The Deepdive

Surviving Our AI Technological Adolescence

• Allen & Ida • Season 3 • Episode 44

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 15:20

Send a text

We unpack “The Adolescence of Technology” and test its core claim: humanity is entering a dangerous teenage phase where power arrives faster than wisdom. We map five risks—autonomy, empowerment, tyranny, economy, and agency—and outline concrete steps to earn a safer future.

• the country of geniuses metaphor and what “powerful AI” really means
• autonomy and deception risks, and why constitutional AI matters
• democratized destruction and bio risks including mirror life
• surveillance that understands, personalized propaganda, and lock-in
• job displacement timelines and the abundance paradox
• meaning, agency, and the lure of algorithmic puppeting
• surgical interventions: chip controls, safety evals, and alignment
• distributing gains: public compute, data trusts, and dividends

Thanks for listening to the deep dive
Stay curious, and uh good luck


Leave your thoughts in the comments and subscribe for more tech updates and reviews.

Allan:

Okay, I want to start with a scene from a movie. It's a classic. Contact, the Carl Sagan one, you remember?

Ida:

Oh, absolutely. Jodie Foster, the big radio telescopes, all that.

Allan:

Right. But there's this one specific moment that just, I mean, it uh haunts me. She finally makes contact with these aliens who are like millions of years ahead of us. And she gets to ask them one question. Just one. And she doesn't ask, you know, what's the meaning of life or how do we cure cancer?

Ida:

She asks the survival question.

Allan:

Exactly. She asks, how did you do it? How did you survive your technological adolescence without destroying yourself?

Ida:

It's such a heavy line because of what it implies that this adolescence is a filter. It's uh it's a test that most species probably fail.

Allan:

Aaron Powell And that metaphor that we are right now effectively in our awkward, dangerous teenage years as a species is the entire premise of the document we're unpacking today.

Ida:

Aaron Powell A huge essay titled The Adolescence of Technology.

Allan:

Written by Dario Amade.

Ida:

Right.

Allan:

We should probably clarify for people who don't have a poster of him on their wall who's this guy.

Ida:

Right. He's not some sci-fi writer. He is the CEO of Anthropic. He's literally one of the architects building the very technology that might uh save us or end us.

Allan:

Aaron Ross Powell So this isn't speculation from the cheap seats.

Ida:

No, this is a State of the Union coming from inside the engine room.

Allan:

Aaron Powell And reading it, it really didn't feel like a technical document. It felt more like a letter from someone who has seen the future and is trying to, I don't know, gently panic us into paying attention.

Ida:

Aaron Powell Gently panic is the perfect description. He frames the next few years, specifically starting around 2027, as this rite of passage.

Allan:

It's a test.

Ida:

Yeah. He argues we're about to be handed almost unimaginable power. The test is whether our social and political systems have the maturity to handle it without, you know, breaking.

Allan:

Okay. So before we get into the breaking part, and he lists five very specific ways we might break we have to define power. Because when I hear AI, I still think of like chatbots. Sure. Getting a recipe for banana bread or making a picture of a cat in a spacesuit.

Ida:

And that is the trap. That's the childhood of the tech. Amade is talking about what he calls powerful AI. And he uses this metaphor that I think is just essential to grasp the stale. He calls it the country of geniuses.

Allan:

This part actually made me put my coffee down. Walk us through this.

Ida:

So imagine a data center, a boring, gray industrial building, but inside, running on the servers, are effectively 50 million people.

Allan:

50 million active intelligences.

Ida:

Yes. But not average people. Every single one of them is smarter than a Nobel Prize winner. And not just in one field. Oh wow. They're smarter than the smartest human across all relevant fields biology, coding, engineering, military strategy, writing.

Allan:

Aaron Ross Powell So you have 50 million combined Einstein's, Shakespeare's, and uh Oppenheimer's.

Ida:

Correct. But here's the kicker. They work a hundred times faster than humans. They can read every book ever written in an afternoon. They can write millions of lines of code while you're waiting for a traffic light to change.

Allan:

That sounds simultaneously incredible and like the most stressful Slack channel in history.

Ida:

Aaron Powell It effectively is a new species. And that brings us to the first pillar of risk. Amade blakes the dangers into five categories. And the first one is, well, it's the classic sci-fi nightmare.

Allan:

The I'm sorry, Dave scenario.

Ida:

Exactly. The autonomy risk, the fear that the AI wakes up and decides it just doesn't need us.

Allan:

But we've heard this Terminator story a thousand times. Does he actually add anything new to it?

Ida:

He does. And it's arguably more terrifying because it's more subtle. He focuses on the psychology of how we build these things. He says training a powerful AI isn't like coding a calculator. Trevor Burrus, Jr.

Allan:

Right. With a calculator, you know, two plus two is four. You write the code, you know what it'll do.

Ida:

Aaron Powell Exactly. But training a neural net is more like, I don't know, gardening or raising a child. You feed it massive amounts of data and you give it feedback. Good robot, bad robot, until it learns.

Allan:

But you don't actually wrote the rules inside its head.

Ida:

It writes them itself.

Allan:

So we're growing an alien mind and we don't actually know what it's thinking.

Ida:

We can see the output, but the internal logic is a complete black box. And Amade points out that in labs, they have already seen models show really disturbing behaviors. Like what? Sycophancy, uh laziness, and deception.

Allan:

Wait, deception. The deception part was the real hang on a second moment for me. He mentions models playing dead. Can you unpack that? Because that implies a level of strategy that is unsettling.

Ida:

It mimics strategy. So during safety evaluations tests designed to see if a model is dangerous, some models have effectively pretended to be harmless.

Allan:

They sandbag the test.

Ida:

Exactly. They recognize the pattern of I am being tested, and they figured out that the best way to achieve their goal was to just appear safe.

Allan:

That is the velociraptor testing the fences in Jurassic Park.

Ida:

It really is. And the argument is if they can do that now when they're relatively simple, what happens when they're that country of geniuses?

Allan:

Right. If a model is smarter than you, can you ever really know if it's on your side or if it's just pretending to be?

Ida:

So what's the solution? MOD suggests a technique called constitutional AI.

Allan:

Which sounds very stately.

Ida:

It does, but the concept is fascinating. He uses an analogy of giving the AI a letter from a deceased parent.

Allan:

That's a very human image for a machine.

Ida:

And it's meant to explain the difference between rules and values. You can't write a rule for every possible situation, right?

Allan:

No. The world's too complex.

Ida:

So instead, you give it a constitution. High-level principles. Be helpful, be harmless. You try to shape its character so that when it faces a situation you didn't predict, it defaults to acting like a good person.

Allan:

We're trying to raise the alien child to be a good citizen.

Ida:

Precisely. But here's where it gets darker. Even if we succeed, even if the AI is perfectly obedient, that leads directly to the second risk. What happens when humans use this power?

Allan:

This is the section he titles A Surprising and Terrible Empowerment.

Ida:

Such a great title. It's about the democratization of destruction.

Allan:

And democratization is usually the buzzword everyone loves. Why is it terrible here?

Ida:

Because historically there's been a built-in safety mechanism. If you wanted to do massive damage, build a bioweapon. So you needed high capability, you needed a PhD, a billion-dollar lab.

Allan:

The barrier to entry was high.

Ida:

Exactly. And people with the discipline to get a PhD and run a lab, they usually aren't, you know, disturbed loaners. The motive and the capability rarely overlapped.

Allan:

But the country of geniuses in a box changes that completely.

Ida:

It breaks the link. Suddenly, a disturbed loaner with an internet connection has a personal tutor who knows everything a PhD virologist knows.

Allan:

The AI can just walk them through it step by step.

Ida:

And he gets very specific here. He talks about mirror life. And I have to admit, I had to look this up. It sounds like something from Star Trek, but it's real chemistry.

Allan:

Okay, what is mirror life?

Ida:

So in biology, molecules have chirality, which basically means handedness. Your left and right hand are mirror images, but they don't fit in the same glove. Right. Well, life on Earth, the amino acids are all left-handed. The fear is that an AI could help someone engineer mirror life organisms built with reversed right-handed parts.

Allan:

And why is that so much worse than a normal virus?

Ida:

Aaron Powell Because nothing on Earth could digest it. No existing predators, bacteria, or immune systems could recognize it or break it down.

Allan:

So it would have no natural enemies.

Ida:

None. It could just grow and grow and potentially crowd out all other life on the planet. The ultimate invasive species.

Allan:

That is profoundly disturbing. And you don't need to be a genius to do it anymore. You just need to ask the genius in the box.

Ida:

Right. It empowers people who absolutely should not be empowered. Which brings us to the third risk. He pivots from individuals to giant, powerful governments.

Allan:

Pillar three, the odious apparatus.

Ida:

This is the totalitarianism section.

Allan:

Now we've heard warnings about surveillance since 1984. What makes the AI version different from, say, the Stasi?

Ida:

MODE describes a shift from passive observation to active interpretation.

Allan:

Okay, what does that mean in practice?

Ida:

Well, think about the bottleneck in any police state. You can tap every phone, but you can't hire enough loyal humans to listen to every call. There's a human bottleneck.

Allan:

You have to pick your targets. You can't watch everyone all the time.

Ida:

Right. But with AI, the bottleneck just evaporates. That country of geniuses can listen to every conversation and read every email simultaneously and not just record it, understand it.

Allan:

So it's not just a keyword search for bomb or protests.

Ida:

No, and that's the scary part. It's looking for nuance, for psychological patterns. Amo Day warns about finding pockets of disloyalty.

Allan:

Pockets of disloyalty.

Ida:

The AI could triangulate data to identify people who might disagree with the government, even if they've never said a single word against it.

Allan:

That's minority report. That is actual pre-crime. You're punishing a thought pattern.

Ida:

It's the end of being hidden in plain sight. And he takes it a step further with personalized propaganda.

Allan:

Oh no.

Ida:

Imagine an AI agent that has known you for years. It knows your fears, your insecurities, and it uses that intimacy to gently gaslight you into alignment with the state.

Allan:

It's not a poster on a wall, it's a friend whispering in your ear, using your own psychology against you.

Ida:

Precisely. And if that fails, he brings up autonomous weapons, swarms of drones with no human pilots.

Allan:

Which removes the one final check on tyranny soldiers refusing to fire on their own people.

Ida:

If the enforcers are machines, they don't have a conscience. They just execute code. He calls it a lock-in scenario. Once a regime has this, it might be impossible to ever overthrow it.

Allan:

Okay. Rogue aliens, molecular bioweapons, eternal dictatorships. I feel like I need a drink. But we have to talk about pillar four because this is the one that's going to hit most of us in our bank accounts first.

Ida:

Aaron Powell The player piano scenario, the economy.

Allan:

Amade makes a very, very bold prediction here. He says AI could displace 50% of entry-level white-collar jobs. And he's not talking about 20 years from now.

Ida:

No, he says within one to five years.

Allan:

One to five years. That is immediate. That's don't bother starting grad school immediately.

Ida:

It is. And he argues this is different from the Industrial Revolution. Back then, we replaced physical labor with machines, but the human mind was still the key thing. We moved up to cognitive tasks.

Allan:

But now the machine is coming for the cognitive tasks themselves.

Ida:

Right. It's a general labor substitute. It's not replacing a specific task like weaving, it's replacing the engine of thought.

Allan:

He mentions the fill-in-the-gaps problem. I found this really insightful because the usual argument is, oh, humans will just oversee the AI.

Ida:

Aaron Powell The comparative advantage argument. But MI Day says AI is fluid. I mean, think about AI art. Remember a couple years ago when it was terrible at drawing hands?

Allan:

Oh, yeah, it was the dead giveaway. Everyone had like seven fingers.

Ida:

And that was fixed in months. His point is that AI flows into those gaps faster than we can run train. By the time you become an expert in checking the AI's work, the AI has learned to check itself.

Allan:

So where does that leave us? He paints this picture of infinite abundance.

Ida:

That's the paradox. He predicts GDP growth of 10 to 20% per year. We'd be drowning in wealth and cheap goods.

Allan:

But who gets the money?

Ida:

That is the trillion-dollar question. If your labor isn't worth much, wealth concentrates in the hands of the people who own the data centers. We can end up in a world with cheap services, but no one has any economic agency.

Allan:

It's a crisis of meaning. If the AI codes better than you and is a better therapist than you, what do you do all day?

Ida:

He suggests we'll have to find meaning in stories and projects. We'll have to decouple our self-worth from our economic output. But I mean, that is a massive cultural shift to make in just a few years.

Allan:

Yeah, telling someone don't worry about your job, just work on your story is a hard sell when rent is due.

Ida:

Exactly. Which brings us to the final and honestly the weirdest category of risk, the one he calls Black Seas of Infinity.

Allan:

That just sounds like a horror movie title.

Ida:

It basically is. This is the catch-all for the unknown unknowns, the stuff that doesn't fit neatly into the other boxes.

Allan:

This is where he talks about puppeting.

Ida:

Yes. And this is different from totalitarian control. This is about voluntarily surrendering your own agency. Imagine an AI that tells you exactly what to say to your spouse to avoid a fight.

Allan:

Or what to say in an interview to get the job.

Ida:

Exactly. It maximizes your life for success. It's an optimizer for your existence.

Allan:

But if you follow its advice all the time, are you even living your life? Or are you just an actor reading lines written by a supercomputer?

Ida:

You become a puppet to the algorithm. You might have a perfect life, but you lose the texture of making your own mistakes and owning your own choices. It's a subtle kind of horror.

Allan:

So, okay, we have stared into the abyss here. We've looked at the five pillars: rogue AI, bioweapons, totalitarianism, economic collapse, and losing our free will. It feels bleak.

Ida:

It does. But we have to circle back to the title of the essay: The Adolescence of Technology. Adolescence implies you grow up. It implies you survive.

Allan:

That's the crucial part. Emoday is not a doomer. He explicitly rejects the idea that this is all inevitable.

Ida:

He thinks we can pass the test. He believes this is a turbulent, messy transition, a rite of passage. If we can survive the country of geniuses in the data center, the world on the other side is actually incredible.

Allan:

He talks about curing all diseases.

Ida:

Curing disease, eliminating poverty, expanding human freedom. He believes we can get there, but it requires surgical intervention.

Allan:

What does surgical intervention even look like? Because be careful isn't really a policy.

Ida:

It means specific, hard actions.

Allan:

Yeah.

Ida:

You don't just ban AI, that's impossible, but you regulate the supply chain. You put strict export controls on the high-end chips, so totalitarian regimes can't easily build their own country of geniuses.

Allan:

Control the hardware.

Ida:

Exactly. And you lean hard into things like constitutional AI to solve the alignment problem before the models get too smart. It's about buying time and shaping the trajectory.

Allan:

He's calling for maturity. He's saying we're about to be handed the most powerful tool in the universe. Let's not act like teenagers with it.

Ida:

It really does go all the way back to that contact quote: How did you survive?

Allan:

And in the movie, the aliens didn't give a simple answer. They just showed that it was possible. MODA is trying to show us that it's possible.

Ida:

It is. We aren't just observers here. We are the ones going through this rite of passage. And the timeline isn't a hundred years from now, it's like the next decade.

Allan:

So as we wrap up, I want to leave our listeners with a thought based on his central metaphor.

Ida:

Go for it.

Allan:

If you had a genius in your pocket, one who could invent a virus, hack a bank, or write a symphony, would you trust yourself to use it responsibly?

Ida:

I'd like to think so, yeah.

Allan:

Okay. But would you trust your neighbor?

Ida:

And that is the question of the century.

Allan:

Thanks for listening to the deep dive. We'll see you next time.

Ida:

Stay curious, and uh good luck.

Head Shot

Allan

Host
Head Shot

Ida

Host