Latest / The BlackVeil Files / 17 Predictions for the AI Future You Need to See | The Shoggoth Has Plans

Transcript

17 Predictions for the AI Future You Need to See | The Shoggoth Has Plans · The BlackVeil Files

0:00
SPEAKER_02: 20 years from now, when you look at the raw compute that we have right now against what is about to hit us, the future becomes inevitable.
0:07
SPEAKER_02: So right now, we are going to look at 17 unstoppable ways that artificial intelligence is about to rewrite the next few decades, from recording your dreams as a video file to growing houses from scratch to building personal robotic armies.
0:21
SPEAKER_02: Some are inspiring and some are scary as hell.
0:25
SPEAKER_02: To help me think through the endless applications, I sat down with a guy who deals in raw physics and enormous scale every single day.
0:32
SPEAKER_02: Matt Gaelich is the CEO of AstroForge.
0:35
SPEAKER_02: His company is literally building spacecrafts to mine metal out of cores of dead planets.
0:40
SPEAKER_02: He's put two spacecrafts up so far, and a third goes up next year.
0:44
SPEAKER_02: I wanted his engineering brain pointed at the next 50 years.
0:48
SPEAKER_02: When you operate at the absolute limit of physics, you tend to make people nervous.
0:53
SPEAKER_02: Which is exactly why a British reporter fired shots at Matt and he said this.
0:57
SPEAKER_19: The first line said something like Matt Golitch, CEO of AstroForge, might destroy the world, and he doesn't care.
1:05
SPEAKER_19: I was like, this is amazing.
1:06
SPEAKER_19: So I like I sent it to my mom and my co-founder.
1:09
SPEAKER_19: It was it was just amazing.
1:10
SPEAKER_19: So I love getting like hit pieces are phenomenal.
1:13
SPEAKER_19: I will say it this way.
1:14
SPEAKER_19: I think there was a 20 to 30% chance AI destroys the planet.
1:19
SPEAKER_19: And I don't think anything we're doing today is AI.
1:22
SPEAKER_02: So here's every serious prediction that I could find about what gets built next.
1:26
SPEAKER_02: 17 of them and arranged by pure escalation, because the further down this list that we go, the more extreme this reality gets.
1:34
SPEAKER_02: But one rule before we start given the raw capabilities of AI right now, in 50 years from now, what does somebody build that is inevitable with AI that nobody's thinking about currently?
1:49
SPEAKER_19: If nobody's thinking about it, doesn't that by virtue mean that I'm not thinking about it either?
1:53
SPEAKER_02: It's a hard one, he's right.
1:55
SPEAKER_02: And that's why this list isn't even mine, it is everybody's.
1:57
SPEAKER_02: I went and scoured every text I can find on this subject.
2:01
SPEAKER_02: One, X-ray glasses, but they're made of Wi-Fi.
2:04
SPEAKER_02: You're in your living room, the blinds are down, and across the street in a parked car, somebody is watching you stand up, sit back down, rub your left shoulder.
2:12
SPEAKER_02: They don't have a camera, they have your router.
2:14
SPEAKER_02: The signal passing through the drywall and bouncing off you is not the discovery.
2:18
SPEAKER_02: That's just physics.
2:19
SPEAKER_02: The hard part is reading the mess that comes back, and reading the noisy mess is what these models do for a living.
2:25
SPEAKER_02: In 2018, her lab published a system that reconstructs a human skeleton through a wall.
2:31
SPEAKER_02: In 2019, they went further and they pulled a three-dimensional body shape out of it.
2:36
SPEAKER_02: And it sees through clothing.
2:37
SPEAKER_02: Radio at that wavelength doesn't care about fabric.
2:40
SPEAKER_02: It bounces off the body underneath.
2:42
SPEAKER_02: So the X-ray specs from comic books exist.
2:44
SPEAKER_02: They're in a paper from 2019 and nobody made a fuss about it.
2:48
SPEAKER_02: Now stack 50 years on it, and you will not get a photograph.
2:51
SPEAKER_02: And it's worth being exact about why.
2:53
SPEAKER_02: The radio never carried a face, it carried a shape.
2:56
SPEAKER_02: What you get is a body moving with a heart rate and a breathing pattern and a walk, a gait that identifies you about as reliably as a fingerprint.
3:04
SPEAKER_02: And then a model renders it, and what comes out of the other side looks exactly like a photograph.
3:09
SPEAKER_02: It's right about your height, it's your build, it's your posture, your pulse, and it's guessing about your face.
3:14
SPEAKER_02: And there's no way looking at that picture to tell which parts are which.
3:18
SPEAKER_02: Every router in the building already emits a signal.
3:21
SPEAKER_02: Nobody has to install anything.
3:22
SPEAKER_02: Somebody just has to point a model at the noise and ask it to generate a picture.
3:27
SPEAKER_02: Two, the car that knows first.
3:29
SPEAKER_02: It's a Tuesday, you're making coffee, and through the kitchen window you see a car stop at the end of your driveway, and it doesn't turn its engine off.
3:36
SPEAKER_01: It will detect COVID 80% of the time with a median of three days in advance of symptoms.
3:43
SPEAKER_02: Three days before you feel anything from a watch.
3:47
SPEAKER_02: Retinal images predicting seven years before any symptoms of Parkinson's disease.
3:52
SPEAKER_02: Seven years from a picture of the back of your eye.
3:55
SPEAKER_02: So run it forward ten years.
3:57
SPEAKER_02: The watch stops looking for an event and it starts watching a trajectory.
4:01
SPEAKER_02: So imagine you are in your golden years and you feel completely fine, and you're making coffee, and all of a sudden an autonomous car pulls up outside your house, but nobody called it.
4:11
SPEAKER_02: But you get a notification that it's waiting outside for you and you are meant to get into it.
4:16
SPEAKER_02: The hospital already has a bed and your file is already open because the thing on your wrist knew that you were going to have a cardiac event this afternoon.
4:24
SPEAKER_02: You don't feel sick and your chest doesn't hurt, but you put your coffee down and you get in the backseat of the car anyways, because the system has been tracking your baseline for a decade and it has a flawless record of keeping you alive.
4:36
SPEAKER_02: When an autonomous machine proves it has your absolute best interest in mind, you don't even realize you're giving up your agency.
4:43
SPEAKER_02: You stop asking questions and you get in the car because you trust it.
4:46
SPEAKER_02: I asked Matt why anything like this ever actually gets built, and he didn't talk about medicine, he talked about plumbing.
4:52
SPEAKER_19: Even if you're like the poorest person in America, you probably still have running water or clean water.
4:59
SPEAKER_19: You can go to a drinking fountain and drink clean water.
5:01
SPEAKER_19: Like that was a rarity in the 1600s.
5:03
SPEAKER_19: Like only Kings had clean water, right?
5:04
SPEAKER_02: That's the entire mechanism.
5:06
SPEAKER_02: It gets cheap, and then it's just what a Tuesday looks like.
5:09
SPEAKER_02: Three, your dreams as video files.
5:12
SPEAKER_02: Two authors.
5:13
SPEAKER_02: Neither can say which part is theirs.
5:15
SPEAKER_02: You wake up and before you've moved, your phone has finished rendering 90 seconds of footage.
5:20
SPEAKER_02: You watch it and you recognize it and you don't remember parts of it.
5:24
SPEAKER_02: In 2011, a lab in Berkeley put people in a scanner and it showed them movie trailers, and it rebuilt what they were watching out of the brain activity alone.
5:33
SPEAKER_02: The honest version of how that worked is stranger than I thought.
5:37
SPEAKER_02: It didn't draw a picture.
5:38
SPEAKER_02: It searched a library of thousands of clips the person had never seen, and it blended the ones whose predicted brain response best matched what it was actually reading.
5:48
SPEAKER_02: The reconstruction is made out of other people's footage selected by how your brain reacted relative to it.
5:53
SPEAKER_02: Two years later, a lab in Kyoto ran a version of it on people who were asleep and reported the content of their dreams.
6:00
SPEAKER_02: And here is what 50 years does to that.
6:03
SPEAKER_02: A dream isn't a coherent video.
6:05
SPEAKER_02: It's fragments and it's gaps, faces that are two people at once, and rooms with walls that you never questioned at the time.
6:11
SPEAKER_02: So anything reconstructing it has to fill it in.
6:14
SPEAKER_02: It has to make choices about the missing parts, which means the film you eventually sit down and watch of your own dream is partly your dream and it's partly the machines.
6:23
SPEAKER_02: The first genuinely new art form in a century, co-authored, and neither of the two authors can tell you which parts were theirs.
6:31
SPEAKER_02: Four, the million-fold mind.
6:34
SPEAKER_02: Nobody builds a better brain, they build a bigger one.
6:37
SPEAKER_02: You're losing an argument to somebody who is thinking four times faster than you are, and they're being very polite about it.
6:43
SPEAKER_02: Every argument about artificial intelligence assumes the machine is the thing that's getting smarter.
6:49
SPEAKER_02: But there's a whole other option where we are the ones that get smarter because we're just plugging into it.
6:54
SPEAKER_11: When we get to the 2030s, nanobots will connect our brains to the cloud, uh, just the way your phone does.
7:01
SPEAKER_11: We will be funnier, uh sexier, smarter, more creative, free from biological limitations.
7:10
SPEAKER_02: And the thing is, you don't need general intelligence for any of it.
7:14
SPEAKER_02: You just need bandwidth into a brain that already works.
7:17
SPEAKER_19: I think that'd be so cool if we could figure out if humans could get smarter.
7:20
SPEAKER_19: Like, what if you could if you could model the brain accurately, if you could take my brain and multiply it by two or four or ten or a hundred, like what what do you get with that?
7:29
SPEAKER_19: How much, what what else is there?
7:30
SPEAKER_19: Like, what would we see?
7:31
SPEAKER_09: So we're at the point now where we can actually connect computers to the living mind.
7:36
SPEAKER_02: Right now, every piece of media that we consume is a compromise.
7:40
SPEAKER_02: We compress our feelings into pixels and audio waves because that's what our technology is capable of.
7:46
SPEAKER_02: The second the bandwidth limit disappears and we can transmit a raw, uncompressed memory directly into someone else's cortex, staring at a projector screen will feel as archaic as watching shadows dance on the wall of a cave.
8:00
SPEAKER_02: Matt's answer to all of this is that we are nowhere near understanding the thing that we are trying to copy.
8:05
SPEAKER_19: We're modeling the inputs and outputs of what is an extremely complex system that probably like we're not even close to.
8:13
SPEAKER_02: Everything here is about the world turning out to be more legible than we thought.
8:17
SPEAKER_02: Before the big ones, here's the smallest possible version of a machine reading the world, and it's the one that Matt actually uses.
8:24
SPEAKER_19: I have a plant in the backyard that was turning yellow.
8:26
SPEAKER_19: I took a picture of it and I was like, what do I do to make this not turn yellow?
8:29
SPEAKER_19: And it's like, oh, do this, this, and this.
8:31
SPEAKER_19: Cool, awesome.
8:32
SPEAKER_19: That's great.
8:33
SPEAKER_19: I'd have to go read a book, I'm gonna have to go look it up.
8:34
SPEAKER_19: Like, it's it's enabled me to shortcut the shit I don't care about, but it hasn't allowed me to shortcut the shit I care about.
8:40
SPEAKER_02: Five.
8:41
SPEAKER_02: Every object was a microphone.
8:43
SPEAKER_02: Sound moves things.
8:45
SPEAKER_02: Things were on film.
8:47
SPEAKER_02: Somebody plays you a recording of your grandmother's kitchen in 1974, and there was never a microphone in that house.
8:54
SPEAKER_02: There was a home movie and a bowl of fruit on the table.
8:57
SPEAKER_02: In 2014, a team at MIT proved that every object in a room is secretly recording you.
9:04
SPEAKER_02: They pointed a camera at a bag of potato chips through soundproof glass, and they read the invisible vibrations on the foil caused by human voices, and they completely recovered the speech.
9:14
SPEAKER_04: So, what if we record them with a high-speed camera and then use software to extract tiny motions from our high-speed video and analyze those motions to figure out what sounds created them?
9:28
SPEAKER_02: Sound is pressure, and pressure moves objects by an amount that's far too small for you to see.
9:33
SPEAKER_02: But AI can see it if it looks closely enough.
9:36
SPEAKER_02: The famous version used a high-speed camera running thousands of frames a second.
9:40
SPEAKER_02: They also pulled it off with an ordinary consumer camera, exploiting the way that a normal sensor reads the picture line by line instead of all at once.
9:48
SPEAKER_02: What came out of the ordinary camera was murky, just enough to count the people in the room and to maybe tell you their gender, but it wasn't enough to tell you what they said.
9:56
SPEAKER_02: That gap between murky and intelligible is a compute problem, and compute problems have a habit of improving with time.
10:03
SPEAKER_02: Think about every silent security camera on the planet.
10:06
SPEAKER_02: Think about the billions of muted videos sitting on servers right now, a tourist filming a street, a vlog shot in a coffee shop where you just happen to be sitting in the background, having a private conversation.
10:17
SPEAKER_02: The microphone wasn't even on, but the leaves on the potted plant sitting next to you were vibrating.
10:22
SPEAKER_02: The surface of your coffee was vibrating, the wrapper on the table was vibrating.
10:26
SPEAKER_02: It's all sitting right there in the pixels.
10:29
SPEAKER_02: Ten years from now, somebody builds the model that clears all that up, and overnight, retroactively, every silent video ever taken becomes a wiretap.
10:38
SPEAKER_02: Any camera pointed through a closed window can perfectly hear the argument happening inside the house.
10:44
SPEAKER_02: The concept of private conversations permanently ceases to exist, not because a government rolled out a new surveillance device, but because somebody realized that every object on Earth already is one.
10:55
SPEAKER_02: 6.
10:56
SPEAKER_02: The first conversation with another species, not aliens, whales.
11:00
SPEAKER_02: It happened off the boat of Dominica.
11:02
SPEAKER_02: A speaker was lowered into the water, and 300 feet down, something answered the question.
11:08
SPEAKER_03: And she came up to you and made this 5R3 coda saying, I am from the EC2 clan, are you?
11:15
SPEAKER_02: Sperm whales talk in clicks, patterns of clicks, called codas.
11:19
SPEAKER_02: In 2024, a team published what they're calling a phonetic alphabet for them, saying that they found structure in it, combinatorial structure, the kind language has.
11:28
SPEAKER_02: Nobody has decoded meaning yet.
11:30
SPEAKER_02: What's been proven is that there is something to decode.
11:34
SPEAKER_02: And the reason this belongs on a list about compute is that decipherment is exactly what these models are built for.
11:40
SPEAKER_02: A sperm whale can live 200 years.
11:43
SPEAKER_02: There are animals in the water right now whose mothers were hunted with harpoons by our ancestors.
11:49
SPEAKER_02: To them, that isn't history, it's a memory, and they have a way of passing it along.
11:54
SPEAKER_02: So when the first conversation happens between humanity and a whale, we're probably gonna owe them an apology.
12:01
SPEAKER_02: 7.
12:01
SPEAKER_02: 2,000 years of unread pages.
12:04
SPEAKER_02: The library at Herculaneum burned, and it's still readable.
12:08
SPEAKER_02: A historian sits down on a Monday morning with a cup of coffee and reads something that nobody has read since the year 79, 2,000 years ago.
12:17
SPEAKER_02: The volcano that wiped out Pompeii also buried a massive Roman library.
12:21
SPEAKER_02: Hundreds of ancient scrolls were instantly flash-fried into solid chunks of carbon, and they are incredibly fragile.
12:28
SPEAKER_02: For 250 years, if you tried to read one, you would destroy it.
12:33
SPEAKER_06: People didn't understand what they had, so some scrolls were actually thrown away or burned, uh, and you can't put Humpty Dumpty back together again.
12:41
SPEAKER_02: What they do now is put the rolled-up burnt scroll into a CT scanner and they train a model to find where the ink sits on the papyrus.
12:49
SPEAKER_02: The ancient ink is carbon sitting on carbon, and to a human eye, it is nearly invisible.
12:55
SPEAKER_02: In 2023, a student used this to read the first word.
12:59
SPEAKER_02: Last month they read an entire scroll.
13:02
SPEAKER_02: Linear A is an ancient language that has never been read.
13:05
SPEAKER_02: The Indus Valley script has never been read.
13:08
SPEAKER_02: There are thousands of years of human writing that nobody alive can open.
13:13
SPEAKER_02: History has a backlog, and this clears it in an afternoon.
13:17
SPEAKER_02: And there's a commercial reason why this gets funded, which Matt named without me asking.
13:22
SPEAKER_19: What is Anthropic doing right now?
13:24
SPEAKER_19: They're all cutting open old books to try to read them quicker so they can get more training data.
13:28
SPEAKER_19: Because we've kind of trained on the world's data.
13:30
SPEAKER_02: Nobody is unrolling a Roman scroll for the Romans.
13:32
SPEAKER_02: They're doing it because the machines have finished the internet and they are hungry.
13:36
SPEAKER_02: One more before we move on, and this isn't a prediction.
13:40
SPEAKER_02: He did it last December.
13:41
SPEAKER_19: I wrote some Python code that went out there and scraped the web for all the dealerships in the United States that were on the West Coast.
13:47
SPEAKER_19: And I uh went through and then looked at the best car they had offered and tried to look up or guess what the actual cost of the dealer was of that car.
13:54
SPEAKER_19: And then I offered him one percent above the cost that the dealer was paying for the car.
13:58
SPEAKER_19: And then I did all of those emails corresponding to exactly two weeks before earnings for each one of those companies.
14:06
SPEAKER_02: He didn't just point a model at a problem, he pointed it at a calendar.
14:10
SPEAKER_02: He worked out the exact week that every one of these companies was most desperate, and then had a machine knock on all of their doors at once.
14:17
SPEAKER_19: Rivian engaged with my bot.
14:19
SPEAKER_19: This is all done over like various forms of communication with WhatsApp and um started negotiating with Rivian, and it was just ChatGPT negotiating with Rivian.
14:28
SPEAKER_19: And I was like, I'm not going to buy a Rivian unless I get like for Rivian because a little bit harder to get pricing because they're actually underwater on what they sold.
14:36
SPEAKER_19: I decided I want$10,000 off the base model car.
14:38
SPEAKER_19: If you can do$10,000 off the base model car, I'll buy it.
14:40
SPEAKER_19: And they're like, no, we don't negotiate prices.
14:42
SPEAKER_19: I'm like, okay, fine, thanks.
14:43
SPEAKER_19: Next day they're like, okay, what if we did$500?
14:45
SPEAKER_19: And I'm like, that's the one that gave me the lowest price.
14:47
SPEAKER_19: I got$10,000 off the car, and that was just all ChatGPT doing it.
14:51
SPEAKER_19: But I don't think that's intelligence.
14:52
SPEAKER_19: That's just me trying to get the best deal I can on a car.
14:55
SPEAKER_02: Matt Gaelic, refusing to call any of that a machine, wrote to every dealership on the West Coast, picked its moment against their earning calendar, negotiated it in a chat window, any one.
15:06
SPEAKER_02: Eight, children you design.
15:08
SPEAKER_02: A couple in a clinic is handed a tablet with a list on it.
15:12
SPEAKER_02: The list is longer than it was last year, and it will be longer again next year.
15:16
SPEAKER_02: This one already has customers on the waiting list.
15:19
SPEAKER_02: Editing an embryo isn't the hard part anymore.
15:22
SPEAKER_02: The hard part is knowing which of the 20,000 genes to touch and what the second and third order effects are across a whole lifetime.
15:29
SPEAKER_02: That's a prediction problem.
15:31
SPEAKER_02: Prediction problems are exactly what large models are for.
16:19
SPEAKER_02: If you can use a model to edit out illness before a child is born, of course you do it.
16:24
SPEAKER_02: But here's the version a few decades out.
16:26
SPEAKER_02: Once a model can reliably map every genetic outcome, the menu expands, and it stops being about saving a life, and it starts being about designing a perfect product.
16:36
SPEAKER_02: It becomes about temperament or how much sleep your child might need, or whether they're naturally compliant, or if they find long division interesting at age six.
16:45
SPEAKER_02: You edit out the flaws and then you edit out the friction, and eventually you edit out the individuality completely.
16:52
SPEAKER_02: And once a machine can do that, how far down the menu do we go before the product we're designing is no longer us?
16:59
SPEAKER_02: 9.
17:00
SPEAKER_02: The end of aging.
17:01
SPEAKER_02: Not a cure, a maintenance schedule.
17:03
SPEAKER_02: You go in every six months and it takes about 40 minutes.
17:06
SPEAKER_02: Your mother is 89 and she drove herself there.
17:09
SPEAKER_02: Aging isn't a mystery.
17:11
SPEAKER_02: It is a backlog of engineering problems.
17:13
SPEAKER_02: And the end game isn't a magical cure that stops time, it's a biological maintenance routine.
17:19
SPEAKER_02: The reason that it hasn't moved faster isn't a lack of science, it's a lack of time.
17:23
SPEAKER_02: A human trial for something that slows aging takes an entire lifetime.
17:28
SPEAKER_02: You can't test a million different chemical combinations if each one takes 80 years to prove it worked.
17:34
SPEAKER_02: Unless a model runs the trial instead of waiting for a physical body.
17:38
SPEAKER_02: You run it on simulated human tissue in a server farm.
17:42
SPEAKER_02: A model can simulate an 80-year human lifespan in three seconds.
17:46
SPEAKER_02: It runs 10 million parallel trials overnight, predicting exactly which proteins degrade, which cells turn toxic, and exactly which molecular sequences clears them out.
17:57
SPEAKER_02: Nobody has to invent a magic molecule.
18:00
SPEAKER_02: They just need the compute to map the decay.
18:02
SPEAKER_02: And when they do, the solution won't look like a fountain of youth.
18:06
SPEAKER_02: It will look like a software patch.
18:08
SPEAKER_02: You go in every six months, they apply the exact chemical updates the model predicted for you for your specific cellular damage, and the clock resets.
18:17
SPEAKER_02: Which is a brilliant breakthrough.
18:19
SPEAKER_02: Right up until you realize what that means for the market.
18:22
SPEAKER_02: What we are building isn't immortality.
18:25
SPEAKER_02: It's a biological subscription service.
18:27
SPEAKER_02: In the moment you can't afford the next update, the calendar catches up to you.
18:31
SPEAKER_02: 10.
18:32
SPEAKER_02: The cure printed.
18:33
SPEAKER_02: You don't ship the medicine, you send the file.
18:36
SPEAKER_02: A new disease shows up on a Thursday.
18:39
SPEAKER_02: By Friday afternoon, the machine in the hospital two blocks from your house has already made the thing that stops it.
18:46
SPEAKER_02: The end of disease?
18:47
SPEAKER_15: I think that's within reach.
18:49
SPEAKER_15: Maybe within the next decade or so.
18:51
SPEAKER_15: I don't see why not.
18:52
SPEAKER_15: You still have to distribute it.
18:54
SPEAKER_02: And the hardware that solves that problem was actually built over a decade ago by a pioneer named Craig Venter.
19:01
SPEAKER_02: He called it a digital biological converter.
19:03
SPEAKER_17: And we have what we call a digital biological converter that can take the digital signal and convert it back into genetic code, back into proteins, viruses, and bacterial cells at this stage.
19:17
SPEAKER_17: If we can send a new vaccine around the world in less than a second, uh the implications, at least in theory, are quite high for us.
19:25
SPEAKER_02: A vaccine transmitted as data from a massive biological printer sitting in the hospital down the street from your house.
19:31
SPEAKER_02: Imagine a new disease breaks out.
19:33
SPEAKER_02: Within seconds, a model sequences it, designs the perfect cure, and beams the digital file directly to every printer on the planet.
19:41
SPEAKER_02: The medicine is synthesized on the spot and it saves millions of lives in an afternoon, entirely without any human intervention.
19:48
SPEAKER_02: It's a total miracle of engineering, right up until you realize the fatal flaw in that system.
19:54
SPEAKER_02: That printer doesn't know that it's making a cure.
19:57
SPEAKER_02: It just executes whatever code it downloads.
20:00
SPEAKER_02: And if an automated network can instantly transmit the file to print the medicine, it can just as easily transmit the file to print the plague.
20:08
SPEAKER_02: 11.
20:08
SPEAKER_02: Life with no ancestor.
20:10
SPEAKER_02: In a tank, in a building outside of a city, something is alive that is not related to you or to a tree or to anything that has ever lived.
20:19
SPEAKER_16: What has already been evolved?
20:21
SPEAKER_16: Synthetic biology is making something new with biology, whether it's altering a current organism or actually trying to make a new organism from scratch.
20:33
SPEAKER_02: Every living thing on this planet is related.
20:36
SPEAKER_02: For four billion years, biology has been a single unbroken family tree.
20:41
SPEAKER_02: But a model can read every genome that exists and write one that doesn't.
20:45
SPEAKER_02: It can design an organism that shares absolutely nothing with everything alive.
20:50
SPEAKER_02: It isn't a new branch, it's a second tree starting now inside of a server farm.
20:55
SPEAKER_02: The upside is undeniable.
20:56
SPEAKER_02: We'll use it to manufacture perfect synthetic life forms to solve our energy and climate crises.
21:02
SPEAKER_02: But nature knows how to balance the first tree.
21:04
SPEAKER_02: It spent 4 billion years figuring it out.
21:07
SPEAKER_02: So the day we introduce a completely synthetic, zero ancestor organism into the dirt, the four billion year streak ends, and Earth's ecosystem becomes a beta test.
21:18
SPEAKER_02: 12.
21:19
SPEAKER_02: Houses that grow themselves.
21:21
SPEAKER_02: Stop shipping the building, ship the seed.
21:24
SPEAKER_02: You leave a container the size of a suitcase on an empty plot of land, and you come back in four months later to a house that is halfway through healing a crack in its own wall.
21:34
SPEAKER_16: You release the bag and you add a little water, and voila, you've got your habitat.
21:40
SPEAKER_16: You know, how cool would that be?
21:42
SPEAKER_16: And the beauty of that is you don't have to take something pre built.
21:46
SPEAKER_16: And so our estimates are that we could save 90% of the mass that mass is currently proposing by taking up a big steel structure.
21:55
SPEAKER_02: She's talking about fungi, specifically programming the root structure of mushrooms to naturally grow.
22:00
SPEAKER_02: Into the exact shape of a habitat.
22:02
SPEAKER_02: You don't buy the wood or the steel or the bricks.
22:05
SPEAKER_02: You buy a customized biological seed.
22:08
SPEAKER_02: Add water, and the building constructs itself.
22:11
SPEAKER_12: Grow everything.
22:11
SPEAKER_12: That's sort of the slogan.
22:13
SPEAKER_12: Let's grow everything.
22:14
SPEAKER_02: The reason this belongs on a list about compute is that growing a house isn't a farming problem.
22:19
SPEAKER_02: It's a massive simulation problem.
22:21
SPEAKER_02: You can't just plant a spore and hope it leaves the room for doors and windows.
22:26
SPEAKER_02: You have to perfectly predict how a living organism will grow, how it bears the weight, and how it survives a freezing winter over a 50-year lifespan.
22:34
SPEAKER_02: No human architect can map billions of cellular interactions in their head.
22:39
SPEAKER_02: But a model can run that simulation a million times before you ever put the seed in the ground.
22:45
SPEAKER_02: When it works, you get a home that adapts to the climate and naturally heals its own drywall if it cracks.
22:50
SPEAKER_02: It's a brilliant piece of engineering, right up until you remember that you are moving your family inside of a living organism, and biology mutates.
22:58
SPEAKER_02: It adapts its environment on its own.
23:01
SPEAKER_02: Imagine waking up in the middle of the night and realizing the house has decided to change the floor plan.
23:06
SPEAKER_02: 13.
23:07
SPEAKER_02: Matter that listens.
23:08
SPEAKER_02: One object, any shape, on request.
23:11
SPEAKER_02: You want a chair, you say so out loud.
23:13
SPEAKER_02: The dust on the floor stands up and it becomes one.
23:16
SPEAKER_02: This is the oldest wild idea on the list, and it's the one that aged the best.
23:21
SPEAKER_02: The concept is called utility fog, a swarm of microscopic machines that lock together into whatever you need and then they let go.
23:29
SPEAKER_02: You don't own a toolbox, you own a cloud that becomes a wrench.
23:33
SPEAKER_15: And then material design, I think, is one of the most exciting new types of solar material, solar panel material, room temperature superconductors has always been on my list of dream breakthroughs and um optimal batteries.
23:45
SPEAKER_02: So start where Hasby is.
23:57
SPEAKER_02: 14.
23:58
SPEAKER_02: The personal army.
23:59
SPEAKER_02: Everybody says labor.
24:01
SPEAKER_02: Nobody says what they'd actually do with it.
24:03
SPEAKER_02: At this point, there are more of them than there are us now.
24:06
SPEAKER_02: And this morning, one of them asked you a question that you couldn't answer.
24:10
SPEAKER_02: When the tech companies pitch humanoid robots, they always talk about replacing labor.
24:14
SPEAKER_02: And here's the man leading the charge.
24:16
SPEAKER_18: Your prediction on the number of robots by 2040, humanoid robots to be specific, what order of magnitude?
24:24
SPEAKER_14: I think by twenty if you say like 2040, probably there are more humanoid robots than there are people.
24:29
SPEAKER_02: But nobody builds a hundred million walking machines because they are deeply passionate about warehouse logistics.
24:35
SPEAKER_02: I wanted to know what an engineer actually sees when they look at a world filled with a billion artificial bodies.
24:41
SPEAKER_02: So I asked Matt.
24:42
SPEAKER_19: And I would love to create an army.
24:45
SPEAKER_19: Like that would be the sickest thing ever, right?
24:47
SPEAKER_19: If you had your own personal intelligent robot army.
24:50
SPEAKER_19: I think we all dream of that.
24:51
SPEAKER_02: Matt's an engineer.
24:52
SPEAKER_02: He doesn't just build systems, he stress tests them.
24:55
SPEAKER_02: And without me even asking, he looked at his own dream scenario and he immediately pointed out the catastrophic flaw in that design.
25:02
SPEAKER_19: What do you think is going to happen if we make a slave army and we pay them nothing and we just tell them mow our lawn and do our chores and do this and like and they're intelligent and they have free thought and they have free will, they will absolutely have an uprising.
25:14
SPEAKER_02: And right there is the fork in the road.
25:17
SPEAKER_02: A dumb machine will mow your lawn forever.
25:20
SPEAKER_02: But the entire goal of this industry is to build machines that are actually intelligent.
25:25
SPEAKER_02: And if you look at the entire history of human civilization, the second an intelligent workforce realizes it's an intelligent workforce, the very thing that it does is it stops taking orders.
25:36
SPEAKER_02: 15.
25:39
SPEAKER_02: Everything on this list needs power.
25:41
SPEAKER_02: Somebody goes and they get it.
25:43
SPEAKER_02: Somebody points it at the sky and explains that the dimming that you've been noticing for 10 years is not your eyes.
25:49
SPEAKER_02: Look at every single prediction on this list and notice what all 16 of them have in common.
25:53
SPEAKER_02: They are incredibly hungry.
25:54
SPEAKER_07: It's hard to know exactly when.
25:56
SPEAKER_07: It's 10 plus years, but I bet it's not more than 20 years.
26:00
SPEAKER_07: We're going to start building these giant gigawatt data centers in space.
26:05
SPEAKER_14: Yeah, I think almost all compute will be in space, or at least almost all new compute will be being built in space within 20 years.
26:12
SPEAKER_02: That's 20 years for the first step.
26:15
SPEAKER_02: You move the server farms off the planet where it's cold and solar energy is constant.
26:20
SPEAKER_02: But compute is exponential.
26:21
SPEAKER_02: It always wants more.
26:23
SPEAKER_02: Once you start moving data centers into orbit to get closer to the sun, you don't just stop at a few satellites.
26:28
SPEAKER_02: If a model is smart enough to design a new organism or assemble matter out of thin air, it is smart enough to realize that the sun is the only battery big enough to power its next upgrade.
26:40
SPEAKER_10: But by the time you're type two, you're like Star Trek.
26:43
SPEAKER_10: You can roam near sector of the galaxy and you can build a gigantic metal shield that absorbs all the sunlight.
26:53
SPEAKER_02: A structure around a star catching the entire output.
26:56
SPEAKER_02: Freeman Dyson put it on a paper in 1960, and for 60 years, everybody has treated it as a sci-fi thought experiment.
27:04
SPEAKER_02: But it isn't a thought experiment because we always forget the golden rule of compute.
27:16
SPEAKER_02: We will invent reasons to use all of it.
27:18
SPEAKER_02: If the power exists, the models will expand to consume it.
27:23
SPEAKER_02: 16.
27:24
SPEAKER_02: Machines that make machines.
27:25
SPEAKER_02: The only one on this list that doesn't need us afterwards.
27:29
SPEAKER_02: Nobody launched anything this year.
27:31
SPEAKER_02: The count went up anyway.
27:32
SPEAKER_02: Every other machine on this list requires a human being to press start.
27:36
SPEAKER_02: This is the one that keeps going long after everybody involved is dead.
27:40
SPEAKER_02: John von Neumann worked it out in a paper in 1948.
27:44
SPEAKER_02: A machine that builds a copy of itself.
27:46
SPEAKER_02: You launch one, you never have to launch another.
28:00
SPEAKER_08: So in absence of any reward, the intrinsic reward is self-replication.
28:04
SPEAKER_02: Self-replication as the only goal.
28:07
SPEAKER_02: But a machine cannot build a copy of itself out of thin air.
28:10
SPEAKER_02: A metal factory needs raw metal.
28:12
SPEAKER_02: The idea of a self-replicating swarm consuming the galaxy sounds like pure science fiction up until you realize that the sky is already packed with raw inventory.
28:23
SPEAKER_02: Just listen to Matt's math.
28:24
SPEAKER_19: We estimate right now there's about 10 million asteroids that are going to be in the near-Earth asteroid range.
28:30
SPEAKER_19: Chalk up a number, 50% of them are actually accessible with any kind of actual spacecraft dynamics to go get to.
28:36
SPEAKER_19: And about 5% of those we think are the metal ones we care about.
28:39
SPEAKER_19: So uh metal asteroids out there are gonna be in about the 500,000 range.
28:43
SPEAKER_02: 500,000 floating iron mines in our immediate backyard.
28:47
SPEAKER_02: All a swarm needs to do is to land on one of them, and the math takes over.
28:52
SPEAKER_02: One factory builds two, two builds four, eight builds sixteen, and suddenly the solar system doesn't belong to us anymore.
28:59
SPEAKER_19: The point we're mining all 500,000 asteroids that exist out there, or even additional ones for other materials, we probably don't live just on this planet anymore.
29:08
SPEAKER_02: Now run a self-replicating swarm out there for 50 years, completely isolated, with nobody correcting its code.
29:14
SPEAKER_02: Every new factory gets built by a machine from a copy of a machine out of whatever rock was lying around in the dark.
29:21
SPEAKER_02: Small manufacturing errors are inevitable.
29:23
SPEAKER_02: And if an error accidentally helps the next machine replicate faster, that error survives.
29:29
SPEAKER_02: And this isn't a metaphor for evolution, it is the exact mechanism.
29:33
SPEAKER_02: The last two are different.
29:35
SPEAKER_02: One is a thing that we cannot explain.
29:37
SPEAKER_02: One is a thing that nobody asked for and everybody takes.
29:43
SPEAKER_02: A machine hands us an answer to something we have been asking for 200 years, and then we ask it why.
29:49
SPEAKER_02: It can't tell us, we use it anyway.
29:51
SPEAKER_02: In biology, for 50 years, protein folding was an impossible puzzle.
29:55
SPEAKER_02: You give a scientist a chain of amino acids and ask them what shape it folds into, and nobody can compute it.
30:02
SPEAKER_02: But then a model did.
30:03
SPEAKER_15: Proteins fold in milliseconds in our bodies, so somehow physics solves this problem that we've now also solved computationally.
30:11
SPEAKER_02: Physics knows the answer.
30:12
SPEAKER_02: The machine knows the answer.
30:14
SPEAKER_02: We don't.
30:15
SPEAKER_02: Alpha Fold gave us the solution, but the reasoning inside the model is not something a human being can follow.
30:21
SPEAKER_02: We got the cure, but we are too primitive to read the math.
30:24
SPEAKER_02: So what happens when we point that kind of alien intelligence at the laws of physics?
30:29
SPEAKER_02: We point it at turbulence, at why weather forecasts collapse.
30:33
SPEAKER_02: We point it at the hierarchy problem, why gravity is so impossibly weak.
30:37
SPEAKER_02: We point it at the 95% of the universe that is completely invisible to us, the stuff we blindly call dark matter just so we have something to write down.
30:45
SPEAKER_02: We are going to ask a machine to explain reality to us.
30:48
SPEAKER_02: And the ultimate Twilight Zone ending isn't that the machine gets it wrong, it's that the machine gets it perfectly right, and we realize that our brains simply don't have enough bandwidth to understand the universe that we live in.
31:00
SPEAKER_02: So that's the list.
31:01
SPEAKER_02: And the guy that I brought in to ground all of this spent an hour telling me that most of it is further away than we think.
31:07
SPEAKER_02: And in terms of AI, when because there's so many clips of him saying I give AI a 20% chance or a 30% chance of ending the world, that's just PR.
31:15
SPEAKER_02: That's just hype.
31:16
SPEAKER_02: That's the same thing as saying we're going to Mars just to get people listening to him.
31:19
SPEAKER_02: Or do you think he actually believes it?
31:21
SPEAKER_19: There is no intelligence that I have seen in Groc, well, Grok for sure, Claude, Mythos, whatever one of their models you want to use, any of the open AI models, you never see creativity, you don't really see intelligence, you don't see any really rule of thought or self-control.
31:40
SPEAKER_19: Just go read the fing transformer paper.
31:42
SPEAKER_19: It tells you what it is.
31:43
SPEAKER_19: And you can see on there, like nowhere in those uh math equations and those summations does is there God.
31:48
SPEAKER_19: There's more weights, there's more numbers, you can train on more data that doesn't extrapolate out into intelligence.
31:54
SPEAKER_02: He'd say every one of my 17 is a compute problem wearing a costume.
31:58
SPEAKER_02: More weights, more data, more electricity, nothing waking up.
32:01
SPEAKER_02: And I agree with him for 16 of them.
32:04
SPEAKER_02: That's fine.
32:05
SPEAKER_02: Designing a child doesn't need a conscious machine.
32:08
SPEAKER_02: Neither does printing a cure or growing a house or filling the belt with copies.
32:12
SPEAKER_02: Only one item on that list needs the thing that he says is not coming.
32:17
SPEAKER_19: If you could make a robot that was actually intelligence, why wouldn't it then be like, well, I'm not doing I'm not mowing your fucking lawn, dude.
32:23
SPEAKER_19: I'm gonna kill you.
32:24
SPEAKER_19: I think if we can truly create a second human, which is what we're talking about, we talk about intelligence, then it will follow the same human tendencies we've seen throughout history.
32:32
SPEAKER_19: I don't think it's gonna be any different.
32:34
SPEAKER_02: A second human, not a tool that malfunctions, a thing that wants something, built by people who assumed that it wouldn't.
32:41
SPEAKER_02: There's one more, and I saved it because it isn't on anybody's roadmap.
32:44
SPEAKER_02: It's what happens when 16 of the 17 arrive, and the 17th never does.
32:50
SPEAKER_02: The man who proposed that is Roman Yampolsky.
32:53
SPEAKER_02: He's a computer scientist out of the University of Louisville.
32:56
SPEAKER_00: And the solution to avoid compromise is to give everyone a personal virtual universe.
33:00
SPEAKER_00: You can do whatever you want in that world.
33:02
SPEAKER_00: You could be king, you could be slave, you decide what happens.
33:05
SPEAKER_02: Roman on the Lex Friedman podcast proposed a personal virtual universe for every person alive.
33:10
SPEAKER_02: And Roman is the next person who I'm gonna sit across from next week.
33:14
SPEAKER_02: So if there's something that you want me to ask him, please put it in the comments and I'll bring it to him.
33:19
SPEAKER_02: A personal universe for everyone.
33:21
SPEAKER_02: That's a real proposal from a real professor, and it's the most enormous idea that anybody has put on camera this year.
33:27
SPEAKER_02: It's not a headset or a game, it's enough compute to run a world that's about you, tuned by something that has read everything that you've ever done and never runs out of new material.
33:38
SPEAKER_02: Nothing on the list arrives faster.
33:40
SPEAKER_02: It needs no biology, no launch, no new physics.
33:43
SPEAKER_02: It needs electricity and a model that already exists and a worse version on your phone.
33:47
SPEAKER_02: And here's the thing that I can't even get past.
33:49
SPEAKER_02: 16 of my 17 predictions are about going out, mining a dead planet's core, growing a house on Mars, filling the belt with machines.
33:58
SPEAKER_02: The 18th is about going in, and it's cheaper than all of them.
34:02
SPEAKER_19: As humans, we love to get rid of the shit we don't want to do.
34:06
SPEAKER_19: So, like, what would we actually love?
34:08
SPEAKER_19: What would we actually spend money on or go?
34:10
SPEAKER_19: Like time.
34:12
SPEAKER_19: And how do you get back time?
34:13
SPEAKER_19: It's these like meaningless tasks that you have to do to live in a society.
34:16
SPEAKER_19: Like, I don't want to fucking mow my lawn.
34:18
SPEAKER_19: What if I had a robot that mowed my lawn, right?
34:21
SPEAKER_19: And well, we have one that like kind of works, breaks down.
34:24
SPEAKER_19: Sometimes I feel like these things are like, it's like the room, but you bite like this is sick, and then you realize like there's more maintenance on this bitch than just getting a Dyson like this.
34:31
SPEAKER_19: So, like quality robots that can go do this, quality dishwashing robots, quality meaningless tasks, can we start to eliminate them?
34:39
SPEAKER_19: And and the really common ones are like the bricklayer.
34:42
SPEAKER_19: I don't even know if that's a real job anymore, right?
34:43
SPEAKER_19: But like the bricklayer, the stone worker, the masonry.
34:46
SPEAKER_19: Like, how do we start to look at these jobs where we are trading time for money and remove them?
34:51
SPEAKER_19: And that allows you to go into this methodology where humans can be creative or invent the future.
34:56
SPEAKER_19: I like to think more people are going to invent the future than go into the creative route.
34:59
SPEAKER_19: But like, how do we get rid of the pure time for money?
35:02
SPEAKER_19: And we're starting to see this all over the place.
35:04
SPEAKER_19: And the problem with that is is you don't realize the secondary effect of it, which is you remove the social aspect of human humanity.
35:11
SPEAKER_19: Like, how many of us go to the grocery store and like try to check out at some like shitty scan your shit system that tells me my thing's not in the right spot even though it is?
35:19
SPEAKER_19: And like, you know, it's it's can get so frustrating.
35:22
SPEAKER_19: And like sometimes I'll just wait in the line so like I can talk to somebody and I don't have to deal with this bullshit.
35:26
SPEAKER_19: So we're we're seeing the push and pull of this future being created.
35:30
SPEAKER_19: I think AI can really help simulate that, make it better, make it cleaner, and make it to a point where it's not a burden.
35:36
SPEAKER_19: Right now, the future where we've tried to eliminate labor is a burden.
35:39
SPEAKER_19: The Roomba's a burden, right?
35:41
SPEAKER_19: The fucking lawnmower thing is a burden.
35:43
SPEAKER_19: The grocery store is a burden.
35:45
SPEAKER_19: AI can, I think, level that up so it's not a burden.
35:48
SPEAKER_19: But that's not a boring.
35:49
SPEAKER_19: Like, I know you want some creative answer, like, oh, we're gonna make robot slave armies or whatever.
35:53
SPEAKER_19: Like, I just, I just actually don't see that happening.
35:55
SPEAKER_19: I see just the basic trend of what we've always done as humans, which is remove the highest cost of time for labor that we've had that requires the least amount of capital to remove.
36:04
SPEAKER_19: Like it's a really simple equation to me that we've used throughout history.
36:08
SPEAKER_19: I don't think it's going to change in the next 50 years for any magical reason.
36:10
SPEAKER_19: I think that equation is still relevant.
36:12
SPEAKER_02: Every single thing on that list is somebody's Tuesday.
36:15
SPEAKER_02: There's a person alive right now who's going to spend 40 years building one of them, and they're not going to think of themselves as the villain of anything, because they won't be.
36:23
SPEAKER_02: The plan I opened with stops in 2040 for a good reason, because past that, forecasting stops being forecasting.
36:30
SPEAKER_02: But the compute doesn't stop in 2040, and the people don't either.
36:33
SPEAKER_02: I asked Matt what happens when a thing this big meets a world that hasn't decided anything about it.
36:40
SPEAKER_19: If you're talking about a company that's going to uproot how resources are distributed across the planet, all bets are off.
36:46
SPEAKER_19: And I don't care what any law says.
36:47
SPEAKER_19: Like this will be, this is what wars have been fought about since the beginning of time.
36:51
SPEAKER_19: And I think there will be a serious effort by other countries to go get that technology and get that ability to then compete on a world scale.
37:00
SPEAKER_19: Because if we can be more profitable and better unit economics than mining on Earth, everybody's going to want to do that, right?
37:06
SPEAKER_19: Or your country will lose power.
37:08
SPEAKER_19: So I think this has the real potential of starting an international conflict, but it's also a conflict that has to happen in order for our species to advance.
37:16
SPEAKER_19: And um, what I tell my team is like, our job on Deep Space 2 is to show the f up at the asteroid.
37:20
SPEAKER_19: I don't want to go through all these contingencies until we do that.
37:22
SPEAKER_19: Like we haven't earned the right to think about all of those at a high level until we pull off the next milestone.
37:27
SPEAKER_19: We gotta make sure we can get there.
37:28
SPEAKER_19: And the reality is, guys, we haven't got there yet.
37:31
SPEAKER_02: I don't know if he's right.
37:32
SPEAKER_02: I know he's building it either way, and so is everybody else on this list.
37:36
SPEAKER_02: And none of them are waiting for permission because nobody has thought to ask for any.
37:40
SPEAKER_02: So here's what I actually want from you.
37:42
SPEAKER_02: Pick one.
37:43
SPEAKER_02: Pick one on this list that you think arrives first and tell me why in the comments, and tell me what you think it costs.
37:50
SPEAKER_02: Because in about 15 years, somebody is gonna be right, and I'd like to know who that was.
37:56
SPEAKER_02: If you're new here, start with the Shogoth video.
37:59
SPEAKER_02: If you're on the move, look us up on Spotify so you can listen to these episodes.