Ancestor Simulation Script (Patreon)

As our video games become more and more life-like, it’s becoming clear that at some point, perhaps soon, our simulations will be indistinguishable from reality. If that’s true, how do we know it didn’t already happen? Could we be in a simulation now?

The idea that we’re just part of a simulation sounds outlandish and, frankly like science-fantasy. However some very serious scientists and philosophers and tech billionaires think it possible, and perhaps even likely. Before I get into details, I want to share with you a conversation I recently had about this idea with my colleague at the Hayden Planetarium in New York City – in fact, with its director, Neil deGrasse Tyson, along with comedian Eugene Murman as part of Neil’s StarTalk Radio show.

(To see the conversation:: https://www.youtube.com/watch?v=6Ng1wwW6AeE )

The conversation goes in lots of fun directions. You can check out more on StarTalk Radio – link in the description. Now, Neil uses an entertaining example of us being a Sims game for hyper-advanced alien basement-dwellers. That’s a hilarious image, but he’s illustrating a very serious point: that under certain assumptions, virtual minds should vastly outnumber real minds in our universe. If so, shouldn’t we be virtual minds? Let’s dive deeper into this rabbit hole and decide if this actually makes sense.

But before we do so, we should be clear about the type of simulation we’re talking about here. Let’s avoid the idea that the entire universe is simulated right down to every atom, electron, or vibrating quantum field. That’s a much deeper rabbit hole - one we’ll need the holographic principle to explore. It’s a whole big thing. We’ll get back to it.

Instead, today I want to talk about the idea that it’s our experience of the universe that is simulated. That we are simulated minds in a virtual universe that has just enough detail to convince us of its reality. I’m talking about so-called ancestor simulations, an idea proposed by Oxford University’s Nick Bostrom. It goes like this:

Bostrom argues that in the future it will be possible to simulate the action of all of the neurons of a human brain, and to simulate the sensory input to that brain with enough fidelity to convince the simulation that it’s a real person. That's not much of a stretch. There's a good chance we'll be able to simulate the action of all neurons in one human brain within a couple of generations. However Bostrom crunches the numbers to show that a super-advanced civilization could do this on such a scale that these sorts of virtual minds vastly outnumber real minds. Why would they bother? For science! He proposes that an advanced civilization may want to run simulations of its own history to study the behavior of the types of minds that lived that history. He calls these “ancestor simulations”.

Let’s look at the numbers. The human brain has 100 billion neurons and well over 100 trillion synapses. It’s been estimated that the entire operation of a single brain could be simulated with somewhere between 100 trillion to 100 quadrillion binary operations for every second of time that the brain experiences. Bostrom argues that it doesn’t take anywhere near that much computing power to then simulate an external environment with the fidelity needed to fool the virtual brain that its environment is real. That’s debatable because the environment needs to be perfectly consistent with respect to all measurements made by all brains in that environment. But whatever, let’s go with Bostrom’s assumption.

A full human ancestor simulation would simulate all humans that ever lived. Going back 50,000 years it’s estimated that around 100 billion people have lived and died. An average 30-year livespan gives each of them a billion seconds, and each of those seconds requires 10^14-10^17 operations. Multiply those numbers together and you get 10^34-10^37 binary operations to simulate all of human history. Bostrom says 10^33-36, but potato potato. You can mess with any of the numbers and still remain within those few factors of 10. 

So how long would that take to compute for a super-advanced civilization? Bostrom uses Robert Bradbury’s estimate that a computer the size of a large planet – a so-called Jupiter brain – would be capable of performing 10^42 operations per second. In other words, it would be capable of simulating the entire mental lives of all humans in history a million times over - every single second.

Just one such computer would generate an insanely large number of life-long mental experiences that are indistinguishable from the type of mental experience you and I are experiencing right now.  That’s true even if you scale back - say to a computer the size of the moon, or if you assume several more orders of magnitude in the computing power needed to run the simulation. Bostrom claims the following, which he calls his simulation argument: If ancestor simulations are something that even some civilizations end up creating – so if they advance far enough AND decide it’s a good idea - then most of the self-aware minds that ever come into existence will be simulated ones. Therefore we are in an ancestor simulation.

This sort of existential angst about disembodied brains being more common that real ones didn’t start with Bostrom. The thought experiment is similar to that of the Boltzmann brain. The idea is that in an infinite multiverse, in should be vastly more common for particles to randomly assemble into a brain that is having exactly your current experience of the world than for particles to randomly produce Big Bangs. We talk about it in our last episode.

Both ancestor simulations and Boltzmann brains require us to invoke something like the Copernican principle. It tells us that we aren’t in a special place in the universe – we’re on a typical planet around a typical star in a typical galaxy. With one exception – our place in the universe must have been able to produce and sustain us – so we’re somewhere habitable. That last addendum is an application of the Anthropic Principle – we must observe a universe, or a part thereof – that can have observers.

Copernican reasoning, with a dash of the Anthropic Principle, tells us that we should be the most typical – the most common type of observer that could possibly be having our current experience. So if the virtual minds of an ancestor simulation are vastly more common than the minds of the original, living creatures, AND if the simulated experience is completely consistent with OUR experience, then it’s more likely we are those more typical observers. I should note Bostrom is on record as placing the odds at less that 50% that we’re a simulation. Why? Because he thinks it just as likely that either all civilizations die out before being able to create vast-scale ancestor simulations, or essentially no super-advanced civilizations choose to make them.

The ancestor simulation idea suffers from some of the same issues as the Boltzmann brain idea. We already talked about Sean Carroll’s argument against concluding that we are Boltzmann brains: as soon as we do so, we must also conclude that we probably don't have the capacity to have carried out that line of reasoning in the first place. A Boltzmann brain is as delusional about the consistency of its mental faculties as it is about its existence before that instant. Similarly, in the case of ancestor simulations, upon deciding we are simulated we acknowledge that there’s no experiment that we can do to prove that we are NOT. The hypothesis is unfalsifiable. Bostrom himself points out that upon being found out by one of its resident minds, the simulation can be instantly edited or rewound. In fact this edibility is a necessity. These simulations can only cover a tiny fraction of the universe, so they are prone to inconsistencies. It’s far more computationally economical to edit out the discovery of these inconsistencies than it is to simulate to much of the environment that inconsistencies don’t happen.

Another big issue is just the potential for spectacular over-reach in using these Copernican and Anthropic arguments. Bostrom even weighs in on this with regards Boltzmann Brains. He posits a type of Bayesian reasoning – so, assessing the probability of a hypothesis being true by taking into account prior probabilities. Imagine a philosopher telling a cosmologist that we must surely live in a universe capable of producing the most brains, because that universe would give the maximum probability of our own existence – the one thing we know for sure is true. Therefore scientists should come up with cosmologies that generate the most minds. The universe that has the most minds must be the one that we’re in. Bostrom labels this the presumptuous philosopher problem, and warns against such reasoning. I wonder if he needs to reject his own ancestor simulation idea on the same grounds.

Highly specific scenarios like ancestor simulations or Boltzmann Brains generate impossibly large numbers of minds that are identical to our own. But where do you stop? With a little imagination, yet more mind-factories can be conceived, and simulated us’s multiply exponentially. Which type of fake you are you? Just for now, I strongly recommend that we proceed as though we are real live observers, part of the original space time.