The idea of space colonization has captured the human imagination for centuries, but only recently has anyone tried figuring out how it might look in practice. In this talk from Effective Altruism Global 2018: San Francisco, Stuart Armstrong explores how a technologically mature humanity might explore the cosmos. 

A transcript of Stuart's talk is below, including questions from the audience, which we have lightly edited for clarity. You can also watch the talk on YouTube and read it on effectivealtruism.org.

The Talk

So, this talk is entitled, “The Far Future of Intelligent Life Across the Universe.” But it might also be entitled, “The Surprisingly Near and the Almost Unimaginably Far Future of Intelligent Life Across the Universe.” Or “The SNatAUF Future”, as we refer to it.

But the Almost Unimaginably Far will be a brief parenthesis at the end. The Surprisingly Near is that it is very, and please notice the quotes, “easy” to expand across the universe and colonize every single galaxy that we could ever reach.

There's two points to the talk. The fun, technical stuff is the second. The first is why had nobody put these numbers together before I did?

This is arguably the oldest science fiction book from the second century, where people on a boat ended up on the moon where there was a war with the empire of the sun and various things happened. As you can tell, it's not exactly the hardest sci-fi that you can get.

But we've had galactic empires. We've had harder sci-fi and we've had things that look at, say, the economics or the social aspects or the environmental aspects of spreading across the galaxy. But it was a surprise to me that no one had sat down and said, “Okay, colonizing the universe, how would we do it? What are the numbers that we'd look at?”

Robert Bradbury started that. This is his table on disassembling planets, but he died before taking that any further, and I used his numbers with great gratitude. Especially the ones for the planet Mercury.

So that's the first lesson of this talk, that there are questions out there that can be analyzed that have not been. There's still important unanswered questions that people have a vague, intuitive idea as to how space colonization might work, but no one's actually sat down to write it out. And there's probably going to be things like that, and things far more near term than space colonization.

And now on to the sort of second lesson of the talk that was just really fun, collecting all the numbers, putting the pieces together from the different fields of physics and engineering and seeing what I could do with it.

So, the challenge. Humanity wants to colonize the universe. How? Energy and materials? Probe designs? Where do we go? How long does it take? The basic questions. This is going to be using the methods of so-called exploratory engineering, which means that we know what we can't do, we know what we can do, we're kind of going to move into this nebulous area there.

And obviously, so if we assume, say a few thousand years, which rounds to zero on a cosmic scale, what could humanity achieve if we continue as a technological civilization?

Now again, we shouldn't turn to stories here, but we're going to use some guiding principles. The first one is that if it's been done in nature, we'll be able to do it ourselves at some point. So I'm assuming some form of AI and some form of replication will be doable. Because we're really quite good at co-opting nature or copying it eventually.

And the second thing is that, I'm just going to assume the task can be automated, because automation is something we're really good at. And what this means is that scale is not an insurmountable barrier. Just because something is big, doesn't mean that we can't do it. We'll get to that.

So the real limiting factors are going to be: energy, material and time. So, the first thing: energy.

Now, the obvious thing is we have a sun. Let's Dyson it. There's various designs for Dyson spheres, the one you see in science fiction is always the shell, which would tear itself apart and would just plunge into the sun if you actually tried to do it. A swarm, on the other hand, consists of lots of orbiting solar collectors. Or you could have a bubble, which consists of solar sails and panels, that are just sort of floating, using the energy of the sun itself to stay in orbit.

We're going to go for the swarm, which is the simplest and easiest design. And here I'm going to always assume conservative options. When I get to the part about disassembling planets, you may question the use of the term conservative, but actually the conservative assumption is that we can take planets apart. The daring assumption is that we'll have super materials so we won't need to do it.

We need a convenient source of material for this, close to the sun. Hmm. As I said, sorry Mercury, it's nothing personal. But here are some numbers. The mass of Mercury, let's assume 50 percent of the mass is usable, mainly hematite oxygen and iron. As I said, this is a conservative assumption. We'll probably be able to use most of Mercury if we get this done.

A sphere in Mercury's orbit, using its useful mass, would have about half of a millimeter thickness. And that is plenty, because what we're mainly thinking is huge mirrors. That's the majority of the Dyson Swarm. Huge mirrors floating in space, orbiting, concentrating solar light down on some engine or some cell that actually does the extraction of the energy.

And this is, I say, if we allow ourselves to dream a bit here, and look at more extreme materials, then we can get much thinner swarms, which means we could use a large asteroid rather than a planet. And the sort of, various Mercury Conservation Societies, which will no doubt spring up, will probably be grateful to us for that.

In this diagram, we get energy. We mine stuff on Mercury. We get that stuff into orbit. We make solar collectors from that stuff. And we use that to get energy. And the cycle continues. And we build mining equipment and factories as we go along.

Completing the loop is the main assumption here. If you get this loop, you get exponential feedback. And then the details don't really matter. If you don't get the exponential feedback, you won't be able to take Mercury apart in any reasonable amount of time. But if you do get the loop, even if it's an inefficient loop, you will.

So, what I assumed was that it took five years to process the material and move it into orbit and get useful energy out of it. As I say, 50 percent of the material usable, 1/10 efficiency for moving the material off the planet. 1/3 efficiency for solar captors. And initially, our seed is one squate kilometer of solar captors on the surface of Mercury.

This seed by the way is something that we could do, I don't know, in 15 years, if the entire world felt like it. So, what happens then? Well, on a log scale, this is the amount of power that we get. So for the first five years, nothing happens because material is just being placed. Then we get a leap as the first generation of the Dyson Swarm comes online. Then a second leap as the second generation comes online. The one made by this one. And then it smooths out gradually. And I stopped the curve at that point because if you look at the mass of Mercury, that's what it's done in the same amount of time.

As long as you can close the exponential loop, you can can take Mercury apart in, well, this is completely unrealistic because if we do this, we would do it in 15 years or less. We probably won't do it this way. As I said, we'll probably use large asteroids, but realistically, if we can get the feedback, it'll be much faster than this.

So, that deals with getting the energy. By the way, this made it into the Daily Mail on one of those infuriating, technically correct, but completely missing the point, articles. It said, first of all, that scientists aim to begin projects within 25 years. I don't know who those scientists are. I'd like to meet them. But also, it claimed that this would solve the world's short-term energy problems, which, yes, it would. I mean, a nuclear bomb solves your lack of heat. This has kind of completely missed the point.

But anyway, we've got the energy. Now we need to launch. And what are we going to launch? Well, we have various mass drivers, quench guns, or lasers that we should point at solar sails from our probes. I'm imagining a mixture of the two. And then we launch.

What do we send? Well, we send the so-called von Neumann probes, self replicating probes. Here's a design for a von Neumann probe, take a happy couple, add more happy couples, give them stuff to eat. Give them data, manufacturing capability, add an engine, wrap it all up, and here we have a von Neumann probe that can eventually create copies of itself. Extremely inefficiently, hundreds of tons, limited acceleration and the speed of replication is the speed of human reproduction, and maturation.

There are some designs for a self-replicating lunar factory. NASA was doing some cool stuff in the 80s. But if you adapt this design you get a 500 ton payload. This is far too much, but this is a useful upper bound for what we might have in mind.

And this is the smallest design, the self-replicating, Merkle-Freitas HC molecular assembler.

But what I said, and what we're assuming is that we can co-opt or copy nature. And in nature, we have vibrio comma, the sort of smallest general environment replicator at 10 to the -16 kilograms. We've got E. coli, which is much more robust, the smallest seed is at 10 to the -9 kilograms. That's interesting because seeds create microscopic structures. The smallest acorn is a gram. And the acorn is very interesting because this is a potential huge factory, solar powered huge factory for the production of more acorns.

If we copy that design we can sort of give it a leg up, because we can give it extra energy at least at the beginning. Like give it some nuclear power. You need legs, drills, roots, to extract resources. Again biology has this so we can assume that our probes will also have this. And I'm going for a final mass of 30 grams for this. Many people think this is ridiculous, but it's not clear whether people think it's ridiculously high or ridiculously low.

Data storage limits, I'll skip over this. Basically it's that there's ample space to store as much as we want, including the whole population of earth as uploads, some compressed uploads if we felt like it.

Now, decelerating, I'm going to look at using a rocket. What I've recently discovered is that you can escape the tyranny of the rocket equation, which I though you couldn't. If you decelerate using a gun, you can actually do it more efficiently than if you use a rocket.

If you wanted, in the questions, you can ask me the difference between using a gun or a rocket for deceleration. So these are the, again, conservative estimates. This is the relativistic rocket equation. It is incredibly nasty. Final mass of that. And here, basically I'm looking, you need to decelerate from various speeds, using various engines. How much reaction mass do you need to end up with your 30 gram replicator?

Matter - anti-matter, fusion, and fission. And I'm looking at the three scenarios across the diagonal. So this is the amount of reaction mass that you need to decelerate. And rounding these up a bit, are model probes of five kilograms, 15 tons, and 35 tons, entered at these various speeds of launch.

There's another way of decelerating, which is basically, don't bother. The Hubble drag means that if you aim for the most distant of distant galaxies, by the time you get there, you'll be moving with practically no relative velocity. This also means that for other galaxies, you'll arrive slower than your launch speed, so deceleration also becomes easier. So, I'll also model this scenario.

Now this is where the sort of social, or story model of science fiction leads us astray. Colonization is often thought of, well, not even at the galactic level. You start at a star, you go to another star, you get resources, you spread across. This is basically the European age of exploration ported to space.

But, here's another way of doing it. Let's go everywhere all at once. Launch probes to every single galaxy that you could ever reach. Small technical details: we might need to go around the milky way, so some of the galaxies, you have to go in two stages, but that's a minor thing.

This is our exponentially increasing universe.

And this is how far we can reach in co-moving coordinates at 50 percent of the speed of light, 80 percent, 99 percent and 100, well at light speed itself. If we allow ourselves to re-accelerate, then we can go further. So if we stop and then re-accelerate, we can go further. And this is the number of galaxies that we can reach from a low, if we go at half light speed of just a mere hundred and sixty million galaxies to a high of four billion.

These are putting all the numbers together. The numbers of probes, there's a certain redundancy. There's a bit I've cut out here, which is what do you do with space dust? You need a certain redundancy. There's a redundancy of 40 for the high speed. We just need a redundancy of 2 for the low speeds.

Total energy requirements. We have our sun. We've Dyson'd it. How long does it take to get this amount? Well, for the worst possible scenario, the fusion launch, we need six hours of the sun's energy. To power the launch to every single galaxy we could ever reach at these speeds. And if we go for the 500 ton replicator, it goes up to 11 thousand years. Again, we can use more stars if we want to. We can pause along the way. Do it the old fashioned way.

But again, on the cosmic scale, these numbers approximate to zero. So what is going to happen if we consider that we can expand so easily, hence aliens can expand so easily. By the way, the initial impetus of this analysis was to make the Fermi Paradox much, much worse. Because we don't need to just worry about why don't we see aliens in our galaxy, but why have they not come here yet? And why have the aliens of nearby galaxies not reached us yet? Especially because the earth is a late-coming planet amongst the earth-like planets. Anyway that's another thing.

If life is rare, you'll get this: mainly isolated bubbles. If that's the case, then this on the left is where we get to the really far future.

And on there right, there's the scenario where there's lots of lifes and then we'll encounter them as we expand, and we'll have to negotiate around the frontier. The problem there is external coordination. What do you do when civilizations meet each other? There's attack versus defense, what is the balance when you're in the cosmos? Can you have better technology at this stage? Or is pretty much, are you, at least for warfare, are you pretty much maxed out?

What about scorched earth? If you say, well if you attack us, we'll destroy our resources, it'll just cost you. You'll gain nothing. There are negotiations, extortions, threats, surrender, there's an interesting dynamic there. It's very hard to predict but these are the considerations that we'll have if there's lots of civilizations that'll encounter each other.

In the empty scenario, it's a question of internal coordination. How do we stop the different branches of humanity or descendants of humanity from breaking apart? This is the amount of mass that we could eventually reach. This is that mass taken as a mass energy. And this is the number of erasures that we could theoretically do with that amount of energy.

Now, remember what I said about the very far future? Computing is more efficient when it's cold. If we wait a long, long time. And by a long long time, I mean trillions of years. Don't be so ridiculous, this is, you won't even notice trillions of years that go by. But if we wait a really long time, the only heat that we will get will be from the event horizon of the cosmic expansion of the universe itself. And then the temperature will be extraordinarily low. And then if we take all this energy, and run it, this is the amount of erasures that we can do. The Landauer Limit, or the Bekenstein Bound. One or the other of that.

Erasures are important 'cause that's the thing that gains entropy. You can't avoid getting entropy when you erase stuff on a computer. So if you have reversible computation, maybe quantum reversible computations, how often do you need to erase? And then that'll give you, by the way, ten to the one hundred and twenty-two is a very large number indeed. I think that's one of the biggest understatements I've every said.

But this gives you the idea of the amount of computation that you could theoretically run. I say theoretically because we have a problem getting this mass energy. This mass into energy, and then waiting that long. The problem is that proton decay is just too damn fast.

Proton decay is on the scale of ten to the thirty four years or so. And it seems that the protons will decay before the black holes start evaporating. And that is very annoying 'cause otherwise we could just chuck out all the stuff into the black holes, wait, and harvest the energy as they start evaporating. But this is theoretical maximum.

The point of this was, nobody had done these computations before me and Anders Sandberg and Bradbury, while he was still alive, did some of them, and we sort of put it together, so you can really put together interesting stuff that is there. And the second thing is, this is really cool and shows the potential massive amounts of value that could exist in the universe.

And if we can ensure that that's flourishing value, that would be really nice. Thank you.

Q&A

Question: What about aliens? I think you touched on this a bit with the external coordination. So, I'm gonna spin that as like, which of the external coordination problems would you expect to be most challenging?

Stuart: Whether there is a principled way of resolving disagreements or not. So, whether you have to have wars or not basically. But we can talk about that more later. Because that gets esoteric.

Question: Someone wanted to hear your explanation of the difference between using a rocket and a gun for deceleration.

Stuart: When you use a rocket, you have to accelerate or decelerate your fuel that you will then use to continue to decelerate. If you use a gun… if you're accelerating the bullet by shooting the gun, that's all in one impulse, and you don't get the fuel for the bullet that is accelerated to the same extent.

You can also, this is probably more realistic, eject your bullet, have it deploy a solar sail and then just paint it with a laser from your gun, and that will avoid the rocket equation as well.

It's basically, you're just decelerating the payload. If the payload carries the fuel, you have the rocket equation. If the payload does not carry the fuel, and the fuel is stored in the gun or the rest of the thing itself, you can avoid the rocket equation. It's Eric Drexler who came up with that and it took me a long time to convince myself about it, but the equations work out.

Question: Where do you place the greatest probability on a great filter?

Stuart: I put the great filter early, at the very beginning of life, pre-life or at the very beginning of life. If I had to bet, I would bet on, either pre-life or mitochondria kind of thing, or maybe oxygen. But, we can talk about that more in office hours, because there's a lot of interesting other points there.

Question: You said that science fiction often talks about the British colonization model, but you think we might be able to go to every galaxy simultaneously. Is there an assumption then, that you would have enough energy early on in the process to do this simultaneously?

Stuart: Yes. For the small probes, you need six hours of the sun's energy to launch all the probes to all the galaxies. So yes, we have some slack.

I'll also just mention that we have alternate designs. Maybe we just want to hit a single super cluster, and if we have heavier probes, we can just hit super clusters and radiate out from there. There's other designs that we could do. But we can just blast every galaxy with that kind of energy.