Anthropic Principles and Existential Risks

• authors Katja Grace

Existential risks are more likely than you think.

Imagine that God is going to toss a coin. If it comes up heads, he will create 10 people in isolated numbered rooms. If it’s tails, he will create 1,000 people, similarly housed.

You learn all this from a poster on the wall of the room you have suddenly found yourself in. Though interested in the deeper meaning of your strange origin, you find yourself most curious about how God’s coin came up. Are you one of only ten people or one of a thousand?

At first you think there is a 50 percent chance of either. The coin was fair, and everyone who was created was bound to find themselves in a room like this either way. You have no new evidence.

But then you think about it a different way. All of the people who could be created, under either heads or tails, had the same chance of existing — 50 percent. Intuitively, it seems then that you have the same chance of being any of them. And virtually all of them only exist if tails came up, so your existence is strong evidence that tails did in fact come up. Further to this, you reflect that if God repeats this experiment, most people ever born will be born on those occasions when tails comes up.

Just as you are rejoicing at the realization that you are very likely to have 999 brothers and sisters, you suddenly find the first argument persuasive again. And then the second and the first. You may be relieved to read at the bottom of your poster that the answer to this question is a matter of debate among those who have spent much longer thinking about such things.

The thought experiment you just found yourself in is God’s Coin Toss. It was invented by the British philosopher Nick Bostrom. The two answers correspond to two popular kinds of anthropic principles.

A diverting puzzle, you may say, but who cares? God is apparently not a gambling man, so this situation will never arise. And even if you do find yourself in such a situation, the world is unlikely to end if you guess wrong. But there you could be mistaken. To see the deadly fix your anthropic reasoning may illuminate, you need to know that you are in the Great Filter, and also what that is.

If you look very closely at the sky, you will see a fantastically huge number of stars, about 10^22. If you look among these stars for signs of vast alien colonization explosions sweeping out from them, you will see none. This means the chance of a given star giving rise to a civilization which visibly colonizes beyond its home star is less than miniscule.

Now picture the path of necessary steps between a lifeless star and a civilization with the power to colonize other stars. First, the lifeless star needs to have planets. At least one planet needs to give rise to life. That life needs to develop high intelligence. We just worked out that there is a tiny chance of a given star reaching the end of this path. This means at least one of these steps is very hard, most likely more of them. This set of steps is called the Great Filter, because stars get filtered out as they progress along the path.

The big question about the Great Filter is whether most of the hard bits are between steps that humanity has already passed, or whether a lot of difficulty remains in our future. If we have passed most of the hard parts, our star is incredibly unusual. We are alone amongst a vast number of primitive stars. This is actually the optimistic case, because if we have passed most of the filter already, we are quite likely to make it to the end of the path from here. We can expect to colonize space without much difficulty.

On the other hand, if much of the Great Filter’s strength is in steps we have yet to pass, this star’s cargo is ordinary. Aliens at our level are two-a-penny, but most of them don’t last. They get filtered later. Since in that case we have no known advantage over the masses, humanity is also very unlikely to make it through the filter.

It’s very important to make it through. The worst ways to get filtered in future steps include many human extinction risks such as disasters from nanotechnology or biotechnology. Some extinction risks could not be filters because they would destroy such a huge chunk of the universe at once that we can tell they did not obstruct life in the past. For instance, vacuum-state decay would destroy everything including the prevailing physical constants, so it is not one of the filters we are looking for.

Other future filters we may face would not immediately destroy us, but would still prevent the vast future of happy lives possible if humanity can expand beyond its home star. For instance spacefaring technology may be much harder than we imagine. Or ecological or social collapse may send technological civilization into a dark age from which it turns out to be virtually impossible to reascend.

Aliens may be colonizing the galaxy imperceptibly. This can be thought of as a large filter at the last step, a step of being visible after expanding into space. If this is the only large filter in our future, we may yet be safe, or perhaps someone else is enjoying the universe’s lush resources. This unlikely scenario is the best available with most of the filter in our future. Those are much more worrying than filters in our past.

We want to know whether most of the filter is in our past or our future. Preferably, we want to know that it’s in our past. The anthropic principles I mentioned earlier can help answer this question.

Picture an array of possible universes we might be in, some with most of their filter strength before us, some with most of it after us and some with a more even mixture. Some of these worlds are more likely than others, from what we know about things like evolution and space travel. Some of these worlds have more creatures at our stage along the path to space colonization. Those where less of the filter’s strength is in steps we have passed will generally have more creatures at our level, because there will be more solar systems containing such creatures.

You are actually in something very much like God’s Coin Toss again. This time, there are many possible universes you might be in instead of just two possible worlds. The populations of the universes span a range of many orders of magnitude. The main difference that might matter is that these possible universes contain many other creatures at different stages of development whom you know you are not. For instance pterodactyls, neanderthals or space worms. Do the anthropic principles tell us anything interesting here? Since we don’t know which, if any, of these principles is correct, lets try both of them.

The second principle is easiest to apply. It said in God’s Coin Toss that you are a hundred times more likely to be in the world with one hundred times more people. If you apply the same reasoning here, you find yourself much more likely to be in one of those universes where our stage of development is teeming with occupants. That is, a universe where less of the filter’s strength is in our past. This is a universe where we are more likely to be doomed.

The first principle is a little more complex to apply here. In God’s Coin Toss, we learned that since everyone sees what I see, I haven’t learned anything. In general, it says I’m more likely to be in worlds where a greater fraction of people see what I’m seeing. In God’s Coin Toss the fraction was one hundred percent in either world, so we didn’t update our probabilities. The situation is more complex in the Great Filter, since there are other creatures. What proportion of people who see what I see depends on who I count as everyone. If I count only creatures at our own stage of development —those with technology at about our current level— then the situation is identical to God’s Coin Toss. I learn nothing and find myself no closer to doom than I previously thought. However, if I use a wider reference class, counting more creatures in the observers, I will generally begin to receive bad news again. I will explain this step by step.

Suppose I include some creatures from past stages of the filter in the reference class, such as intelligent creatures with significantly less technology than us. Now in worlds that have larger filters between them and us, there are less of us relative to them. That means such worlds are less likely. Intuitively, in those worlds you would expect to be a common earlier creature, not a rare late creature. Since we know the total strength of the filter is pretty big, past filters being weaker means future filters are likely to be stronger. Just looking at past filters, this principle says we are more likely to be in trouble.

Now lets look at future filters. Suppose that I include some creatures from future stages in the reference class and none from the past, such as creatures similar to us but with nanotechnology. Now worlds with larger filters between our stage and the later stage or stages become more likely. This is because worlds with larger filters after us contain fewer creatures at those future stages relative to creatures at our current stage. Our experiences of being early are proportionally more common. Again, we find ourselves more likely to be in a world with bigger filter steps in the future, and so in more danger.

We have seen that if we extend our reference class in either direction (or both), danger looms. Naturally, if we extend it in both directions, we get a combination of these effects and danger still looms. At least it looms larger than you thought it did. Remember that the likelihood of existential risks at this point depends on the prior probabilities you held on being in the various universes, including over which future filters are more likely as well as the anthropic principle you use. You have merely shifted your expectations enormously in the direction of pessimism. If, for instance, you originally thought it virtually certain that we are alone in the universe, you should continue to be fairly optimistic about our chances.

The two principles I showed you make different predictions about details such as the extent of the probability shift and the effect of timing. However, since both principles agree that we have underestimated the risk of future filters, we need not resolve the disagreement over which principle is correct to update our expectations of danger. We need only be confident that something like one of these principles is correct. While we look for a more promising principle, perhaps we should direct some of our attention to the urgent task of investigating and averting extinction risks.