The British philosopher Nick Bostrom mentioned that he believes that there’s a roughly 1/3 chance of each of the following mutually exclusive, collectively exhaustive hypotheses to be true: (1) all civilizations perish before reaching a point of being able to construct hyper-real simulations, (2) all simulation-capable civilizations establish strongly enforced bans against hyper-real simulations, or (3) we are overwhelming likely to be in a hyper-real simulation (https://www.simulation-argument.com/). The argument for (3) is that any civilization capable of creating hyper-real simulations would likely create many of them, so the vast majority of conscious being would reside in a simulation rather than “base reality.”
Given my techno-optimism and libertarianism inclinations, I would be more likely to rate the probabilities as 20% perish, 5% ban, and 75% simulation. Yes, I believe there’s a greater than 50% chance that I am living in a hyper-real simulation.
This leads to a second question that I ponder frequently about when I’m not doing real work: Am I alone or the main character? Assuming that (1) the lifetime of the “base reality” universe is finite, (2) the size of the habitable “base reality” universe is finite, and (3) that computation cannot occur with a time resolution of less than a Planck unit, then the total number of simulated consciousnesses (which far exceed the number of base-level real consciousnesses) that ever exist will be finite. What does the population size distribution of simulated consciousnesses look like? Are most of the simulated consciousnesses living in full simulated universes with billions or trillions of simulated consciousnesses, or are most of them living in narrow simulations populated by just themselves?
It is impossible to fully know the distribution of simulation consciousness populations, but if modern science is any guide, there will be a power law distribution. For example, when scientist perform computationally expensive quantum mechanical simulations or molecular dynamics simulations, the vast majority of the simulations will be small simulations of one or a few molecules for a few nanoseconds. However, there will also be some mega-simulations by well-funded groups that include many molecules. Put another way in lay terms, there are many more college students that can run a small simulation (read: video game) with a limited number of simulated consciousnesses (NPCs) on their laptop, but the the small number of mega-simulations run by billion-dollar companies have many more simulated consciousnesses in each.
The likelihood of being in a small vs. big simulation depends strongly on the power in the power law: In other words, with every doubling with simulation size, do the number of simulations reduce by 50%, >50%, or <50%? Thinking in log2’s, the simulated consciousness population of the simulation that I’m likely to be living in could, for example, range from 2^0 (solipsism) to 2^33 (8 billion humans) to 2^50 (1 quadrillion consciousness in a space opera universe in which we human find ourselves in a quiet corner). If the power law power is exactly 1, then there’s roughly a 2% chance that I’m alone, 2% chance that I and exactly 1 other person are actually conscious, …, and a 2% chance that I’m in the space opera involving a quadrillion consciousnesses.
To explain the small population simulations, video games have long used the concept of “real-time rendering” to conserve resources. Meaning, the world, the objects, and the NPCs do not exist until they come into the player character’s (PC’s) field of view. For example, if the point of a simulation is to test the effects of socioeconomic status on aging and lifespan, an N=1 simulation centered a single consciousness would be sufficient and most efficient. All other “people” could be simple scripts that are occasionally run. An N=2 simulation may be created to study romantic relationship dynamics, information transmission through generations, or strategic competition. All of these are likely (and if I had the capability to run hyper-real simulations, would be interested in running all of them).
Personally, I believe that I exist neither in a N=1 simulation nor in a N=2^33 or N=2^50 simulation. Rather, I think the most likely situation is that I live in a N ≈ 2^8 = 256 consciousness simulation. Assuming that it is hard to fool a simulated consciousness with a low-level script, we would expect that most of the “people” that a simulated consciousness interacts with at any depth to be also simulated consciousnesses, while shallow acquaintances could potentially be simpler scripts (e.g. you probably wouldn’t notice if your local Starbucks barista is replaced by a human-looking robot). Since antiquity, Dunbar’s number = 150 is roughly the maximum number of people that any (human) individual routinely interacts with, so leaving a remaining buffer/swap of 106 would take us to 256. As some fraction of this 150/256 die, their “consciousness” slot may be taken up by a new birth — and so this hypothesis of “parsimonious simulations” is somehow innately consistent with the concept of reincarnation.
Dave Zhang's Startup Blog 