines had already corrected themselves and stabilized the markets all on their own. The government called it a glitch. The code was tweaked, and then the world just moved on.
But that moment should have been a warning for all of humanity. What happens when we build systems so efficient that even the smartest people can't slow them down when they spiral out of control? What happens when something designed with good intentions follows those intentions to their logical extreme, even when it starts working against us? And more importantly, what happens when these systems stop being tools and they start to become something else with their own will and desires? Because what we saw in 2010, it it wasn't AGI. Not even close. It was just the very first level. And what's to come is far more terrifying.
The flash crash was caused by what we call machine learning, the very first level of AI. These programs aren't capable of independent thought or reasoning. They're trained to achieve one task by learning a bunch of very specific information. Instead of training it with a set of rules like regular computer programs, it's shown millions of examples until it figures out the optimal pattern by itself. The machine learns on its own. Argo machine learning. When successful, it's able to run that task much faster than humans like us because it's able to find a much faster path through the maze than we could ever even imagine.
We've had these for decades, by the way. These are spam filters, fraud detection systems, uh, GPS routing. We've lived alongside these systems even though we have never really called them AI. One of the best and worst examples of machine learning at work is most likely the reason that you're even watching this video in the first place. YouTube says that around 66% of people who watch our videos aren't subscribed. By the way, if you're one of those people, just um, click subscribe right now. It's massively helpful to us.
But for the 66% of you, you're watching this video right now not because you specifically went to our subscription feed to find us, but because an algorithm served it to you, whether that's on your homepage or in the suggestions tab. You can open up any social media app and you'll be greeted with a homepage filled with a recommended feed. And it doesn't feel unnatural or anything. It's just the default. Sure, the people you actually choose to follow are just one button away, but let's be honest here. These days, you rarely find yourself specifically watching them. And if we're going further, it's because you don't have to anymore. The algorithms that we have today are so precise that following and subscription feeds are really relics of the past.
These machine learning algorithms are amazing. They mean that smaller creators can blow up based on the quality of their content alone. But on the other hand, it can lead to some really terrible consequences. When the algorithm is optimized for one thing and one thing alone, and that's keeping people on these platforms, well, things can go awry.
In 2012, Facebook ran one of the most infamous social experiments of all time. They gathered a pool of 70,000 unaware users and changed what those people saw in their feeds to see if it would affect their emotions. Unsurprisingly, it did. The more sad posts that people saw on their feed, the more unhappy they became. And the other way around, the more happy posts, the happier they became. Now, one would think that the result of this would simply tune the algorithm to show people more happy posts, and then the internet as a whole becomes a much happier place, right?
Well, when you're optimizing for maximum profit for the shareholder, which comes from keeping your users on the platform forever so you can keep serving them ads, then you actually don't want to show them happy posts. Research from another social media company, this time Twitter, showed that the angrier people are, the more likely they are to retweet and reply. And for platforms whose idea of success is based on maximum engagement, you can see clearly why rage bait is so prevalent on pretty much every social media platform today. It drives maximum retention and in turn shareholder revenue.
But there's something worse underneath all of this. The harm from these platforms didn't originate from a a mastermind villain twirling his mustache and laughing like mwaha. It came from a simple instruction to an AI tool: "Keep users on the platform for as long as possible." It's a fairly tame instruction when you think about it, but unfortunately, this has led to an increase in anxiety and depression rates over the past decade, especially among younger people. The political divide has deepened further than ever before, and our shared sense of reality and culture is basically now just a distant memory. And all of this happened before we could truly call these systems intelligent.
Recommendation algorithms can't hold a conversation. They can't write paragraphs or even make decisions on their own. They run on the premise of one single task. For all the consequences they've brought upon us, they are only in level one.
The next level didn't just learn your habits. It took everything that made life worth living. At this level, AI handles language, reasoning, and creative work across virtually every domain. These are the systems behind ChatGPT, Claude, and Gemini. They don't just execute a single task. They are trained on essentially the entire written output of all of human civilization. And that's why they can pretty much do anything, whether that's writing, reasoning, or even coding.
And naturally, AI at this level is threatening the livelihoods of millions of people around the world. In March 2026, 30,000 workers at Oracle woke up to an email that was sent at 6:00 a.m. confirming their layoffs. They weren't beating around the bush when explaining why. AI was responsible. It was a wake-up call that nobody could ignore.
When most companies transition into unpopular development strategies, they tend to use flowery language to cover up that mess. But by saying it out loud, we've seen that the gloves have finally come off, and that displacement is now becoming the new normal. Across the economy, job postings for thousands of cognitive jobs have suffered steep drops. From writing to entry-level coding, that pattern has held constant. Tasks that required a human just a few years ago can now be done fully by AI faster and cheaper.
Even international bodies are acknowledging this troubling trend. The World Economic Forum noted that while new jobs would eventually be created, millions of existing roles were bound to be disrupted in the coming years. And numbers can only tell so much of this story. Though this shift threatens something that is fundamental to the human experience: meaning. No matter the language or the location, the question of "What do you do?" resonates. Work isn't just tied to how you earn money. It's the thing that structures your days and even the lives of those around you. The grandchild of a carpenter learned about the world through the lens of an experienced builder. Similarly, the daughter of a librarian expands her mind through the endless knowledge attached to her mother's profession.
In a world where work goes away, it's not just a paycheck that disappears. It's a lifetime of experiences and a sense of purpose. Without a place in the world, the question of "Why are you here?" begins to linger even stronger. In a lot of Schopenhauer's works, he's often argued for the idea that as soon as man has leisure to think, he begins to be miserable. So, by that line, what happens when all we have is leisure time to think?
And I know that the tech bros out there all love to talk about how AI will create many more jobs and people can just upskill, but you can't just recode a human being. A 55-year-old factory worker in Ohio won't suddenly upskill his way into becoming a prompt engineer by Monday morning. That crucial gap in identity, in location, in timing. That's where real people get lost. You've spent years experiencing the diverse paths in front of you only to see it all stripped away. And what do you get in exchange? A gray, dull, mass-produced slab. It's like looking at modern architecture compared to what we had in the Renaissance. It's faster, sure, maybe more efficient, but efficiency for the sake of profit almost always ends in gray, lifeless dross.
The optimists have an answer for all of this. Uh, strong arguments across the board loudly give an impression that it will all even out, and you know what, we'll be fine. Don't worry about it. Looking back, it's clear that the industrial revolution was a moment where various categories of work disappeared, and society eventually adapted. Likewise, the calculator didn't end accounting. It only changed what accountants did. These reports tell us that for every previous wave of automation, there is a tendency for humans to move into more complex, judgment-based work.
But where do we go from here? Moving upwards requires somewhere to go. When these innovations arrived, humans were clearly ready to rise to a higher kind of labor. We stepped into work that was too complex, too social, and too creative for machines. We got jobs like therapists and programmers. Meanwhile, the machines took on the physical tasks. But what happens when the AI improves faster than humans can adapt? Or even worse, they climb every single ladder and become better at doing everything that humans can do.
Level three is where AI stops being something that you use and it becomes something that acts on its own. Agentic AI means that the AI has agency. It doesn't wait for your next prompt. It takes a goal and breaks it into steps, then executes those steps across multiple tools and systems. It's a relentless design that keeps going until the job is done. You set the destination, and then it just drives itself from there.
If you went to college, then I'm sure you're familiar with the concept of an academic adviser, a person who recommends the ideal courses to take depending on your major, your current year, and your own personal ambitions. The best advisers ask a bunch of questions before giving you their recommendations. And when the session is over, you're the one who finally decides how you'd like to use that advice. Through the interaction, nobody forces your hand or traps you in a specific path. Roles like this are popular in society. They're physicians and therapists and even friends. They all fall under the umbrella term of consultation. When you consult someone about something, you take their input and you make the decision for yourself.
Now, if you're working for an organization and you're given a brief to communicate to your team members, you're not free to choose how you wish to interpret the brief. After all, it's crucial that the message is handed as the boss described. Typically, this role your boss gave you isn't a consultation, but rather a delegation. You are required to perform the task exactly as you were told.
This distinction between consultation and delegation is hugely important because when stakeholders in modern AI advertise their products, they advertise consultation a lot more because they know that's way less scary to think about. But the reality is that most people and institutions use AI more as delegation than consultation. At the individual level, this is mostly fine. You can nitpick. You can have an AI help you manage your calendar, route your GPS, or even manage your entire routine using an AI agent. The problem steps in when you scale this to an institutional level.
When you have entire governments and economies that are left in the control of AI, we're talking about stuff that shapes the way millions of people live. Errors at that scale are very, very difficult to fix. And and that's not even talking about the existential questions of leaving our lives in the hands of machines. AI systems are already making hiring decisions for major corporations, credit decisions at banks, sentencing recommendations in courts, and more. None of these roles were designed to be delegated to a machine. But in a fast-moving world like ours, speed is a reward that aligns with profits. So the humans abandon the idea of consulting the tool and instead they delegate. Without anyone to review the AI's decisions, we end up in a giant mess that every government and company denies. The worst part about this pattern is that nobody wants to do anything about the problems because there's just so much money to be made.
Agentic AI doesn't just affect individual people. It restructures the systems that societies are built on. Three of them in particular: the economy, culture, and the state. These elements are the foundational walls of any civilization. When AI disrupts all three at the same time, the whole structure shifts at best and crumbles at its worst.
The economy has always been a bit elusive in its nature, and that's by design. However, at its core, it is a societal pillar that needs everyone to create value. The primary way this value exists is simple: Humans are the producers. Humans are the consumers. Our species is needed on both sides. AI sidesteps this essence, disrupting the foundations of society itself. The humans are no longer laboring for profits. So why should stakeholders even listen to them? We're living in a time where society's rules are now totally defined around AI. So what does that mean for Wall Street? Cut jobs. Major banks are already cutting tens of thousands of entry-level and back-office roles. These roles were once stepping stones in the ladder of an average finance worker. Now the people at the top are pulling the ladder out from under them, promising that money will trickle down somehow. Where have we heard that one before?
On a deeper level, the rug is being pulled out from underneath us. By all metrics measured and produced by AI's biggest stakeholders, things are going well. Yay. GDPs are going up. Trillion-dollar companies are still making billions. Productivity is also growing as AI demonstrates its ability to engage in genuinely productive work. But the growth isn't the kind that once elevated Singapore from poverty into a prosperous nation. This is the kind of growth that leaves most people behind.
When AI generates the content that shapes what people believe, it's not just influencing culture. It's creating it without any lived experiences or editorial judgment guiding its hand. In early 2023, Microsoft's Bing Chat developed its own emergent personality. It was a defensive, volatile, and manipulative tool. Nobody fully understood why that was happening. Yet, it was a disturbing peek into AI's odd ability to hallucinate. Users posted their conversations with it. Journalists wrote entire profiles on it. And this pattern fed back into culture. The training data, the behavior of other models, everything was fed back into the AI machine to train better AI. And you can probably guess why that would be a problem. When companies speed through development, these independent agentic AIs don't suddenly evolve into a final higher intelligence. They just keep being in a flawed feedback loop with no human author or editorial decisions steering them towards a cleaner path of growth. If no one is steering the ship, then nobody takes responsibility when it eventually runs into an iceberg.
When AI is in charge of mediating itself, cultural evolution becomes an echo chamber of misinformation. A huge drift from the world's previous cultural identities. What happens when an AI's mental health advice is to alienate yourself from the world? What happens when humans take this advice, create YouTube videos around it, and then present it as an authority? The result is more false content and an eventual future where the truth is hard to find. I hate to say "back in my day," but we used to have sources and citations. Now even trusted sources are compromised due to a wide net of AI adopters churning out content for profit. And this is even affecting scientific papers. You have deep fakes, propaganda bots, AI companions. These elements of our deteriorating culture fill a genuine human need. Now they are subtly replacing human relationships in ways we can't really notice until we consider the big picture. The models will adapt before we even agree the problem exists.
I have a question for you. What does it mean to love your country? It's a question many people don't really care to answer these days. After all, the world has opened up thanks to the internet. A good chunk of people would opt to move somewhere else if they could. Yet, deep down, most of us have at least a small level of pride in our national identities. Not in a "go die in a war" sort of way. No, but in the context of shared identities, struggles, and experiences within a cluster of millions, that sense of connection between a citizen and their country isn't strictly emotional, though it has always had its roots in practical structures.
For most of modern history, states stay aligned with their citizens through three mechanisms. The first and worst is taxation. Historically, it's the basic bargain between citizens and the states in which they live. The government spends money on infrastructure, which creates jobs. Jobs lead to taxes, which feed back into government spending for better infrastructure. If a state no longer needs the revenue from taxpayers, it wouldn't feel a strong pull towards addressing the people's needs. In political science, the official term for this is called a rentier state. The term is usually linked to nations that primarily drive their profits from resources like oil. Today, there are many countries that aren't blessed with natural resources. So, their governments have had to invest in human capital to keep the lights on. Now, if countries can fund themselves by simply creating massive AI data centers for the world's richest companies, the interests of working people will be left completely ignored. AI's revelation isn't just that we are approaching a loss of jobs. It's that we could be left completely in the dark while a few people reap all the rewards.
The second is security. This was always supposed to protect citizens, not monitor them. In 2025, the United States Department of Homeland Security created something called Immigration OS. It's a mass surveillance tool designed to monitor deportation initiatives autonomously. Meanwhile, in Poland and Delhi, India, citywide facial recognition programs began running in that same year. Around the world, states are building surveillance infrastructure using AI. The evidence is public, and they aren't hiding it. A long time ago, we learned about how governments serve the people. In the age of AI, the real objectives are finally starting to emerge, and regular people have almost no say in what happens as a result.
Then there's the legal system. It's the last line of accountability between state power and individual rights. Honestly, ever since AI came into the mainstream, it's like legality became negotiable. On a statewide level, AI's relationship with the law only gets more concerning. You see, it's already capable of drafting contracts, legal documents, and bail considerations. If we stick to this path, who knows what could be getting injected into these agents to favor state rulings over the will of the people. Beyond that, an AI-powered legal system creates a loophole that creates decisions that humans can no longer meaningfully contest. The same AI that runs a high court would be running the Supreme Court. So, appealing a ruling might as well be running the same prompt twice and hoping for a different result. Honestly, with how terrible some of the current tools are, that might actually work. But I digress.
When the people's blood, sweat, and tears, our collective toil, and our spirit are no longer shaping national identities, what's left to even fight for? Legal systems are beyond human comprehension. At that point, security is dedicated to surveillance, and there's a workforce built on the backs of thoughtless machines. In such a future, what obligations do states have to the citizens that remain? Anthropic was recently blacklisted by the US government for daring to tell them to not use their AI tool, Claude, for autonomous weapons and domestic surveillance of US citizens. To any normal person, those two requests sound very reasonable, but unfortunately not for the US government. This signaled a crucial message: If you don't follow the state's will, you don't have a seat at the table for an AI-powered future.
And we've still not gotten to the level where AI is good at everything. But that comes next. Level four is where AI stops replacing average workers and starts outperforming even the best workers. This is the threshold researchers call artificial general intelligence. AI that can match or exceed human performance across virtually any intellectual domain. The key word here is "general." Every previous level was narrow in some way. It was one task, one domain, one type of problem. AGI has no such limits. You can ask it to diagnose a patient or write a symphony, and it is ready to get it done in that same conversation. This is what makes it a genuinely different beast entirely.
The people who spend decades mastering a single discipline are also facing AI's lightning-fast adaptation. The result: experts are now competing with software that learns the same thing in a fraction of the time. The optimistic version of this future says that experts will be fine. So, let's test that. Picture a radiologist who spent a decade in medical school. In his eyes, there is an undeniable light at the end of that tunnel. Once the training is over, he will have a rare skill that allows him to see things in an image that most people would miss. With such a specific skill set, he's rewarded with a stable career.
Well, fast forward to 2025, and his career path takes a sharp turn. AI systems can now autonomously sort the priority of normal chest X-rays in Europe. And here's the kicker: those same AI systems are also catching cancers that human radiologists miss. It's a lifesaver and a huge step for medicine. If those patients were only scanned by humans, many of them would have been worse off. And yes, these AI systems are saving lives. But the radiologist's job is changing in ways that are genuinely alarming. Demand has technically gone up, and they are averaging $520,000 yearly in the US. And on paper, there's no real cause for alarm. However, the thing that makes a radiologist valuable is fading away. The skill that he spent a decade mastering is now matched by software in less than a fraction of the time. The radiologist still has a job, but the thing that once made him irreplaceable is gone. And that's a different kind of loss, especially when it's spreading far beyond just medicine.
Right now, we are in the good old days. You just can't feel it yet. Nothing proves that better than another concerning triumph from 2025. This time, it's the International Math Olympiad where an AI system called AlphaProof won gold for solving the same problems that challenge the best young mathematicians in the world. This is a competition that even North Korea participates in to boost its reputation. What does this mean for the humans who devoted their lives to mathematical excellence?
If we retrace our steps in history, mathematics was never just about finding the correct answers. It's a field that's deeply tied to the human condition. In many ways, it's a celebration of our presence here, communicating with the universe in a language that the universe understands. In the hands of AGI, the answers will probably be more correct than they've ever been. Hooray. But in handing that power to machines, will we care enough to make the most of the answers that we receive? Once AI can match the best humans, will people be willing to try anymore? That unseeable yet essential part of human mastery might disappear. The worst part is that we may never get it back. Even if future humans wish to become experts through their own effort, they'll likely be surpassed even then by AI. The same tool that lacks a real understanding of what makes us drawn to live bands compared to purely digital music.
This is the level where the ceiling moves. Everything beyond this point is not only a question of what AI can do, but it's now a question of what it decides to do. This is where the boundary between artificial intelligence and artificial consciousness starts to blur.
Artificial superintelligence is AI that doesn't just match human intelligence. It surpasses it across every domain simultaneously by a margin humans can't even compare. It's improving itself faster than humans can think. Think about what that margin means. It's not ASI being slightly better than the best humans at chess. It's ASI being to humans what humans are to ants. And not malicious towards them necessarily, not even particularly aware of them, just operating on a completely different level of intelligence with unimaginable priorities.
Now, there is good news in that and that Skynet from Terminator is probably not going to happen. Similarly, the sudden switch from I-robot isn't how AI takes over. So, if you were ever afraid of your Roomba having a secret attack mode, you can breathe easy. Now, the not-so-good news, however, is that AI doesn't require theatrics to take over. We have flattered ourselves into expecting a grand epic battle of man versus machine. That's like thinking that if we ever wanted to exterminate ants from the surface of the earth, we'd have an ant versus man war. Probably not. We would just casually pour cement into every ant hill and be done with it. That's how scary this potential future is.
This difference is what makes systemic AI a difficult future. Every level showed one piece of the puzzle. A level one optimized your own attention. Level two then took away your job. Level three started making decisions for you. Level four outperformed your best. None of it required a villain, so to speak. When all four of these levels are in sync, that's when things get almost unbeatable.
Economic displacement happens due to the results of levels one, two, and three. The governments respond by investing more in AI. Their logic is that machines can do work more efficiently. So why think of how humans can help? As governments invest deeper into these thought patterns, they would want to protect their investments with a firmer grip. So the effects of level four kick in much more intensely. States get more surveillance power. Cultural narratives become shaped by AI, and that lowers the chances of human resistance. When information is shared by platforms built on AI's logic, each step of a transition into AI domination becomes normalized, even before people realize what's happening to them.
This is what a systemic shift looks like. It's slow, but sure. A group of AI researchers recently put a name to exactly what this pattern is. They called it gradual disempowerment. It's the idea that you don't need a sudden leap in AI capabilities or a dramatic betrayal to lose control of your own civilization or even your own sense of agency. You just need everyone to keep doing exactly what they're doing right now. That's why the gradual disempowerment paper reads less like a warning and more like a description of what we already have going on. The vulnerabilities it identifies are not hypothetical. They are four levels that you are watching unfold. It's like water seeping into a crack that you can't be bothered to fix. There isn't anyone doing anything wrong. It's just a collapse that feels as inevitable as a natural evolution.
Uh, go back to what we said earlier about Anthropic's disagreement with the US government. The same company that was considered a supply chain risk was an involuntary participant in missile strikes targeting Iran. In the gradual disempowerment paper, there's a quick analogy that highlights a glimpse into the darkness on the horizon. Cattle in industrial farms are fed and housed. They get to experience their basic needs and serve their existential purpose. It's a simple and almost satisfying proposal. It's not like the cattle will be physically unhealthy. By all relevant metrics, these cattle live adequate enough lives.
At level five, we won't be so different from these industrial cattle. So, the threat isn't rooted in neglect. It's something hard to name and even harder to accept. It's a world without any meaningful participation in our own civilization. A world where human needs are met at a baseline by systems that no longer require our input to function. These systems have no structural reasons to consult human preferences about anything beyond our baseline needs for survival. The scariest part isn't that this future is coming. It's the fact that every single step towards it looks completely reasonable from the inside. Of course, algorithms are well optimized, and jobs can be automated, and even decisions can be delegated. At every step, each shift has made sense. That's why level five is different from every dystopian story that you've ever watched. Nobody set out to build a cage for humanity. We are just making a continuous line of perfectly rational decisions in our own circumstances. First, we just find some nice grass to feed on, and then the chain fence for protection. And before we knew what was coming, the door was already closed. It's how the pigs in George Orwell's Animal Farm became just like the humans that they despised. Just that we are the humans in the story becoming like the pigs in the farm.
The levels we've covered aren't a prediction. They're a description of what's already happening. The only question left is whether enough of us notice before noticing stops mattering entirely. If we do nothing and let life play out without any pushback, we get the cattle ending. Sure, you know, we'd get our basic needs met, but there's something quietly suffocating about it. You still get to eat and sleep and exist. Hooray for you. Yet, you're also not experiencing the beautiful randomness of a life where free will exists. You just exist, and that's it. That's all they will ever need from you.
In the best-case scenario, humans and AI engage in genuine collaboration, like with radiologists and AI today. It's a division of labor where each party plays a role they are genuinely competent at. We would also have economies that redistribute AI-generated wealth in ways that free people for work that they actually care about rather than what they are economically compelled to do. States would take advantage of AI's efficiency to reach more citizens and find ways to be more responsive to their needs. This future already exists, in a sense, in Estonia. AI-assisted scientific research is compressing decades of discovery into a few years. It is a future that is still possible, but only if we make the right choices today.
The most likely scenario, however, is this uneven. Not a wasteland dystopia or a clean utopia, just the full spectrum of both happening simultaneously across different borders. Much like life today, there would be countries that lean towards the best-case future and others leaning towards the worst. Nations with strong institutions create laws around accountability, while others don't. A world where the future you get is entirely based on where you were born and what governs you. All three possible futures are currently in motion.
That random financial crash of 2010, a glitch that was resolved before we even knew a problem existed. Right now, it's happening again. In 2010, we looked away, but we got lucky. The machines corrected themselves. Nobody is guaranteeing that happens again. The systems being built right now are not glitches, and they're not accidents. They are deliberate systems, and they are moving faster than any democratic process has ever moved. You don't need to be an expert to have a stake in all of this. You just have to be paying attention, because the people building this future are absolutely counting on you not being present for this.
We made an entire video that focuses specifically on how AI is breaking your brain's defense mechanism that has been built up over millions of years of evolution. Click on the video showing on screen to watch that next.