Latest / 80,000 Hours Podcast / We can guess what intergalactic war would look like. And strangely, it matters.
Transcript
- Rob Wiblin: Let’s talk intergalactic war in space, billions of years from now. What would something like that actually look like, according to our best understanding of physics? I recently came across a really interesting analysis of exactly this question from the AI researcher Beren Millidge. And I want to walk you through it because it’s incredibly fun, remarkable how much we can likely predict this far in advance, and — as I’ll explain — it might even be decision-relevant to humanity in the relatively near future. If you spend any time thinking about the far future — and by “far future” I mean billions of years from now — one question that really matters is whether our universe mostly ends up peaceful or mostly ends up violent. Does it settle into a patchwork of civilisations, each quietly doing their own thing? Or does it end up looking more like a science fiction movie, with enormous space empires locked in some sort of endless conflict? The answer to this probably turns on a surprisingly answerable question. In an all-out war between two galaxy-spanning civilisations, does the defender or the attacker have the advantage? If we want a peaceful future, we’re going to want to hope that it’s the defender, because in that case violence won’t pay and so it probably won’t exist. Or at least not much of it anyway. Here’s the setup: imagine two civilisations, each occupying their entire home galaxy and harnessing the energy of every star inside it. Both have maxed out their technology tree. There’s nothing of meaningful use left for them to invent. These are what physicists call Kardashev III or K3 civilisations. To give the attacker their best shot, and to keep their commute manageable, we’re going to imagine that these two galaxies are positioned right next to one another. And we’re also going to assume no wormholes, no faster than light travel, no time machines — just known physics and engineering pushed to their absolute limits. What might a war between these two vast galactic powers actually look like? I’ll show you now. There are really only three viable classes of weapon for civilisations at this scale, each with very distinct strengths and weaknesses. First, beam weapons — specifically, incredibly powerful lasers. It sounds crazy, but it’s consistent with the laws of physics and material science to be able to target a laser at a planet-sized object in a neighbouring galaxy. It simply seems possible to build mirrors large and precise enough, and there’s almost no material in the intergalactic medium to scatter the laser or cause it to lose precision or strength. The best case for an attacker is if a K3 civilisation can successfully coordinate huge numbers of Dyson spheres across the entire galaxy to fire synchronised lasers at a single point in another galaxy. We can’t be sure whether a K3 civilisation can actually pull this off, but we’re not aware of any engineering constraint that clearly prevents it. If they can make it work, they would be able to deliver 10^35 [watts] of energy at any given planet-sized target in a neighbouring galaxy. The gravitational-binding energy of Earth — the energy needed to completely blow the planet apart — is about 10^32 joules, just 1,000th as much. So a laser like this would destroy Earth in a few microseconds. And then of course, the laser could be targeted somewhere else. The key benefit of such lasers, for the attacker, is that they come with no warning. The photons travel at the speed of light, the fastest speed possible. The first photon that tells you’re being shot at is the laser shot. So that sounds awfully offence-dominant, but this approach has a very important drawback: lasers can’t change trajectory once launched. If you’re firing from Andromeda at the Milky Way, the round trip of information is about 5 million years. You see where something was 2.5 million years ago, and then your laser takes another 2.5 million years to get there. A lot can change in 5 million years. Star positions are sufficiently predictable over that kind of timeframe to hit them. But anything that’s deliberately moving, even a little — habitats, fleets, computational infrastructure set on eccentric orbits — could be absolutely anywhere within an enormous volume of space by the time your beam arrived to try to hit them. That makes the intergalactic laser devastating against fixed targets: Dyson spheres locked around immobile stars, infrastructure built around the supermassive black hole at the galactic centre, planets on completely predictable orbits (like Earth). But against a civilisation that has deliberately chosen to spread out and move about entirely at random, you’re basically going to end up trying to hit a needle in a haystack the size of an entire galaxy. You can try to make your laser broader to hit more things, but its power just quickly becomes too low to do serious damage. The underlying issue is that space is just unfathomably empty. And that emptiness is a huge asset to any defender. Which brings us to our second weapon: usually referred to in the tiny literature that exists on all of this as relativistic kill vehicles, or RKVs. These are basically physical objects accelerated to some large fraction of the speed of light and hurled at the enemy galaxy: bullets, more or less. The kinetic energy at these speeds is so enormous that they don’t need to carry explosives, they just hit things and trivially obliterate them. Now, RKVs have one massive advantage over beams: because they’re objects physically closing the distance, they can carry onboard sensors and steering equipment to update their targeting as they approach the thing that you want to destroy. They’re not shooting blind from effectively 2.5 million years ago. So in the final tens of thousands of years of flight, they can survey the defender’s galaxy, pick a target, and adjust course. And over such long distances, even really modest onboard thrusters would allow you to home in on some individual habitat and keep steering into it right up to the moment of impact. The downside? Well, naturally they travel slower than light, and that means the defender gets to respond. If you’re shooting from 2.5 million light years away, an RKV launched at 90% the speed of light would give a defender up to 250,000 years of warning, at least if they spotted the missile when it was launched. Even if it was only spotted when already inside the galaxy, that’s still decades or centuries of reaction time. And this reaction time is a big benefit to the defender — because they don’t actually need to entirely destroy an incoming RKV, which would be a little challenging. It’s enough to simply nudge it slightly off course. Space, again, is almost entirely empty. So if you push it a little bit sideways, enough to offset its own steering, it’s going to sail past everything off into the empty void forever. Defensive laser installations — the same kind that any responsible property owner would build on their Dyson spheres anyway — can melt the surface of incoming RKVs, blind their sensors, and apply a small push to shove them off target. A push that adds up because they’re travelling so fast and for so long. Because the defender is sending pure energy a short distance, rather than mass an enormous distance, the energy required to deflect an RKV is just vastly less than the energy the attacker had to spend launching it and accelerating it. That asymmetry — cheap to deflect, expensive to launch — is the heart of the defensive advantage against these intergalactic bullets. The third type of weapon is the classic science fiction invasion fleet: ships that cross the void and then slow down, establish a beachhead in the opposing galaxy, and try to go on to conquer territory. This is the least promising approach, at least in the early stages of a war. Invasion ships just have a fundamental problem that beam weapons and RKVs do not: they’ve got to decelerate at the destination, and decelerating is the opposite of stealthy. You’re going to end up ejecting an exhaust plume, which is incredibly visible to the people you’re trying to surprise. You are essentially releasing a giant beacon, screaming, “I am here, come kill me!” for literally thousands of years as you slow down. The deceleration issue also makes it energetically expensive to send a large functional payload, because you have to accelerate to nearly the speed of light, carrying with you not only your invasion fleet, but also all of the equipment you’re using to speed up, and all of the equipment you need to slow down at the other end — which is basically the same amount again. On top of that, invasion ships — unlike RKVs, which are basically just bullets — they’re going to be expensive, kind of internally complex, and therefore pretty fragile objects. So even a modest laser shot could probably break them apart. And even if you get past all of those barriers, you arrive fighting an entrenched civilisation with all of its industrial capacity right there — while your supply lines stretch back millions of light years, leaving you just completely isolated. You would effectively be expending a lot of resources to send over something that can be blown up by the defender for pennies on the dollar. So invasion fleets could, at best, become relevant late in a war, once a defender is already very weakened by the first two types of weapons we’ve talked about. So what should a sensible, paranoid defending civilisation do in response to all of this? First and foremost, don’t be a sitting target. Do not cluster your people or infrastructure around fixed, predictable points like stars and planets. Instead, you’ve got to starlift. Pull the hydrogen out of your stars and distribute it to billions of mobile habitats scattered across the galaxy on randomised, shifting orbits. Each one of these could easily carry enough fuel to support a fusion reactor that continues running for billions and billions of years. Apparently, this is somewhat less energy efficient than a nice orderly set of Dyson spheres, but it would make your civilisation virtually impossible to wipe out through distant bombardment. Attacking a paranoid, diffuse K3 civilisation set up in this sort of way would be, as Millidge puts it in his essay, “like trying to punch holes in a fog.” Second, build massive redundant sensor networks: giant interferometer spanning light years, deep space sentinel probes extending far out into the galactic halo and beyond. Anything unusual — an object too big, moving too fast, or with the wrong spectrum for natural interstellar debris — that’s got to get flagged and tracked and probably destroyed. It sounds expensive, but in the scheme of the resources that we’re playing with here, as a K3 civilisation, it’s all very affordable, I assure you. A paranoid civilisation would also probably want to seed diffuse-dust minefields along likely attack vectors from nearby galaxies. This would slow down incoming RKVs and light them up for defensive systems to easily see. Third, maintain a powerful invisible second-strike capability. Even if you’re hit without warning, you should be able to launch a devastating counter-bombardment. And of course this doesn’t help you survive the first strike, but it raises the cost to the attacker, making war even less attractive to them than it otherwise would be. And fourth, maintain the ability to detect and neutralise any enemy self-replicating probes. You know, von Neumann machines that manage to slip through and start building up inside your galaxy. If they could get away with this, this would kind of be the attacker’s ideal and their long game. You want to get replicators established internally where they can operate and attack with much shorter time lags. Stamping these out, as we’ve talked about, is really doable with your galaxy-wide monitoring system. But the challenge is that you do need to have a bunch of redundancy, so this system can work almost perfectly — even if you’re facing a huge influx of lasers and RKVs targeting your defensive infrastructure. As best as we can tell, the fundamental dynamics here strongly favour the defender, so long as they prepare ahead of time. The attacker has to cross millions of light years, expending enormous energy to accelerate mass or coordinate their beams. They’re operating on information that’s millions of years old. While the defender, by contrast, is right there: short communication distances, real-time sensor data, all their industrial and laser capacity immediately available to respond in some intelligent way. And the asymmetry gets more extreme with greater distance. We set things up by imagining a best-case scenario for an attacker: two galaxies in the same local group, you know, only 2.5 million light years apart. Scale that up to a warfare between galactic clusters separated by tens or hundreds of millions of light years, and offence essentially becomes an exercise in futility. So even in the best case, the attacker is extremely unlikely to completely destroy a prepared K3 defender. They could seriously damage a naive or early one — you know, a civilisation that’s clustered all of its infrastructure around fixed predictable points. But a paranoid distributed mobile civilisation, they can survive even a surprise attack with relatively manageable losses, and then settle in for a war of attrition that lasts millions of years, in which the attacker really can’t win at all and isn’t gaining anything much from. This makes large-scale intergalactic warfare probably irrational, in a cold economic sense. The attacker would almost certainly expend far more resources prosecuting the war than they could ever hope to gain through conquest. The only exception here might be an ideologically motivated attacker that values the destruction of the other civilisation for its own sake. And even they kind of face the problem that their campaign would be super visible across the observable universe, potentially inviting intervention from other third parties. And if they have some concrete goal, like stopping the other civilisation from doing something that they oppose, that they dislike, they might well be able to get the same result much more cheaply through some form of negotiation or trade. Now you might reasonably ask: why should anyone care about hypothetical wars between galaxy-spanning supercivilisations? I think there are a few genuinely important takeaways here. First, if we succeed at building aligned AI, and if we colonise the Milky Way galaxy — or eventually spread beyond it — the defence-dominance of space warfare suggests that we probably can’t be easily wiped out by hostile aliens or AIs from elsewhere in the universe. Once a mature civilisation is established in a galaxy, it’s extremely hard to dislodge. And so the universe likely ends up as a stable patchwork, each galaxy belonging to whoever colonised it first, with very little incentive for conquest. Even a pure paperclip maximiser, for instance, would get fewer total paperclips by invading neighbouring galaxies than by just quietly converting its own galaxy into paperclips, defending it and sitting tight. That’s a much more peaceful long-term picture than the ‘Dark Forest‘ scenario so popular in science fiction, where every civilisation is terrified of being destroyed by every other one. The second takeaway is less comforting though. It cuts the other way for us humans right now. Everything I’ve just described about defence-dominance depends on being a technologically mature, distributed, paranoid K3 civilisation. We are emphatically not that. We’re a single-planet species with all of our eggs quite literally in one basket, on a body whose orbit is predictable billions of years in advance. Any Kardashev III civilisation would have telescopes able to map the Earth’s surface in reasonable resolution, and they could probably vaporise everything on Earth with a well-aimed laser beam. The fact that this hasn’t happened is actually good evidence that there isn’t a K3 civilisation anywhere near our galaxy, at least not a genocidal one that wants to eliminate us as competitors. Right now, and for as long as we remain clustered around a handful of predictable objects — Earth, Mars, a Dyson sphere around our sun — we are at a maximally vulnerable point. Fortunately, the chances that another civilisation reaches the technological maturity to spot us and decides to wipe out Earth in the next 1,000 years: that’s really pretty low, seeing as they haven’t managed to do so in the previous 1 billion. And we could almost certainly have distributed backups of our civilisation set in random orbits around the solar system within 1,000 years. But it’s kind of crazy to think that we could feel pressured to do something like that so soon. The third takeaway is that probably not many of the universe’s resources end up wasted, expended, or more in the long run, because mature civilisations are so hard to dislodge. The universe’s long-term allocation of resources is likely determined almost entirely by what happens in the relatively brief settlement phase: a form of cosmic land grab that plays out over a relatively short few million years. Whoever grabs a galaxy first likely keeps it, and that means there’s greater potential for what complex life originating from Earth does in the next few centuries to be sticky, permanently. But is it good for humans and our descendants to take over this galaxy, as opposed to some other alien species later on? Well, that depends on whether we’re better than average — that is, better than the random background replacement civilisation that might come along later — which itself is downstream of our intelligence, our wisdom, our moral values, our ability to coordinate and to avoid harmful internal conflicts, our willingness to do the right thing (whatever that turns out to be), our capacity to avoid slipping into tyranny by random individuals or small groups with strange and cruel values, and our ability to make prudent decisions together. No pressure.