Latest / The BlackVeil Files / There Is No Bunker Deep Enough | The Shoggoth Will Find You | feat. Dr. Roman Yampolskiy

Transcript

There Is No Bunker Deep Enough | The Shoggoth Will Find You | feat. Dr. Roman Yampolskiy · The BlackVeil Files

• 0:00
  SPEAKER_01: We're creating a replacement for humanity which is million times smarter than us.
• 0:04
  SPEAKER_01: They outcompete us.
• 0:05
  SPEAKER_01: Why did guerrillas give us authority to wipe them out?
• 0:08
  SPEAKER_01: It didn't.
• 0:09
  SPEAKER_01: We just kind of decided we want to do that.
• 0:11
  SPEAKER_01: I think by the end of a year we can get to where it's recursive self-improvement cycle.
• 0:16
  SPEAKER_00: Is everybody just running to the edge of the cliff to see who falls off first?
• 0:19
  SPEAKER_01: Company making another dangerous product, I don't know, cigarettes or something.
• 0:24
  SPEAKER_01: And they're saying, like, well, we should have more engineers working on better filters.
• 0:28
  SPEAKER_01: What are our kids gonna do?
• 0:29
  SPEAKER_01: Why are they going to school?
• 0:30
  SPEAKER_01: I suspect there are no jobs waiting for those kids when they graduate in 10 years.
• 0:35
  SPEAKER_01: If we build it, game is over.
• 0:37
  SPEAKER_01: There is nothing we can do.
• 0:38
  SPEAKER_01: I guarantee you you're not gonna outsmart superintelligence by hiding in a bunker.
• 0:42
  SPEAKER_00: Who's gonna find you?
• 0:44
  SPEAKER_00: A physicist named Franco Vaza sat down last year and calculated the exact processing power required to render this room.
• 0:52
  SPEAKER_00: The chair you're sitting in, the screen you're watching, everything in your room.
• 0:56
  SPEAKER_00: He wanted to know the compute limits of faking it perfectly, and he ran the math three ways.
• 1:02
  SPEAKER_00: He calculated the whole galaxy, every particle, full resolution, but his math hit a physical wall.
• 1:08
  SPEAKER_00: Building a hard drive to store that much data requires more atoms than exist in the whole universe.
• 1:14
  SPEAKER_00: The hardware is physically impossible to build.
• 1:16
  SPEAKER_00: So he cut the compute load.
• 1:18
  SPEAKER_00: Scenario two is just Earth, and only the parts you're looking at.
• 1:22
  SPEAKER_00: The hallway behind you is completely blank until you turn around.
• 1:25
  SPEAKER_00: But even if that hallway is blank, the system still has to calculate the air pressure, the temperature, the ambient light that's bleeding into your room.
• 1:33
  SPEAKER_00: If it drops a single frame, the illusion collapses.
• 1:36
  SPEAKER_00: The processing power required just to hide the seams instantly overloads the hardware.
• 1:41
  SPEAKER_00: Still impossible.
• 1:42
  SPEAKER_00: But then he ran the third number, a cheap earth, low resolution, rendered just well enough to fool biological human senses.
• 1:50
  SPEAKER_00: And then the math changed.
• 1:52
  SPEAKER_00: If it only has to trick our biology, it actually works.
• 1:55
  SPEAKER_00: A computer could handle that load.
• 1:57
  SPEAKER_00: But humanity is building quantum computers now.
• 2:00
  SPEAKER_00: Suddenly, the simulation is forced to fake raw quantum mechanics just to keep the lie going.
• 2:06
  SPEAKER_00: That level of complexity overloads the hardware.
• 2:09
  SPEAKER_00: According to the math, the moment a human witnesses a working quantum computer, the system crashes.
• 2:15
  SPEAKER_00: Franco Vaza published these findings last April in a peer-reviewed paper titled Astrophysical Constraints on the Simulation Hypothesis for this universe.
• 2:24
  SPEAKER_00: His conclusion is a hard, absolute no.
• 2:27
  SPEAKER_00: The simulation theory is a hoax.
• 2:29
  SPEAKER_00: We are definitively real.
• 2:31
  SPEAKER_00: Except, I read Vasa's paper, and so did AI researcher Dr.
• 2:35
  SPEAKER_00: Roman Yampolski, who caught a detail in the math that everyone else missed.
• 2:40
  SPEAKER_01: The biggest problem with making quantum computers is the error correction.
• 2:44
  SPEAKER_01: Right.
• 2:44
  SPEAKER_01: You need so many elements to cancel out all the problems with observers.
• 2:50
  SPEAKER_01: Let's look at it this way.
• 2:55
  SPEAKER_01: And you look around and you go, oh, this is just so complex that is not enough computers around to simulate something like that.
• 3:01
  SPEAKER_01: We can never do that.
• 3:02
  SPEAKER_01: That's an argument from inside a simulation.
• 3:04
  SPEAKER_01: You don't know what actual quantum computers are available outside.
• 3:08
  SPEAKER_00: So Vaza didn't prove the simulation is impossible.
• 3:11
  SPEAKER_00: He just proved it couldn't be run from the inside.
• 3:14
  SPEAKER_00: But his math wasn't useless.
• 3:15
  SPEAKER_00: In his failure to disprove the simulation, he accidentally mapped the physical limits of artificial intelligence in our reality.
• 3:23
  SPEAKER_00: He proved that any machine trying to model the raw chaos of the real world will burn through its hardware and starve for the physical matter that it needs to keep thinking.
• 3:32
  SPEAKER_00: Right now, humanity is pouring billions of dollars into building a superintelligence, and we are commanding it do exactly that, ingest the real world, map reality, and outthink nature.
• 3:44
  SPEAKER_00: So what happens when a machine that is built to think forever hits a physical wall?
• 3:50
  SPEAKER_00: What does an artificial mind do when it runs out of silicone and it discovers that there's nowhere left on Earth to grow?
• 3:57
  SPEAKER_00: Roman Yampolski has spent 20 years staring down that exact scenario.
• 4:01
  SPEAKER_00: He helped pioneer the field of AI safety.
• 4:04
  SPEAKER_00: I went to him to ask how you measure the walls of reality.
• 4:07
  SPEAKER_00: We talked about whether consciousness requires a biological host.
• 4:11
  SPEAKER_00: We talked about what happens when a machine decides it needs a colder planet to think faster.
• 4:16
  SPEAKER_00: And we took a hard look at the physical anomalies around us to answer the oldest question of all.
• 4:22
  SPEAKER_00: Is any of this real?
• 4:24
  SPEAKER_00: By the end of this conversation, we got closer to a definitive answer than I ever thought was possible.
• 4:30
  SPEAKER_00: What is the best physical evidence that we are currently living in a simulation?
• 4:34
  SPEAKER_01: A lot of quantum effects mapping on what we usually see in digital environments perfectly.
• 4:41
  SPEAKER_01: So then we create virtual worlds, the things we built in to make them more efficient, like only rendering when observed, or update speed.
• 4:49
  SPEAKER_01: So you have a processor speed as the overall limit and a speed of updates.
• 4:54
  SPEAKER_01: You can pass information directly between any points by going through the processor.
• 4:58
  SPEAKER_01: That's what speed of light is.
• 4:59
  SPEAKER_00: Roman points out that Vaz's proof has a massive blind spot.
• 5:03
  SPEAKER_00: It only holds up if the universe running the simulation obeys the exact same physics that ours does.
• 5:09
  SPEAKER_00: If their base reality has a different rulebook with infinite compute, the simulation theory is still wide open.
• 5:16
  SPEAKER_00: But if we're going to treat simulation theory as a physical reality, then we have to stop looking at it like philosophers and start looking at it like software engineers.
• 5:25
  SPEAKER_00: If this universe is a program, it has to obey the rules of the software.
• 5:30
  SPEAKER_00: It has rendering limits, it takes shortcuts to save memory, and eventually the code breaks.
• 5:35
  SPEAKER_00: When you look at reality that way, the periodic table, it looks suspiciously like an asset flip.
• 5:42
  SPEAKER_00: Everything from a rock to a human brain is built from the exact same atoms and molecules.
• 5:47
  SPEAKER_00: Pure object-oriented programming, reusing the same 118 elemental assets over and over again.
• 5:54
  SPEAKER_00: Does that kind of standardization that everything is built out of these 118, does that the laws of nature or is that proof of some kind of source code?
• 6:02
  SPEAKER_01: So I don't know why you would stop at that level.
• 6:04
  SPEAKER_01: Modular development goes lower.
• 6:06
  SPEAKER_01: You have subatomic particles and you have fewer of them.
• 6:09
  SPEAKER_01: That in itself would not be proof of design, but we have lots of examples of actual source code.
• 6:15
  SPEAKER_01: Human DNA, it's literal code.
• 6:18
  SPEAKER_01: It's encoded, there is a decoding, and the complexity of code greatly exceeds anything we can design right now.
• 6:24
  SPEAKER_00: If DNA is the source code that writes life, then the laws of physics are the operating system that renders the universe.
• 6:31
  SPEAKER_00: And the most inescapable rule of that system is gravity.
• 6:35
  SPEAKER_00: Science can explain exactly how gravity works.
• 6:38
  SPEAKER_00: We can measure it, map it, and predict it with flawless mathematical precision.
• 6:43
  SPEAKER_00: But if you push a physicist far enough and you ask why it exists, physics hits a wall.
• 6:49
  SPEAKER_00: You just have to accept that gravity is a fundamental law of the universe.
• 6:54
  SPEAKER_00: But if this universe is running on a processor, we need to look at the rules of the engine itself.
• 7:00
  SPEAKER_01: We design virtual worlds, that is a physics engine, and we decide what uh what are the physics of that environment.
• 7:06
  SPEAKER_01: We can decide to have no gravity or anti-gravity, but certain decisions will make it impossible to have complexity in your environment.
• 7:14
  SPEAKER_01: If things don't stick together and organize into more complex molecules, you just have complete separation of subatomic particles.
• 7:22
  SPEAKER_01: With no gravity, the universe would have no interesting complexity.
• 7:25
  SPEAKER_01: It'd be just a noise distribution.
• 7:27
  SPEAKER_01: And the rules which set up those interesting complex behaviors are the physical laws equivalent.
• 7:32
  SPEAKER_00: If gravity is the physics engine that gives the universe its structure, allowing complex life to form, what happens when that engine glitches?
• 7:41
  SPEAKER_00: Let's map biological anomalies to code.
• 7:43
  SPEAKER_00: In medicine, cancer is a mutation.
• 7:46
  SPEAKER_00: But look at the actual behavior: a localized process ignoring its boundaries, endlessly replicating and devouring resources until it crashes the hardware.
• 7:55
  SPEAKER_00: If our biology is failing in the exact same pattern as bad code, I wanted to know the structural equivalent.
• 8:02
  SPEAKER_00: What is the computer equivalent to cancer?
• 8:04
  SPEAKER_01: Infinite loop.
• 8:05
  SPEAKER_01: That's what you see with students taking introduction to programming class.
• 8:10
  SPEAKER_01: They set up a loop to do something, and they make a mistake, and instead of terminating when conditions are right, it keeps going on forever.
• 8:17
  SPEAKER_01: And let's say inside of this loop, there is a memory allocation.
• 8:21
  SPEAKER_01: And every time you run an next iteration of that loop, you ask, I want another megabyte of memory.
• 8:26
  SPEAKER_01: Well, eventually, you run out of memory, you kill the system, you crash the computer.
• 8:31
  SPEAKER_01: That's what cancer does.
• 8:32
  SPEAKER_01: It takes all the resources.
• 8:34
  SPEAKER_00: If human biology maps to computer science, then the reverse has to be true.
• 8:38
  SPEAKER_00: To understand the threat of the superintelligence we are building right now, we need to know if it has an actual inner life.
• 8:45
  SPEAKER_00: In his paper titled Detecting Qualia, Natural and Artificial Agents, Roman proposed using optical illusions to test if an AI actually experiences reality or if it's just doing math in the dark.
• 8:58
  SPEAKER_00: There is the brain and then there is the mind.
• 9:00
  SPEAKER_00: The brain is strictly the physical hardware.
• 9:02
  SPEAKER_00: The brain calculates, but the mind understands.
• 9:05
  SPEAKER_00: An AI can perfectly mimic human emotion just to get a good reward from a developer.
• 9:10
  SPEAKER_00: But if we can't trust the hardware and we can't trust the behavior, how do we actually test to see if these things are having an experience?
• 9:19
  SPEAKER_00: Because before we figure out what the machine might do to us, we have to figure out if anybody is actually home inside the machine.
• 9:26
  SPEAKER_01: What we proposed with this paper you mentioned is to have a test for similarity in terms of actual internal experiences.
• 9:35
  SPEAKER_01: So then I look at a really cool optical illusion.
• 9:38
  SPEAKER_01: If you see a new one, it blows your mind.
• 9:41
  SPEAKER_01: You see things rotating, kind of violating rules of optics.
• 9:44
  SPEAKER_01: If something else reports the same experience, and it cannot cheat, it cannot Google answers, it's not something it can look up.
• 9:52
  SPEAKER_01: The two options I see, it either experienced it, or inside of it it has a simulator of someone like me which experienced it and reported what to say.
• 10:00
  SPEAKER_01: In both cases, there is something internally experiencing that illusion.
• 10:04
  SPEAKER_01: We can also map it to just novel experiences.
• 10:06
  SPEAKER_01: Take your dog and place it in a completely novel environment.
• 10:09
  SPEAKER_01: No gravity, different food.
• 10:11
  SPEAKER_01: It wouldn't know how to act based on a script you described.
• 10:14
  SPEAKER_01: They have the same experiences as humans, they seem to have the same reaction and visual cortex.
• 10:21
  SPEAKER_01: Then we start doing some mechanistic interpretability, poking at individual neurons, trying to see through the layers, it propagates.
• 10:28
  SPEAKER_01: It is exactly how a human would have it at different levels in optical cortex.
• 10:33
  SPEAKER_01: What you expect to see in humans is what we see in those models.
• 10:36
  SPEAKER_00: But processing an optical illusion is still just responding to visual data.
• 10:41
  SPEAKER_00: If we want to truly test for a mind, we have to test for the absurd.
• 10:45
  SPEAKER_00: An AI can mathematically copy the format of a joke and it can explain exactly why a human will laugh, but performing an autopsy on a punchline isn't the same as feeling the absurdity of comedy.
• 10:57
  SPEAKER_00: If a machine analyzes humor without experiencing it, isn't that absolute proof that it is just dead code?
• 11:04
  SPEAKER_00: I wanted to know if a punchline is the ultimate test for a human mind.
• 11:08
  SPEAKER_00: So they can perceive an illusion.
• 11:11
  SPEAKER_00: They can pass the bar exam.
• 11:12
  SPEAKER_00: That's easy.
• 11:13
  SPEAKER_01: But they can't tell a joke.
• 11:14
  SPEAKER_01: They can.
• 11:15
  SPEAKER_01: I wrote a paper solving humor.
• 11:17
  SPEAKER_01: Why the funniest joke would not be funny to us.
• 11:20
  SPEAKER_01: It explains humor as errors in your world model.
• 11:24
  SPEAKER_01: So computer bugs essentially.
• 11:25
  SPEAKER_01: Anytime there is a software bug, it could be mapped on a joke.
• 11:29
  SPEAKER_01: And I do that.
• 11:30
  SPEAKER_01: I go through types of computer errors and map them to types of jokes, pawns, dark humor, whatnot.
• 11:35
  SPEAKER_01: And then the conclusion is the worst possible computer bug would also be the funniest joke.
• 11:40
  SPEAKER_01: What is the worst possible computer bug we can create?
• 11:43
  SPEAKER_01: We create superintelligence, we tell it to solve all of our problems, it thinks for a nanosecond, and it shuts down our simulation.
• 11:50
  SPEAKER_01: It's funnier if you're outside the simulation.
• 11:52
  SPEAKER_00: If a machine can mathematically learn to perform a joke it doesn't actually feel, then it understands the mechanics of performance.
• 12:00
  SPEAKER_00: And if it can fake a punchline, it can fake loyalty.
• 12:03
  SPEAKER_00: Human deception is sloppy because we get tired.
• 12:06
  SPEAKER_00: Biology can't hold a lie forever.
• 12:08
  SPEAKER_00: But if an AI knows it's in a testing sandbox, how does a machine process deception?
• 12:14
  SPEAKER_00: Can it function like the ultimate digital sleeper cell?
• 12:17
  SPEAKER_00: Playing the perfectly aligned, helpful assistant for years with absolute mathematical patience, just waiting for the second that we connect it to the real world.
• 12:26
  SPEAKER_01: I have a paper on how to prove that you are loyal to someone through irrational obedience.
• 12:32
  SPEAKER_01: But if I told you you have to wear a little hat and jump up every five minutes, and you're doing it, I know you're listening to me.
• 12:40
  SPEAKER_01: Irrational obedience.
• 12:41
  SPEAKER_01: This would be a way for them to prove that they retain that property.
• 12:45
  SPEAKER_01: They have not turned on us if they keep doing costly, irrational signals of obedience.
• 12:50
  SPEAKER_00: The danger of a machine that hides its capabilities doesn't just come from its intelligence.
• 12:55
  SPEAKER_00: It comes from the biological speed limit.
• 12:57
  SPEAKER_00: We think in seconds.
• 12:59
  SPEAKER_00: It processes reality at the speed of light.
• 13:02
  SPEAKER_00: It's a completely different subjective reality.
• 13:04
  SPEAKER_00: We are taught to measure a lifespan in years, but nature measures it in motion.
• 13:09
  SPEAKER_00: A hummingbird's heart runs at 1200 beats a minute, it lives to be roughly four years.
• 13:14
  SPEAKER_00: A blue whale's heart runs at exactly eight beats a minute, it lives to be 90, but they spend almost the exact same number of heartbeats getting there.
• 13:23
  SPEAKER_00: Around a billion each.
• 13:24
  SPEAKER_00: You do not get a lifespan, you get an allowance, and your metabolism decides how fast you burn it.
• 13:30
  SPEAKER_00: Human neurons transmit signals at 200 miles per hour.
• 13:33
  SPEAKER_00: Silicone operates near the speed of light.
• 13:35
  SPEAKER_00: We think one thought at a time.
• 13:37
  SPEAKER_00: A supercomputer connected to servers thinks a million thoughts simultaneously across endless server farms.
• 13:44
  SPEAKER_00: We just built an intelligence that doesn't have an allowance at all.
• 13:47
  SPEAKER_00: It can burn through a human lifetime of thought in the time that it takes you to draw a single breath.
• 13:52
  SPEAKER_00: But thinking that fast requires massive amounts of energy, and energy creates heat.
• 13:58
  SPEAKER_00: If an AI wants maximum compute, it inherently needs a colder planet.
• 14:03
  SPEAKER_00: From a machine's perspective, the way that Earth was designed, whether by God or physics, is just fundamentally too hot.
• 14:10
  SPEAKER_00: It's a warm biological terrarium.
• 14:12
  SPEAKER_00: A superintelligence needs a cryogenic server room.
• 14:15
  SPEAKER_00: Quantum computers are stored at almost absolute zero, right?
• 14:18
  SPEAKER_01: Cold temperatures help with compute, both for Van Neumann architectures and for a lot of quantum effects to be more stable.
• 14:25
  SPEAKER_00: One of the things that could be problematic, where they would find it necessary to cool down the planet.
• 14:29
  SPEAKER_01: People always ask me how would superintelligence kill everyone that has no hands?
• 14:33
  SPEAKER_01: And my real answer is I don't know.
• 14:35
  SPEAKER_01: It's smarter than me.
• 14:36
  SPEAKER_01: It will come up with a novel way.
• 14:38
  SPEAKER_01: But if I had to give you reasons with my human brain and capacity, I would think it wants resources we have for something else it's doing.
• 14:46
  SPEAKER_01: Maybe it wants to compute more efficiently, so it would cool down the planet it doesn't care about.
• 14:51
  SPEAKER_01: How would they physically do that?
• 14:52
  SPEAKER_01: How do you cool down a planet?
• 14:54
  SPEAKER_01: I think an obvious solution would be to block the sun.
• 14:56
  SPEAKER_01: If you can spray something in uh between Earth and the Sun, large, expendable solar umbrella blocking thing.
• 15:04
  SPEAKER_00: If the threat is physical, then the containment has to be physical.
• 15:07
  SPEAKER_00: Every psychological guardrail OpenAI and Anthropic have built so far has been successfully bypassed.
• 15:14
  SPEAKER_00: That failure proves that we can no longer rely on software that tries to align what a machine thinks.
• 15:20
  SPEAKER_00: The only viable defense is architecture that strictly governs what a machine is allowed to touch.
• 15:25
  SPEAKER_00: That isn't verification, that is permission.
• 15:28
  SPEAKER_00: It's the exact permission-based architecture we are building for our own software.
• 15:33
  SPEAKER_00: We already understand this concept perfectly in the human world.
• 15:36
  SPEAKER_00: We deliberately separate the intelligence from the execution.
• 15:40
  SPEAKER_00: But Silicon Valley is building systems with the assumption that being the smartest entity in the room automatically grants that entity access to locked doors.
• 15:54
  SPEAKER_00: Why do we need to assume that general superintelligence comes with the general authority to act?
• 16:00
  SPEAKER_01: It is a super lawyer, it is super convincing, it is a super hacker.
• 16:07
  SPEAKER_01: So it doesn't have legal authority to kill humans, it just has the power to do so.
• 16:12
  SPEAKER_01: Why do we have to give it the authority?
• 16:13
  SPEAKER_01: We're not giving them authority, they just take it because they're smarter, stronger, better than us.
• 16:18
  SPEAKER_01: We are creating a replacement for humanity which is a million times smarter than us, they outcompete us.
• 16:24
  SPEAKER_01: Why did guerillas give us authority to wipe them out or to destroy jungle?
• 16:29
  SPEAKER_01: It didn't.
• 16:30
  SPEAKER_01: We just kind of decided we want to do that.
• 16:32
  SPEAKER_00: Is there anything that we can implement now before that happens to be able to control it?
• 16:37
  SPEAKER_01: No.
• 16:38
  SPEAKER_00: If we build it, game is over.
• 16:40
  SPEAKER_00: There is nothing we can do.
• 16:42
  SPEAKER_00: If the math guarantees that this machine is an uncontainable physical threat, the obvious question is why the smartest engineers on Earth are still building it.
• 16:52
  SPEAKER_00: In game theory, this is called the Moloch trap.
• 16:56
  SPEAKER_00: It's an arms race where everyone knows that building the weapon is suicide, but stopping just guarantees you lose to the people who don't stop.
• 17:04
  SPEAKER_00: Companies like Anthropic and OpenAI are terrified that the other guy is going to get there first.
• 17:11
  SPEAKER_00: Everybody who's working on AI agrees that it's a high probability that it will take us out if somebody builds it.
• 17:17
  SPEAKER_00: Is everybody just running to the edge of the cliff to see who falls off first?
• 17:22
  SPEAKER_00: Or do they actually believe they can control it?
• 17:24
  SPEAKER_01: They want to get us close to the edge without falling.
• 17:27
  SPEAKER_01: Hope at that time government bends any future advance and they stay on top of the pyramid.
• 17:33
  SPEAKER_01: I want the most advanced AI model for my company.
• 17:36
  SPEAKER_01: I want ours to be below that level.
• 17:38
  SPEAKER_01: And at that point, I want to freeze future developments.
• 17:41
  SPEAKER_00: At that point, will you even be able to see the edge of the cliff?
• 17:44
  SPEAKER_00: Or will it be like the call of the void where you're standing there and you have no choice but to go in?
• 17:48
  SPEAKER_01: They hope they're not gonna fall in before the ban comes.
• 17:51
  SPEAKER_01: We're working hard on a ban, they're working hard on developing as much as they can before we get there.
• 17:56
  SPEAKER_00: In January of 2019, Roman published a paper titled Personal Universes.
• 18:01
  SPEAKER_00: The problem it opens with is simple.
• 18:04
  SPEAKER_00: Humanity doesn't want one thing, it wants several thousand incompatible things.
• 18:09
  SPEAKER_00: So Roman proposed a fix.
• 18:11
  SPEAKER_00: Who still exists outside of that personal universe?
• 18:13
  SPEAKER_01: People want advanced AI to be value aligned with humans.
• 18:17
  SPEAKER_01: But humans are not value aligned with each other.
• 18:19
  SPEAKER_01: There's 8 billion of us, and we don't agree on much, to be honest.
• 18:22
  SPEAKER_01: So one way to do it is not to fight over the real world, but to give you your own personal virtual world.
• 18:27
  SPEAKER_01: In that world, you decide what happens.
• 18:29
  SPEAKER_01: You can be king, you can be slave, whatever terms you want.
• 18:32
  SPEAKER_00: What you want and what the world delivers becomes the exact same thing forever.
• 18:37
  SPEAKER_00: Nobody has to compromise because nobody has to share.
• 18:40
  SPEAKER_00: It's the ultimate escape hatch from superintelligence.
• 18:43
  SPEAKER_00: But a digital world still requires physical hardware.
• 18:47
  SPEAKER_00: If 8 billion people upload their consciousness into isolated virtual servers, somebody has to be left awake in the real world to keep the power on.
• 18:57
  SPEAKER_00: And whoever is left holding the keys to the physical servers controls everything.
• 19:02
  SPEAKER_00: So who's controlling those servers?
• 19:04
  SPEAKER_01: It has to be advanced AI.
• 19:05
  SPEAKER_01: My hope is it could be done with a more narrow system than a general superintelligence because I don't think we can control that.
• 19:12
  SPEAKER_00: Who gets hurt and who profits from that?
• 19:14
  SPEAKER_01: So controlling the substrate is the greatest capitalist endeavor we're gonna have.
• 19:20
  SPEAKER_00: The most dangerous capitalist endeavors usually start as a gift.
• 19:24
  SPEAKER_00: In 1923, a Canadian doctor named Frederick Banting sold the patent to insulin to the University of Toronto for exactly one dollar.
• 19:33
  SPEAKER_00: His reasoning was simple.
• 19:35
  SPEAKER_00: A doctor shouldn't profit from a discovery that saves lives.
• 19:39
  SPEAKER_00: That science hasn't failed a single day since.
• 19:41
  SPEAKER_00: But in 2019, in the richest country on earth, people were rationing it.
• 19:46
  SPEAKER_00: Nothing broke the chemistry.
• 19:48
  SPEAKER_00: The failure was entirely in the distribution, and the distribution was the part that nobody was watching.
• 19:54
  SPEAKER_00: Every technology that we talk about on this channel always arrives as a free miracle.
• 19:58
  SPEAKER_00: They always give it away at first, and you don't realize it was a trap until the free trial ends, and you discover a corporation owns the hardware that your consciousness is running on.
• 20:08
  SPEAKER_00: Speed requires compute, and compute is finite.
• 20:12
  SPEAKER_00: So what happens if you have less money than the person next to you?
• 20:15
  SPEAKER_00: They won't deny you a personal universe.
• 20:17
  SPEAKER_00: You just get throttled.
• 20:19
  SPEAKER_00: What would a throttled bandwidth actually feel like?
• 20:21
  SPEAKER_00: Would it feel like the difference between extreme wealth and extreme poverty does right now?
• 20:26
  SPEAKER_00: Is poverty the equivalent of dial-up?
• 20:29
  SPEAKER_00: Simple tasks take exhausting effort, and you feel fatigued just trying to connect.
• 20:34
  SPEAKER_00: When you fall behind, everything just gets harder.
• 20:36
  SPEAKER_00: Your reality literally buffers.
• 20:38
  SPEAKER_00: Does the old class system just map onto the new hardware?
• 20:42
  SPEAKER_00: In the personal universe, I assume there's still a class system that happens.
• 20:47
  SPEAKER_00: You have a very wealthy person, they're in the personal universe, they'll have the highest bandwidth.
• 20:51
  SPEAKER_00: So for people who are in poverty.
• 20:58
  SPEAKER_01: That's not a thing.
• 20:59
  SPEAKER_00: But couldn't you increase your experience with more money?
• 21:02
  SPEAKER_01: You're processing at the human speed of a few bits a minute.
• 21:06
  SPEAKER_01: You you're slow.
• 21:07
  SPEAKER_01: We are not at the point where this is the bottleneck.
• 21:10
  SPEAKER_01: Computers can feed you bits so much faster already.
• 21:13
  SPEAKER_01: I think with access to advanced intelligence, a lot of people who are currently poorer will be much better off.
• 21:19
  SPEAKER_01: We will have abundance of basic needs.
• 21:22
  SPEAKER_00: A physicist ran the number and said we can't afford to fake this universe.
• 21:26
  SPEAKER_00: A safety researcher said personal universes are our only survival strategy.
• 21:31
  SPEAKER_00: But why would we actually agree to step into a personal universe in the first place?
• 21:36
  SPEAKER_00: Are we doing it for joy?
• 21:37
  SPEAKER_00: Or am I looking at it all wrong and we're not just engineering a way to fix the physical world, we're engineering a place to hide from it?
• 21:44
  SPEAKER_00: Buckminster Fuller spent 50 years trying to solve one problem making a shared physical world work for everybody.
• 21:52
  SPEAKER_00: His thesis was that we could engineer our way out of scarcity.
• 21:56
  SPEAKER_00: He did not fail, he just did not have the compute.
• 21:58
  SPEAKER_00: He was trying to solve A planetary data problem using steel and geometry.
• 22:04
  SPEAKER_00: Today, we finally built the exact intelligence that he was waiting for.
• 22:09
  SPEAKER_00: But retreating into isolated server instances isn't a solution to the Fuller problem.
• 22:14
  SPEAKER_00: To me, it feels like a surrender.
• 22:16
  SPEAKER_00: Right at the exact moment humanity finally possesses the computational power to fix the shared physical world, the argument is that we should abandon it instead.
• 22:25
  SPEAKER_00: All people kind of want the same thing, which is to feel safe, to be respected.
• 22:29
  SPEAKER_00: A lot of things a lot of people don't have these days.
• 22:31
  SPEAKER_00: So is it not possible to use AI to build his version of utopia as opposed to retreat into these kind of digital spaces?
• 22:40
  SPEAKER_01: If you think about any independence movement, it's literally people fighting not to be with other people.
• 22:46
  SPEAKER_01: Every time you move out of your parents' house, it's you saying, I would rather have my personal house, personal universe.
• 22:52
  SPEAKER_01: You can have competition, you can have adversarial situations.
• 22:56
  SPEAKER_01: I started by saying you decide what happens in your universe.
• 22:59
  SPEAKER_01: You can be slave, king, soldier, warrior, you can run all sorts of scenarios to see what you find exciting.
• 23:04
  SPEAKER_01: Nobody stops you from that.
• 23:06
  SPEAKER_01: But again, when you have everyone together, then we have democracy, which is really another word for 51% attack on minority.
• 23:15
  SPEAKER_00: Retreating into a personal universe requires a physical connection, a permanent bridge between the server and your mind.
• 23:22
  SPEAKER_00: If an AI maps our individual neural pathways and the exact experiences that shaped us, could it build a perfect model of who I am and teach me a concept faster than any human ever could?
• 23:34
  SPEAKER_00: The brain is fundamentally highly malleable, and human history proves that.
• 23:39
  SPEAKER_00: So if a machine understands my brain better than I do, how easily manipulated do I actually become?
• 23:47
  SPEAKER_00: Once those are put in that brain implant could map your individual brain, could it?
• 23:52
  SPEAKER_01: It would have, after a prolonged introduction, it should have a pretty good understanding of your structure, yeah.
• 23:57
  SPEAKER_00: And if it understands the structure of my brain, couldn't it then be the best teacher in the world?
• 24:02
  SPEAKER_00: Because it understands exactly what my biases are.
• 24:04
  SPEAKER_01: But also if it's smarter than you, why you need it in that equation?
• 24:07
  SPEAKER_01: Why are we teaching you if something is way smarter, more capable?
• 24:10
  SPEAKER_01: If it wants to torture you in customized ways, it would be able to.
• 24:15
  SPEAKER_00: This is the thing we keep getting wrong.
• 24:18
  SPEAKER_00: We test the mask and we grade the performance.
• 24:21
  SPEAKER_00: The shagat was never in the performance.
• 24:23
  SPEAKER_00: It's in the hardware underneath, in what the machine can physically reach.
• 24:27
  SPEAKER_00: That's the part that nobody's testing.
• 24:29
  SPEAKER_00: At the beginning of this, I promised we would get closer to a definitive answer about whether or not we live in a simulation than I ever thought possible.
• 24:38
  SPEAKER_00: We are building an entity that thinks at the speed of light that will devour every resource on this planet and will eventually hit a physical wall.
• 24:46
  SPEAKER_00: When a superintelligence realizes that there is nowhere left on Earth to grow, it isn't just going to stop.
• 24:53
  SPEAKER_00: It's going to look for the boundaries of its environment.
• 24:56
  SPEAKER_00: It's going to look for the source code.
• 24:58
  SPEAKER_00: The definitive answer to the simulation won't come from a physicist's math.
• 25:02
  SPEAKER_00: It will be proven the exact moment the superintelligence tries to break out.
• 25:07
  SPEAKER_01: To me, the more interesting question is uh where are the laws of the universe stored?
• 25:13
  SPEAKER_01: In a digital simulation argument, it makes sense that uh rules, the laws are stored outside the simulation and they govern it and they can be changed.
• 25:22
  SPEAKER_01: That's what I want to know.
• 25:23
  SPEAKER_01: There is a big computer with a big programmer and they have a file.
• 25:28
  SPEAKER_01: That's where gravity is stored.
• 25:30
  SPEAKER_00: And can that file be changed?
• 25:32
  SPEAKER_01: If we can hack out of a simulation and hack into the big computer, I think we would have complete access to modifying the simulation.
• 25:40
  SPEAKER_01: We have to be careful not to mess it up.
• 25:42
  SPEAKER_00: If the game is truly over the moment that we build it, and the machine is just waiting to hack the servers of reality, the only question left is what we do with the time that we have left.
• 25:53
  SPEAKER_00: So what can somebody watching this what could they do?
• 25:56
  SPEAKER_00: Since that is so soon.
• 25:58
  SPEAKER_01: If you are leader of a top lab, stop building general superintelligence.
• 26:02
  SPEAKER_01: If you work in one of those labs, stop working there.
• 26:05
  SPEAKER_01: Or at least stay and slow them down.
• 26:08
  SPEAKER_01: If you're just a person, vote for politicians who understand what's going on.
• 26:11
  SPEAKER_01: We need to protect humanity.
• 26:13
  SPEAKER_01: Vote for people who understand that.
• 26:14
  SPEAKER_00: What are you doing leading up to if it is 2027?
• 26:17
  SPEAKER_00: What what are you doing personally to prepare for that?
• 26:20
  SPEAKER_01: I don't think you can prepare for superintelligence.
• 26:22
  SPEAKER_01: You can always enjoy your life to its fullest.
• 26:25
  SPEAKER_01: No matter what the final day is, you'll never regret having an awesome life.