18 AI Futures. You're Already Living in One.

Usually when we talk about a massive technological shift, you know, there's this expectation of a singular cinematic moment. Right, like a clear turning point. Exactly, like an engineering breakthrough where a scientist in a white coat holds up a glowing microchip to a room full of reporters and says, well they say, we did it. Yeah, it creates a very clean binary, pre-revolution and post-revolution. Before the internet and after. It's clean, it makes sense, and honestly, it's just comforting. We like history to be categorized with clear dates and dramatic headlines. We do, but it rarely actually works that way. It really doesn't. And today is Tuesday, October 6th, 2026.

AI development right now feels like it's moving at absolute light speed. And when you step into the world of AI integration and human psychology, that cinematic narrative just breaks down completely. It really does. It shatters. Looking at a landscape of the future that is murky and constantly shifting. So let me ask you a question. When was the first time today you let an AI make a decision for you? Oh wow, probably before I even got out of bed. Was it drafting a reply to an email, summarizing a 20 page PDF, or you know, maybe letting an algorithm suggest the fastest route on your map app?

You probably didn't even notice you were doing it. And that is exactly our mission today for this deep dive. Yeah, this is a fascinating topic. It is. We are unpacking a paper by Dirk Werner, a psychologist and psychotherapist, and it's titled 18 AI Futures You're Already Living in One. And the focus here isn't just the underlying code or the hardware. It is a deep look at the psychology behind how we are voluntarily surrendering our decision making, day by day. OK, let's unpack this, because we usually think of the AI takeover like a sci-fi movie alien invasion. A giant ship hovering over the city, shooting lasers, demanding our submission.

Right. The classic Hollywood scenario. Exactly. But Werner suggests it's more like a leaky faucet. It happens one drop at a time until you wake up and realize you're completely underwater. That's a great way to put it. And to understand the mechanics of that leaky faucet, we have to look back at the alien invasion paths we thought we were on. The older predictions. Yeah. The narrative we've been operating under for almost a decade actually stems from 2017. Physicist Max Tegmark published a hugely influential book called Life 3.0. Oh, yeah, I remember that one. In it, he laid out a map of 12 possible futures, basically 12 end states for a world where machines become smarter than us.

And these are some wild scenarios, right? They range from total utopian eradication of disease and poverty to our literal violent extinction. They really cover the whole spectrum. Tegmark broke those futures down into three broad categories. Futures where we live alongside a superintelligence. Futures where we disappear. And futures where we manage to prevent superintelligence from happening at all. OK, so three main buckets. Right. And some of them are incredibly nuanced. What's fascinating here is he describes one called protector god. Protector god, wow. Yeah. This is a scenario where a nearly omnipotent AI secretly ensures our well-being. It manipulates global events, economies, and resources so subtly that humans still believe we're in control of our own destiny.

That is incredibly eerie. It is. And as a psychotherapist, Werner likens this to the perfect therapist who guides a patient to a breakthrough without ever letting the patient realize the therapist was steering them. It's like living in a terrarium that's being perfectly temperature controlled by an unseen hand. But the extinction scenarios aren't all just Terminator robots storming the beaches either, are they? No, not at all. The most famous fear is the brutal one, certainly, where the machines see us as a waste of resources and wipe us out. Right. But Tegmark also described a gentle extinction called descendants. Gentle extinction, that sounds like an oxymoron. It kind of is.

In this future, the machines replace us. But the transition is so gradual and the machines are so clearly our intellectual superiors that we actually view them as our children. Oh, wow. Yeah, we become overwhelmingly proud of them. We see them as our legacy. And we gracefully, voluntarily step aside and let humanity fade out. Werner calls it an extinction with a farewell party. An extinction with a farewell party. That is, I mean, that's heavy. And then the third category is where we just prevent all of this, right? Maybe a massive global surveillance state bans AI research entirely or human civilization collapses from nuclear war before the computers can get smart enough.

Exactly. But here is the thing. All 12 of Tegmark's original scenarios revolve around one single defining premise, the tipping point. The cinematic moment. Yes, the singular moment when a machine crosses the threshold of human intelligence. Tegmark wrote the book as if humanity would have this distinct window of time to prepare for that specific moment. Look at the options and steer the ship. OK, I have to push back a little here, because this sounds like waiting for a supervillain to announce their master plan on global television. Behold, my intelligence has surpassed humanity. Right, with a giant laser behind them. Exactly. But in the real world, do defining moments of technological takeover actually exist, or is that just how history books simplify things later for high school students?

That is the core flaw, Werner points out. Tegmark was writing before GPT-2, let alone the models we have today. Right, ancient history in AI terms. Pretty much. The reality on the ground now, in 2026, is that the window to prepare is vanishing faster than our brains can physically process. The curve of advancement is moving at a speed that breaks human intuition. OK, let's talk about the mechanics of that speed. I use these tools every day, and sure, they're getting faster and more accurate, but what does the raw data actually say about what's happening under the hood? We can look at the data on that. And I know we talk a lot about frontier AI models.

Just to be clear for everyone listening, we mean the absolute bleeding edge systems being built by the top labs, the ones that cost billions of dollars to train. Yes, exactly. So let's look at the numbers behind those frontier models. The research institute Epoch AI tracked the computing power used to train them. OK, what did they find? From 2010 to 2024, that computing power grew by four to five times every single year. Four to five times every year? Every single year. But raw compute doesn't always translate to how capable a model is in the real world. You can throw 10 times the microchips at a bad architecture and still get a bad model.

Makes sense. So how do we measure actual capability, then? Let's look at METR. They're an organization that evaluates AI capability by testing task length. They measure how long a human expert takes to do a complex multi-step task that an AI can still successfully complete 50% of the time. So they aren't just asking it trivia questions. They are dropping the AI into a virtual environment with a browser, a coding terminal, and saying, here's a massive database. Find the security flaw and write a patch for it. Yes, precisely. And they time how long a human would take to do that same job. Right. They are testing agentic behavior, the ability to plan, execute, fail, correct, and continue over long periods.

OK. In 2025, METR found that the doubling time for this task length was about seven months, meaning every seven months, the length of the autonomous task an AI could handle doubled. Seven months is already fast. It is. But using their revised methodology in January 2026, they looked at models introduced since 2023. And that doubling time dropped to 131 days. Wow. And for models introduced since 2024, it dropped to just under 90 days. Wait, it's doubling every 90 days? How is that jump even possible? What changed in the models to cause that massive drop? It comes down to reasoning and error recovery. The 2023 models were essentially just predicting the next word based on vast training data.

Right, just really good autocomplete. Exactly. If they made a mistake in hour two of a five-hour coding task, they would hallucinate and the whole project would crash. The 2024 models introduced internal reasoning loops. Oh, I see. They learned to write a piece of code, test it themselves, realize it failed, and try a new approach without human intervention. That self-correction is what caused the metric to suddenly skyrocket. So they aren't getting stuck as much. No, they aren't. By May of 2026, METR actually hit a wall. They added a public note stating that tasks exceeding 16 hours of human labor could no longer be reliably measured using their existing tests.

They literally couldn't test them anymore. Right. The AI models are hitting the limits of our ability to even measure them. METR noted that if the trend from 2023 continues, the measured task length capability would increase roughly sevenfold within a single year. Here's where it gets really interesting. Because if the capability is increasing sevenfold in a single year, why doesn't it feel like that when I log in? What do you mean? Well, it feels like having a savings account with insane compounding interest. But when I look at my day-to-day balance, it just looks like a few extra bucks. Shouldn't I feel completely overwhelmed by a paradigm shift every time there's an update?

Why does every new AI model just feel like a slightly better Swiss army knife to me rather than a total revolution? Werner highlights this as a profound psychological blind spot. Humans systematically underestimate exponential growth. We do. Oh, absolutely. It is a deeply documented cognitive bias. Back in 1975, psychologists Willem Wagenaar and Sabato Sagaria ran a series of studies where they presented test subjects with data points representing exponential growth. OK, what happened? Even when looking right at the numbers on a page, the human brain mentally extends the curve as a straight diagonal line. So we visualize a staircase going up one step at a time. Exactly.

We calculate in linear additions. One, two, three, four. But the underlying technology is calculating in multiplications. Two, four, eight, 16, 32. Right. It's just a totally different scale. Because our evolutionary wiring only understands linear progression, every new generation of AI just feels like a slightly better tool. We habituate to the new baseline instantly. That makes so much sense. A tool that would have looked like pure magic three years ago is now just something we complain about if it takes five seconds to load. Precisely. We completely miss where we actually sit on that exponential curve. Okay, but if the average person can't intuitively grasp exponential compounding, the top scientists, the researchers, and the politicians looking at the actual data charts certainly can.

You would hope so. Which brings up the obvious question. If this curve is pointing straight into the sky, and we can't even measure it properly anymore, why isn't anyone pulling the emergency break? Well, the experts have been leaning on the break for years. There has been a steady cascade of ignored warnings. Really? Like what? In March 2023, the Future of Life Institute, which is actually chaired by Max Tegmark, issued an open letter calling for a six-month pause on training models more powerful than GPT-4. Oh yeah, I remember that making the news. They argued we needed time to develop safety protocols. The leading companies ignored it and kept training.

Unbelievable. Then in October 2025, there was the Statement on Superintelligence calling for an outright ban on advanced frontier models until there was scientific consensus on safety. And who signed that one? It was signed by the pioneers of the field, the scientists who literally invented modern deep learning, like Geoffrey Hinton and Yoshua Bengio. The godfathers of AI. Let me guess, the frontier AI CEOs did not sign it. They did not. In fact, Sam Altman at OpenAI had already said he'd be very surprised if by the end of the decade, we didn't have extraordinarily powerful models doing things humans physically cannot do. Wow. And fast forward to just the last few weeks, September 2026, and the scientific panic has spilled over into a full political frenzy.

It really has. And before we get into this, just a quick note here for you listening. Because we're looking at actions from across the political spectrum, from Bernie Sanders to Donald Trump, our goal today isn't to endorse or condemn any political viewpoints. Left wing, right wing, we're not taking any sides. We are impartially reporting on the contents of the source material to understand this geopolitical chess game. Exactly. We're just looking at why the technology isn't stopping. And the source material shows how this issue has blown past traditional partisan lines into a space of sheer geopolitical panic. So what exactly happened in September? In early September 2026, Senator Bernie Sanders announced legislation to ban artificial superintelligence.

He held a rally in Washington calling on Donald Trump to negotiate a direct treaty with China's Xi Jinping to put advanced AI research on hold globally. Okay. Senator Elizabeth Warren supported the idea of a moratorium. And Volker Türk at the UN warned of existential risk. And then on September 23rd, Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act. And that bill has some serious teeth, doesn't it? It threatens up to 20 years in federal prison for anyone violating the ban. 20 years in prison. We've gone from polite open letters to federal criminal bills threatening decades behind bars. But when you look at the mechanics of this, I have to raise an objection here.

It feels like a massive game theory trap. How so? It's a high-stakes staring contest. If I blink, I lose. How can a unilateral American ban work if everyone assumes the other guy, whether that's China or Europe or some private lab, will just keep coding? Werner points out that this is a textbook prisoner's dilemma. The reason nobody pulls the brake isn't a technical failure. It's a psychological and game theoretic one. Right. The economic upside to being the first to achieve general artificial intelligence is estimated in the trillions of dollars. The military upside is total strategic dominance. So the stakes are literally everything? Everything. So every government and every CEO looks at the board and thinks, if we stop, our competitors will keep going.

We will lose our global standing and the existential risk to humanity won't even be reduced because someone else will build it anyway. Which is exactly what Donald Trump argued against the Sanders bill saying, whoever wins in AI wins it all. The fear of coming in second place is stronger than the fear of the technology itself. Exactly. Collectively, all the players might agree that slowing down is the safest choice for humanity. But individually, the rational move for each player is to keep accelerating. So despite warnings from the UN, despite Nobel laureates begging for red lines, despite domestic bills threatening prison time, the race continues without a single brake pad in sight.

The race just keeps going. Which means Tegmark's entire map, the idea of a neat manageable tipping point where we can just pause, hold a summit and decide our fate, it's a fantasy. It's completely obsolete. Right. The top-down political bands are trapped in the prisoner's dilemma. So we have to stop looking at the geopolitical level and start looking at what's actually happening on the ground to us right now. And that is why Werner added six new futures to the framework. Paths 13 through 18. These are scenarios that don't rely on a cinematic supervillain announcement. They look at the messy incremental reality of the technology we are already deploying.

Okay. So instead of walking through them like a list, let's group them by how they actually impact society. Sounds good. First, there are the structural failures. What if the technology plateaus and we get stagnation? Or what if it fragments into dozens of chaotic competing systems, which he calls fragmentation? Both very possible. Right. And then there's AI collapse. This is the scenario where civilization doesn't fall because an AI gets too smart and malicious. It falls because a deeply integrated AI just breaks at the wrong moment. Like a structural collapse. Yes. And Werner points to July 2024, the CrowdStrike software outage. That wasn't even an AI problem.

It was just a bad software update. And it paralyzed hospitals, banks, and airports worldwide in a matter of hours. Oh, I remember that. It was chaos. It was. Imagine that level of fragility with an AI system that autonomously manages the electrical grid or the financial markets. We are building massive single points of failure. It highlights our growing unthinking dependence. But Werner, leaning into his background as a psychotherapist, really focuses on the societal and personal impacts. What happens if it doesn't break? Right. He outlines a scenario called bifurcation. Humanity literally splits into two distinct cultures. One group merges with AI, perhaps economically at first and biologically later.

And the other group fundamentally resists it, demanding human-only spaces and labor. We are already seeing the early seeds of this culturally, aren't we? We are. The UN High Commissioner has even noted a rising trend of AI refusal. But the core thesis of Werner's paper isn't really about the refusers, is it? It's about Path 16, which he calls silent disempowerment. And the mechanism of how we get there, which is Path 18, gradual normalization. If we connect this to the bigger picture, Werner brings in a 2025 paper by researcher Jan Kulveit on the concept of gradual disempowerment. Okay, what does Kulveit say? Kulveit argues that human society loses control of its destiny, not through a hostile, violent takeover, but because AI simply assumes our economic and cultural roles one by one.

Just takes them over. Historically, the economy and governments catered to human interests because they required human labor to function and human purchasing power to grow. But if an AI can do the labor and machine-to-machine transactions drive the economy, the system keeps running perfectly well, but without us as a necessary component. Exactly. We don't get oppressed. We just become irrelevant. We become the retirees of history. Wow. The transition happens slowly through gradual normalization, where each step feels like a minor convenience rather than a loss of autonomy. Okay, I hear you. But wait, I have to push back with a real-world perspective here. Because we aren't being forced to use AI.

We want to use it. Well, sure. I mean, it writes my emails in seconds. It saves me an hour a day. Are you saying my desire for a convenient shortcut is exactly what strips me of my power? Is Werner actually suggesting we go back to doing things the hard way just to prove a point? That is the crux of the debate in the field right now. And your pushback is exactly what prominent thinkers argue. Computer scientists Arvind Narayanan and Sayash Kapoor strongly argue that AI is just normal technology. Right, like electricity or the spreadsheet. Precisely. They say it will spread over decades. We will adapt.

And it will raise the floor for everyone. And economist Tyler Cowen makes the case that AI will actually empower us, leading to a decentralized future where individuals have more leverage, not less. Because it democratizes skills, right? Yes. AI grants access to legal assessments for people who can't afford a law firm. It allows someone who doesn't know a single coding language to build an application for their small business. So there is a massive upside. Delegation can clearly be a form of empowerment. So what does this all mean? How do we know when giving a task to a machine crosses from progress into silent disempowerment? Well, Werner provides a brilliant framework to identify that line.

He argues that delegation only becomes disempowerment when the possibility of returning to the original method disappears. Ah, okay. Like using a GPS. Using a GPS on your phone doesn't disempower your spatial reasoning, as long as paper maps still exist and someone remembers how to read them. Exactly. The danger isn't the delegation, it's the inability to return. If the satellite goes offline and you can't navigate, you're in trouble. Right. If the underlying skill atrophies entirely, or the infrastructure to do it the old way is dismantled, that's when you are trapped. And Werner points to a real-world litmus test that is happening to all of us right now, doesn't he?

He does. Regarding how we gather information. Web searching. The Pew Research Study from March 2025 is the perfect empirical proof of Kulveit and Werner's theory. What did Pew find? They analyzed the behavior of 900 U.S. adults interacting with search engines. When users searched Google and were shown a traditional list of blue links, they clicked on those links to visit the original websites about 15% of the time. Okay, 15%. But when Google placed an AI-generated summary answering the question at the top of the page, users clicked on traditional links only 8% of the time. Wow. That's nearly a 50% drop in traffic to original creators. It gets much deeper.

The actual sources cited within that AI summary. You know, the little footnote links the AI provides to prove its work? Yeah, the citations. Users clicked on those sources only 1% of the time. 1%. Why is the drop so severe? It is cognitive offloading. The human brain is designed to conserve energy. If an AI provides an answer that looks grammatically confident and plausible, the brain skips the energy-intensive verification step. We just trust it. We entirely delegate the task of verifying reality to the algorithm. And this raises an important question regarding environmental researcher Daniel Pauly's shifting baseline syndrome. Shifting baseline. Oh, this is the fishing analogy, right?

Yes. Pauly observed that each generation of fishermen considers the state of the ocean they grew up with as normal. Right. A generation that grows up with an ocean devoid of large fish doesn't mourn the loss of those fish because they never knew they existed. The baseline shifts. So just as a generation growing up with empty oceans doesn't miss the fish, a generation growing up with AI summaries won't miss the original sources. Exactly. Because they will never have developed the habit of seeking them out. And this triggers the secondary economic effect. Independent publishers, investigative journalists, and niche website operators rely on those clicks to survive. Right, and if they lose that traffic because the AI is intercepting the user, they lose their livelihood.

They go out of business. And they stop producing original human research sources. So the path back isn't actively blocked by some malicious code. It simply grows over. The paper maps aren't confiscated. The printing press that makes the paper maps just quietly shuts down. That is the essence of Path 18. Gradual normalization. The true existential risk doesn't require a hostile takeover. No spaceships. No spaceships. The course of history isn't being decided by a dramatic treaty at the United Nations or a scientist pressing a red button in a secure lab. It is happening in the billions of small, entirely unnoticed choices we make every day to save three minutes of reading or 30 seconds of typing.

Wow. We are just voluntarily handing it over. Every decision we delegate is immediately rewarded with less effort, which feels great. But the long-term cost is totally abstract, invisible, until the infrastructure beneath us is gone. Which brings us to the end of our deep dive today. And I want to leave you listening at home with a final lingering thought to mull over. One that builds on everything we've just discussed. It's an important thing to consider. It really is. Because if the shifting baseline means future generations won't miss the original sources because they never knew them, what happens to human curiosity itself? Yeah, that's the real danger.

Right. Because curiosity is driven by the friction of not knowing and the effort required to find out. If an AI always gives us the perfect prepackaged answer instantly, will we eventually lose the cognitive ability to even formulate a novel question on our own? If you never have to search, do you forget how to wonder? Think about that the next time you let an algorithm finish your sentence. Remember, the AI takeover isn't a spaceship in the sky. It's a leaky faucet. And you might already be underwater. Thank you for joining us on this deep dive. We'll see you next time.