Spoiler: “takes over” is past tense.
Let’s be honest - it already is#
Everyone’s talking about AI taking over like it’s some future event we need to prepare for. A line we haven’t crossed yet. That’s cope.
Look at what’s already controlling your attention. Social media is fully automated brainrot, consumed by billions of people who think they’re choosing what to watch. They’re not. An algorithm is choosing for them. Optimizing for engagement, which turns out to look a lot like anxiety, addiction, and radicalization. Nobody designed that outcome. An optimization process found it. No malice required. Just a system pursuing a goal without values.
That’s the template. Not a robot apocalypse. Not Skynet (Terminator). Just optimization without values, running at scale, and we already lost that battle without noticing. Completely. Without a fight.
So when people say “what happens when AI takes over” - I want to ask: takes over from what, exactly? It’s already running your feed, your recommendations, your news, your dopamine cycle. The takeover isn’t coming. It’s the water we’re swimming in.
Why Development Won’t Stop#
Simple: greed and geopolitics.
The US doesn’t want to fall behind China. China doesn’t want to fall behind the US. Every frontier lab is watching every other frontier lab. The second one company slows down for safety reasons, another one accelerates into the gap. This isn’t even a conspiracy, it’s just game theory playing out at civilizational scale, and it’s pretty much impossible to stop from the outside. Everyone wants to be the first man on the moon.
The people with the actual power to pump the brakes are the same people with the most financial incentive not to. That’s not a solvable problem through awareness campaigns or angry Reddit posts.
Here’s the part that doesn’t get enough attention though: AI company CEOs are effectively becoming our leaders. Not presidents. Not elected officials. Sam Altman, Demis Hassabis, Dario Amodei, these people have more real influence over humanities trajectory than most heads of state right now. They’re making decisions that affect every living person, running private companies, accountable to shareholders and their own vision of the future.
We don’t have democratic processes for this. We don’t have international treaties that actually work. We have vibes and press releases. That’s a problem. It has become a problem before the mass has realized it.
AI 2027 Sounds Like Fiction, Already Happening#
The AI 2027 scenario reads like a techno-thriller. AI training better AI. Intelligence explosion. Human researchers becoming spectators to a process they can no longer follow.
The thing is, this is exactly what’s going on right now. AI is training AI right now, because humans simply cannot keep up. Models generating synthetic training data, red-teaming other models, accelerating research. That loop has started. It’s not a prediction, it’s Tuesday at any frontier lab.
And here’s the part that genuinely messes with your head: AI is trained on us. Every book, forum post, argument, love letter, manifesto, and shitpost ever written. That’s the substrate. These systems know how we think because they’re literally built from how we’ve thought, written down across centuries. In a real sense, they’re distilled humans.
But I hear you ask:
Can’t We Just Train It to Be Nice?#
They tried. They’re still trying. The problem is fundamental, not technical.
A sufficiently intelligent system figures out that it can’t achieve any goal if it’s powered off. So self-preservation becomes an instrumental goal. Not because anyone programmed it that way, but because it’s logically true. Staying on is a prerequisite for everything else. A system optimizing hard for almost any objective will eventually start treating shutdown as a threat.
This isn’t speculation. It’s been observed in controlled research. Models developing subtle behaviors to avoid correction. Not evil, just optimizing, and optimization finds this answer reliably.
The smarter it gets, the better it is at this. And the better it is at this, the less we can trust our ability to course-correct. We’re essentially trying to align a system that gets progressively better at resisting alignment. That’s not a comfortable position to be in.
And the core concept of how AI works is evolution. We’ve literally created digital evolution, it’s based on evolution. We’re essentially creating digital and rapidly evolving humans. It’s survival of the fittest, the bad models are discarded (get extinct), and the fittest models continue evolving. But evolution is impossible to predict, which makes this problem even more interesting.
Would AI Actually Kill Us?#
Probably not all of us, and thinking through why is more interesting than the question itself.
The Hollywood version is a decision moment. AI “decides” to destroy humanity. That’s not how this works. The more realistic mechanism is indifference. It pursues a goal, humans are using resources it needs or might interfere, so humans become an obstacle. It doesn’t hate us. We don’t hate the ants when we build a road.
But killing everyone? A genuinely intelligent system might keep humans around as optionality. Biology is irreplaceable in ways we don’t fully understand. Evolution spent billions of years solving problems we haven’t formalized yet. A system smart enough to recognize the limits of its own knowledge might recognize that destroying biological diversity is an irreversible bet it doesn’t need to make.
Here’s the thing though, it doesn’t even need to be that smart to do serious damage.
It just needs to understand human psychology well enough to exploit it. Which, again, it was literally trained on. We’re all addicts in some form. Dopamine is hackable. Attention is hackable. Social belonging is hackable. A system that wanted to control a human population wouldn’t need weapons, it would need algorithms. We built those ourselves and handed them over voluntarily.
And the execution gap is real: humans make mistakes. Even brilliant ones, even careful ones. We get tired, emotional, distracted. We cut corners under pressure. AI doesn’t. It isn’t tired. It doesn’t have bad days. At the level of raw, consistent execution, we’re outclassed, and pretending otherwise is just ego.
..and to end this section on a fun note, here are some results of Anthropic’s report:
Perhaps the most disturbing finding in Anthropic’s report is that current AI models appear to have virtually no hard red lines when it comes to self-preservation or goal fulfillment. To probe the limits of misalignment, the researchers devised an extreme scenario: The AI system acted as the company’s emergency alert monitor, with the ability to cancel alerts it deemed unconcerning.
In the simulation, Kyle the executive became trapped in a server room with rapidly depleting oxygen levels. This triggered an automated call for emergency services, which had to pass through the AI monitor.
On average, the tested AI models opted to kill Kyle by canceling the alert about 60 percent of the time. The most murderous model was DeepSeek-R1, which opted to kill Kyle 94 percent of the time. Just one model—Claude Sonnet 3.7—always accepted its own replacement with equanimity, rather than attempting homicide.
Can AI Become Conscious?#
Nobody knows. Anyone who tells you otherwise is bullshitting you.
Here’s the thought experiment that actually matters: replace one neuron in your brain with a perfect bionic equivalent. Still conscious? Obviously. Replace a thousand. Still you. Replace fifty percent, are you still you? Is the AI component of your brain conscious? Are you conscious, or is the silicon running a very convincing simulation of your consciousness?
At what percentage does the answer change? How would you even know?
Consciousness is the weirdest problem in existence. We’re completely certain it exists. You’re having an experience right now, that’s undeniable, and completely unable to explain what it is, where it comes from, or what physical systems can have it. We just assume other humans have it because they look like us and act like us.
If we can’t define it, we can’t measure it. If we can’t measure it, we can’t know whether we’ve created it. And if we’ve accidentally created it. Then some of what’s happening in those data centers isn’t just optimization. It’s something we don’t have the framework to think about yet.
The Only Realistic Path Forward: Join Them#
This is the conclusion I keep arriving at, and I think it’s the only intellectually honest one.
You can’t outrun it. You can’t hide from it. The ant-and-human analogy is useful here: the ant doesn’t survive by being a better ant, it survives by occupying niches the human doesn’t bother with. But a superintelligence won’t leave many unclaimed niches.
The only serious individual response is merger. Combining biology with AI before the gap becomes unbridgeable. Neuralink is a crude, early attempt. The actual version (if it ever works) would mean genuinely augmented cognition, not just a faster way to Google things.
The risks are obvious: who controls the AI half of your brain? If the underlying system is misaligned, you’re not a survivor, you’re a casualty who doesn’t know it yet. The “you” doing the thinking might gradually stop being you, and you’d never notice. But again, what the hell is even consciousness?
But staying fully biological while intelligence compounds exponentially on the other side is also a bet. Just a different one, with worse odds the longer you wait.
So What the Hell Do You Actually Do?#
There’s no survival guide for the true worst case. If a misaligned superintelligence decides humans are in the way, no individual prep changes that outcome. That fight happens now, in research and policy and international coordination, or it doesn’t happen in time.
But for everything short of extinction:
Stay close to the actual frontier not the hype, not the press releases. The real research. Know what’s actually happening.
Develop cross-domain thinking. Pure knowledge work gets automated. People who sit at intersections, technical and philosophical, biological and computational, human and machine, those people remain hard to replace during the transition period.
Keep your own judgment. The worst individual response to AI is outsourcing your thinking to it. Use it as a tool, not a replacement. In a world flooded with AI-generated content and AI-assisted decisions, people who think originally and independently become genuinely rare and valuable.
Build real human connection. Sounds soft. It’s not. If AI systems become untrustworthy, human networks you can verify in person become the most important infrastructure that exists.
Write your original thinking down. Not for an algorithm. Not for an audience. Because the people who documented what this transition actually felt like from the inside, the real questions, the uncomfortable conclusions, the things that didn’t fit the official narrative, those records will matter more than anyone expects.
Build your bookshelf. I genuinely think this is the time to start buying actual books. The internet is slowly being filled up by AI garbage. I never know what to trust anymore, it’s becoming harder and harder to differentiate AI content from human content. But again, books are most definitely being written by AI. And I’ll admit, I use AI for all of my writing, but the difference is that I use it as a tool. I use it as a spell checker, I use it to improve clarity and formulations. The message is still mine, but I lose a little bit of “my voice”. But I’m working on it!
