Recently, I’ve been mostly-loving the podcast Philosophize This!, and I just listened to Episode #95, Are you living in a simulation? It discusses Nick Bostrom’s paper arguing that the likely answer is “yes”.
I never worried too much about these arguments, on the principle that the answer doesn’t affect what I care about in life, but listening to it, I realized that the basis for this theory is based on old ways of thinking, and the likelihood should go the other way.
As a person who creates simulated realities all the time, we have a secret: most simulations borrow a huge amount from reality (or at least, from their parent reality).
Most of what makes a world– simulated or otherwise– is its data. This is one of the great insights of the machine learning revolution. And you would have a tough time creating that data from scratch even if you wanted to.
If you tell the computer of a holosuite from Star Trek to create a person, it wouldn’t ask you, “Would you like me to simulate the evolutionary process of personhood from first principles?” It will just assume that everything about your simulated person is the same as the way real people work. Cut the person in the simulation, and they’ll bleed like a real person. All the parameters behind their bleeding– the color of the blood, the rate of bleeding, etc.– are aspects of reality.
Plus, simulations have boundaries. For example, if you were interested in creating a simulation of Earth at this point in time, it would be sensible to grab our knowledge of the whole rest of the universe and just plug it in. There’s no reason to create a different night sky.
These boundaries can occur at any level: you can model a planet in the context of the universe, a person in the context of the world, an idea in the context of the brain. In fact, brains do this all the time.
Whereas the original thought experiment had only one data point (our apparent existence prior to world-simulating capabilities), we actually know quite a bit about the question of how much data filters from reality through the simulation boundary of our mind. There are informative arguments from Kantian pure theory and from that logic where Gödel meets information theory. None of it conclusively resolves the mind-body distinction, but neither did the original simulation data point.
But since I started this with an insight from computer science, I’ll complete the loop there too. What do we know about the share of data-vs.-simulation necessary to create a mind? While there’s a lot more we need to learn about the building-blocks of intelligence, the machine learning revolution has taught us that intelligence is hugely built upon data, not on modelling.
So, let’s return to the fundamental thought experiment: the number of simulated realities should exceed the number of real realities. But if you take a random piece of data within any of those simulations, the number of real data points is likely to far exceed the number of simulated data points.
I am not saying that there are not simulated aspects of the world we experience. But I would argue that there is no essential difference between the boogey man of a “simulated reality” and aspects of the universe that we already accept.
Are computers involved in creating our world? Yes, you’re reading a post on one of them now.
Are aspects of you or the world a “copy” from somewhere else? Sure, much of your experience of the world, as simulated by your brain, is just re-applying ready-made templates to the raw facts of the universe.
Is the core “you”, your subjective self, a simulated quantity? Either subjectivity cannot be created on a computer, then no, or it’s an emergent property, and then already modeled on brain hardware.
I imagine that these last points have probably been made plenty of times before by people who are missing the point of the simulated universe problem. And maybe I am too, but in light of the role of real data in simulations, the original question definitely was.